diff --git a/specialists/HANDOFF-2026-03-28.md b/specialists/HANDOFF-2026-03-28.md
new file mode 100644
index 0000000..34712a1
--- /dev/null
+++ b/specialists/HANDOFF-2026-03-28.md
@@ -0,0 +1,239 @@
+# Whetstone Specialist Fleet — Handoff 2026-03-28
+
+**Session date:** 2026-03-28
+**Status:** Phase 0 complete. Two specialists trained and validated. Clean stop.
+
+---
+
+## What This Directory Is
+
+A fleet of ~800KB transformer specialists that replace bounded-output LLM decisions
+in the Whetstone MCP pipeline. Each specialist handles one classification decision,
+runs in under a millisecond, and is trained using the WhetstoneAI Fabricate binary
+and harness unchanged.
+
+**Origin:** This grew directly from the WhetstoneAI Fabricate RSA experiments
+(`/home/bill/Documents/WhetstoneAI_Fabricate/`). The same ~213K-param architecture
+that reached 96.3% in-distribution accuracy on domain routing was applied to
+Whetstone's bounded taskitem decisions. No new training infrastructure was written.
+
+**Full context:** Read `README.md` and `ROADMAP.md` in this directory before
+starting work. They contain the architectural framing, the AST-first design
+correction, the 6-tier automatability hierarchy, and the complete next-step plan.
+
+---
+
+## What Was Built This Session
+
+### Two trained specialists
+
+**`verification_type`** — 5-way classifier
+- Input: sprint step description text (e.g. `"whetstone_validate_schema MCP tool"`)
+- Output: `unit / integration / schema / smoke / docs`
+- Training: 978 examples mined from 260 sprint plan files via heuristic labeling
+- Eval: 172 held-out examples
+- Result: **83.7%** accuracy, grokked at step 1000, stopped at step 3000
+- Checkpoint: `/mnt/storage/fabricate_runs/whetstone_verification_type/`
+- History: `specialists/runs/verification_type_history.json`
+
+**`confidence_tier`** — 3-way classifier
+- Input: structural encoding `"conflicts=N ambiguity=M deps=D prereqs=P queueready=yes/no"`
+- Output: `high / medium / low`
+- Training: 1020 synthetic examples generated from the formula in `TaskitemConfidenceAmbiguity.h`
+- Result: **100%** accuracy, grokked at step 1000, stopped at step 3000
+- Checkpoint: `/mnt/storage/fabricate_runs/whetstone_confidence_tier/`
+- History: `specialists/runs/confidence_tier_history.json`
+
+### Scripts written
+
+```
+specialists/scripts/gen_verification_type_data.py   mines sprint plans → TSV
+specialists/scripts/gen_confidence_tier_data.py     generates synthetic formula data → TSV
+specialists/scripts/train_verification_type.sh      wraps run_grokking_until.py
+specialists/scripts/train_confidence_tier.sh        wraps run_grokking_until.py
+specialists/scripts/infer.py                        single/batch/eval-tsv/REPL inference
+```
+
+### Inference example
+
+```bash
+# From whetstone_DSL root, using fabricate venv:
+cd /home/bill/Documents/WhetstoneAI_Fabricate && \
+.venv/bin/python3 tools/eval_multiclass.py \
+  --checkpoint /mnt/storage/fabricate_runs/whetstone_verification_type/checkpoint.bin \
+  --db         /mnt/storage/fabricate_runs/whetstone_verification_type/fabricate.db \
+  --dataset    /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/verification_type_eval.tsv \
+  --labels     "unit,integration,schema,smoke,docs"
+# → overall: 144/172 = 83.7%
+```
+
+---
+
+## Key Technical Facts
+
+**Training data format** (required by Fabricate binary):
+```
+<integer_label>\t<hops>\t<text>
+```
+Label is an INTEGER index (0-based) matching the `--labels` flag order.
+**String labels in column 1 will crash the binary.** The generators write integers.
+
+**`--labels` flag:** `"unit,integration,schema,smoke,docs"` means label 0=unit,
+1=integration, 2=schema, 3=smoke, 4=docs. Order must match the TSV integers.
+
+**Model architecture** (all specialists use this):
+`--hidden-dim 128 --layers 1 --heads 1 --seq-len 96 --batch-size 32`
+Result: ~213K parameters, ~800KB checkpoint.
+
+**Grokking is normal.** Training logs show "Fresh start" (no prior checkpoint in
+output dir), then accuracy jumps at step 1000. This is expected behavior. The
+harness stops automatically 3 blocks after grokking.
+
+**Label/checkpoint alignment bug** (from RSA experiments — do not repeat):
+If you resume a checkpoint with a changed `--labels` order, the token rows
+misalign. Fix with `tools/swap_label_rows.py` in the Fabricate repo.
+See `WhetstoneAI_Fabricate/HANDOFF-2026-03-28-RSA.md` for full history.
+
+---
+
+## Important Architectural Decisions Made This Session
+
+### 1. AST-first, not tool-first
+
+Whetstone is an AST-as-source-of-truth system. C++, Python, etc. are projections.
+Specialists operate at the AST level — one specialist per decision type covers all
+languages. Do NOT train per-language variants of decisions that live in the AST
+layer. The projection from AST to language is handled by deterministic grammar rules.
+
+### 2. The 6-tier automatability hierarchy
+
+The current `AutomatabilityAnnotation` in `editor/src/ast/Annotation.h` has 5 tiers.
+It is missing the `specialist` tier. The correct hierarchy is:
+
+```
+deterministic  →  formula, zero parameters, always same output
+template       →  deterministic structural composition
+specialist     →  bounded output vocabulary, fuzzy, ~800KB  ← THIS PROJECT
+slm            →  open vocabulary generation, ~1-7GB
+llm            →  full generative, ~14B+
+human          →  not yet formalizable
+```
+
+Boundary rule: if the output can be expressed as a fixed enum → specialist.
+If it requires generating novel tokens → slm or llm.
+
+**`Annotation.h` needs a one-line update** to add `"specialist"` to the strategy
+enum before Phase 2.1 training. This is called out in ROADMAP.md.
+
+### 3. Specialist drives template, not replaces it
+
+The specialist and template tiers work as a pipeline, not alternatives:
+- Specialist picks which template + fills the bounded slot decisions
+- Template deterministically expands those decisions into output
+- LLM handles only the residual open-text portions
+
+### 4. The existing AST annotation schema IS the training target vocabulary
+
+`editor/src/ast/Annotation.h` already defines the output schema for most
+specialists we need to train. Key fields:
+- `AutomatabilityAnnotation.strategy` — the Tacit Codex entropy classifier on nodes
+- `IntentAnnotation.category` — validation/transformation/io/coordination/computation/init
+- `RiskAnnotation.dependentCount` — downstream caller count (input feature)
+- `ComplexityAnnotation.cognitiveComplexity` — 1-10 scale (input feature)
+
+Specialists populate these fields automatically at near-zero cost, replacing
+LLM annotation calls.
+
+---
+
+## Exact Next Steps (Phase 1 + Phase 2.1)
+
+### Immediate: Phase 1 pipeline specialists
+
+Four specialists to train for the taskitem pipeline layer. All follow the same
+pattern as `verification_type` and `confidence_tier`.
+
+**1.1 — `prereq_op_selector` (4-way)**
+Write `scripts/gen_prereq_op_data.py` generating synthetic task descriptions for:
+- `standard` (0): [validate-intake, resolve-dependencies] — default path
+- `needs_review` (1): [validate-intake, architect-review] — breaking/cross-cutting changes
+- `needs_approval` (2): [validate-intake, manual-approval] — experimental/risky tasks
+- `full_gates` (3): [validate-intake, architect-review, manual-approval] — both signals
+Train: `--labels "standard,needs_review,needs_approval,full_gates"`
+
+**1.2 — `worker_type` (4-way)**
+Write `scripts/gen_worker_type_data.py` mining sprint plans with keyword heuristics:
+- `implementer` (0): implement/add/build/write/create
+- `reviewer` (1): review/audit/verify/check/validate
+- `architect` (2): design/architect/plan/structure/refactor
+- `qa` (3): test/coverage/regression/integration summary
+Train: `--labels "implementer,reviewer,architect,qa"`
+
+**1.3 — `gap_class` (5-way)**
+Write `scripts/gen_gap_class_data.py` with synthetic taskitem issue descriptions.
+Classes: under_constrained / environment_gap / planning_gap / cross_project_gap / generator_capability_gap
+
+**1.4 — `escalation_predictor` (binary)**
+Write `scripts/gen_escalation_data.py` from formula: `escalate = confidence < 60 OR ambiguity > 0`.
+Expected 100% accuracy like `confidence_tier`.
+Train: `--labels "proceed,escalate"`
+
+### After Phase 1: Phase 2.1 — `automatability_strategy` (6-way)
+
+**First:** Add `"specialist"` to strategy enum in `editor/src/ast/Annotation.h`.
+
+Write `scripts/gen_automatability_data.py` using the 93 existing `editor/src/*.h`
+headers as labeled examples. Each header encodes the ground-truth tier:
+- Pure formula classes → `deterministic`
+- Schema-to-X converters → `template`
+- Classifiers/scorers with bounded output → `specialist`
+- Text generators with constrained vocabulary → `slm`
+- Open-ended generators → `llm`
+
+Train: `--labels "deterministic,template,specialist,slm,llm,human"`
+
+This is the most architecturally significant specialist — once trained, every node
+in the AST can be tagged with its automatability tier without an LLM call.
+
+---
+
+## What Is NOT Done / Known Gaps
+
+- `infer.py` is a standalone inference tool but not yet wired into any MCP tool
+- No dispatch server yet (`scripts/dispatch.py` — Phase 4.1 in roadmap)
+- `Annotation.h` still has the old 5-tier enum (needs `"specialist"` added)
+- The `verification_type` specialist has 16% error rate concentrated at the
+  schema/unit/integration boundary — the heuristic labeling is imperfect there.
+  Acceptable for now; fix by reviewing ~50 misclassified examples if accuracy matters.
+- JSONL pipeline run records (18,519) are too homogeneous for text-based specialists.
+  Sprint plans are the right source. Do not try to use JSONL for text classifiers.
+
+---
+
+## Connection to WhetstoneAI Fabricate
+
+The Fabricate binary and harness are consumed as-is. Do not modify them.
+
+```
+Binary:   /home/bill/Documents/WhetstoneAI_Fabricate/build/whetstone_fabricate
+Harness:  /home/bill/Documents/WhetstoneAI_Fabricate/run_grokking_until.py
+Eval:     /home/bill/Documents/WhetstoneAI_Fabricate/tools/eval_multiclass.py
+Infer:    /home/bill/Documents/WhetstoneAI_Fabricate/tools/eval_with_vocab.py
+```
+
+The Fabricate venv at `/home/bill/Documents/WhetstoneAI_Fabricate/.venv` has all
+required packages. Use `.venv/bin/python3` for all training and eval calls.
+
+---
+
+## Connection to Tacit Codex
+
+The specialist fleet is the concrete implementation of the Tacit Codex program:
+each trained specialist is a named, deployed piece of collapsed tacit knowledge.
+The `AutomatabilityAnnotation.strategy` field, once populated by the Phase 2.1
+specialist, directly encodes the entropy tier of every AST node — which is the
+per-node Tacit Codex entropy ledger entry.
+
+The same pattern (hard constraints + fuzzy contextual decisions) also applies to
+WHIMP (interaction design) with the 722-metric DuckDB as the input feature vocabulary.
+That connection is documented in the session conversation but not yet in any file.
diff --git a/specialists/HANDOFF-2026-03-31.md b/specialists/HANDOFF-2026-03-31.md
new file mode 100644
index 0000000..162c406
--- /dev/null
+++ b/specialists/HANDOFF-2026-03-31.md
@@ -0,0 +1,93 @@
+# Specialist Fleet Handoff — 2026-03-31
+
+## Where We Are
+
+Switched training infrastructure from Fabricate (C++ binary + SQLite) to **pure PyTorch**.
+Fabricate was causing uninterruptible D-state IO blocking because `/mnt/storage` is a spinning
+HDD and Fabricate writes continuous SQLite WAL data (~100MB/min) even with morphisms disabled.
+
+## What Was Built This Session
+
+### New Files
+- `specialists/scripts/train_specialist_pt.py` — PyTorch training script (replaces `run_grokking_until.py`)
+- `specialists/eval/eval_specialist_pt.py` — PyTorch eval script (replaces `eval_specialist.py`)
+
+### Updated Files
+- `specialists/scripts/capacity_sweep.py` — now calls `train_specialist_pt.py` + `eval_specialist_pt.py`
+- `specialists/scripts/prereq_op_binary_split.py` — same, `evaluate_combined` rewritten in PyTorch
+- Both scripts now use `checkpoint.pt` instead of `checkpoint.bin` + `fabricate.db`
+- TIERS updated: `heads` field added (4/6/8 for small+/medium/large) — PyTorch supports multi-head unlike Fabricate
+
+### PyTorch training script details
+- No SQLite, no WAL. Saves `checkpoint.pt` + `history.json` only.
+- History JSON format compatible with `capacity_sweep.py`: `{"history": [{"step": N, "overall": float, "loss": float}]}`
+- Same CLI interface as `run_grokking_until.py` for drop-in use by sweep scripts
+- Grokking detection + early stop implemented (same logic: loss < threshold OR acc jump ≥ threshold)
+- Checkpoint includes: model weights, optimizer state, vocab, config, labels, step count
+
+## Performance Issue — PARTIALLY FIXED
+
+**Root cause:** DataLoader restarts ~833×/1000 steps when dataset has 300 examples + batch_size=256.
+Each restart has Python overhead → 38s per 1000-step block (expected ~4s).
+
+**Fix applied (untimed):** Load all training data onto GPU once as a tensor, sample via
+`torch.randint` each step instead of DataLoader. This eliminates all Python/CPU overhead per step.
+
+```python
+all_x = torch.stack([row[0] for row in train_ds]).to(device)
+all_y = torch.tensor([row[1] for row in train_ds], ..., device=device)
+# per step:
+idx = torch.randint(0, n_train, (batch_size,), device=device)
+x, y = all_x[idx], all_y[idx]
+```
+
+**This fix is in the code but was NOT timed before session ended.**
+
+## First Task Next Session
+
+1. Time the fixed training script:
+```bash
+cd /home/bill/Documents/CLionProjects/whetstone_DSL
+time /home/bill/Documents/WhetstoneAI_Fabricate/.venv/bin/python3 \
+  specialists/scripts/train_specialist_pt.py \
+  --dataset specialists/data/generated/worker_type_train.tsv \
+  --eval-dataset specialists/data/generated/worker_type_eval.tsv \
+  --labels "implementer,reviewer,architect,qa" \
+  --out-dir /tmp/pt_timing_test \
+  --hidden-dim 256 --layers 2 --heads 4 \
+  --lr 0.001 --weight-decay 0.01 \
+  --max-steps 3000 --batch-size 256 \
+  --stop-after-grokking-blocks 99 \
+  --reset
+```
+
+Expected: ~4-8 seconds per 1000-step block (was 38s before fix), so ~12-25s for 3000 steps.
+If correct, pilot sweep (all 4 gates × 3 tiers) should complete in ~5-10 minutes.
+
+2. If timing is good, run:
+```bash
+bash specialists/scripts/sweep_all_gates.sh --pilot-only
+```
+
+3. Analyze pilot results in `specialists/eval/results/summary.json`, decide which gates
+   need full training vs already good enough.
+
+## Current Baseline Accuracies (small Fabricate models, unchanged)
+- `worker_type`:        88.5%  (implementer residual-category issue, 64% implementer acc)
+- `verification_type`:  84.9%  (schema→integration confusion, capacity limited)
+- `prereq_op`:          69.4%  (standard residual-category, over-fires approval gate)
+- `automatability`:     69.6%  (deterministic 0% — pure residual-category collapse)
+
+Target: 98-99% for all gates. Larger PyTorch models (no multi-head constraint) should help.
+
+## Key Architecture Notes
+- Fabricate constraint: `gpu_transformer` engine only supports `heads=1`
+- PyTorch: any heads where `hidden_dim % heads == 0` — we now use 4/6/8
+- Run dirs are still on `/mnt/storage/fabricate_runs/` — this is fine, `.pt` files are
+  tiny (few MB) and written infrequently (once per 1000 steps), not continuously
+- The old `whetstone_prereq_op_large` run dir on the HDD may still have a D-state
+  Fabricate process associated. Check `pgrep fabricate` at session start; kill if present.
+
+## Venv
+All scripts use: `/home/bill/Documents/WhetstoneAI_Fabricate/.venv/bin/python3`
+PyTorch 2.10+cu128 is installed there. CUDA is available (RTX 3060).
diff --git a/specialists/README.md b/specialists/README.md
new file mode 100644
index 0000000..2ec8b48
--- /dev/null
+++ b/specialists/README.md
@@ -0,0 +1,268 @@
+# Whetstone Specialist Fleet
+
+**What this is:** A system of ~800KB transformer specialists that replace bounded-output
+LLM decisions inside the Whetstone MCP pipeline. Each specialist handles exactly one
+classification or scoring decision, runs in under a millisecond, and can be held
+resident in GPU VRAM alongside thousands of peers.
+
+This is not a LoRA. It is not a fine-tuned 14B model. It is a different architecture
+for a different problem.
+
+---
+
+## Origin: WhetstoneAI Fabricate Experiments
+
+This project grew out of the WhetstoneAI_Fabricate experiments at
+`/home/bill/Documents/WhetstoneAI_Fabricate/`.
+
+Those experiments were training tiny transformer models (~213K parameters) to solve
+structured reasoning problems — math ordering, logic gates, state machines, modular
+arithmetic, DAG traversal. Each model was a binary classifier (true/false) over
+reasoning chains. The key finding was the **Routed-Specialist Architecture (RSA)**:
+
+- A small router model (5-way classifier) reads a raw question and picks which
+  domain expert to consult
+- The chosen specialist (binary classifier) answers the question
+- End-to-end composed accuracy: 96.3% in-distribution, 76.7% OOD after hardening
+- Total model size: router (~800KB) + 5 specialists (~800KB each) = ~5MB
+- Inference: microseconds per query on GPU
+
+The critical realization: **the Whetstone taskitem pipeline is already doing this
+same structure, but with a 14B LLM handling all the decisions.** Every tool call
+involves the LLM choosing from bounded option sets — verification type, prerequisite
+ops, confidence tier, worker role, quality score tier. These are exactly the
+classifier-shaped problems that the Fabricate architecture solves at 800KB.
+
+The WhetstoneAI_Fabricate binary and training harness are reused here unchanged.
+No new training infrastructure was written.
+
+---
+
+## The AST Is Already Rich — Specialists Predict Its Fields
+
+Whetstone is not a C++ code generator. C++, Python, Rust, and all other output
+languages are *projections* of an AST-as-source-of-truth. The AST uses
+language-neutral `conceptType` names (`Function`, `ClassDeclaration`, `IfStatement`)
+that map across 19+ languages through a grammar layer. This means specialists operate
+at the AST level and cover all languages simultaneously — there are no per-language
+variants for decisions that live in the AST.
+
+More importantly: the Whetstone AST already carries annotation metadata that most
+ASTs never have. The annotation schema in `editor/src/ast/Annotation.h` (60+ types)
+includes:
+
+```
+RiskAnnotation.dependentCount         int    how many callers/consumers downstream
+CallHierarchyInfo.callerIds           list   which functions call this one
+BlastRadiusReport.riskLevel           enum   none/low/medium/high (blast radius)
+IntentAnnotation.category             enum   validation/transformation/io/...
+ComplexityAnnotation.cognitiveComplexity  int  1-10 scale
+AutomatabilityAnnotation.strategy     enum   deterministic|template|specialist|slm|llm|human
+AutomatabilityAnnotation.confidence   float  0.0–1.0
+```
+
+NOTE: Current `Annotation.h` is missing `specialist` in this enum. Add it before
+training the `automatability_strategy` specialist. The 6 tiers in order:
+formula → template → specialist (~800KB bounded classifier) → slm → llm → human.
+
+**`AutomatabilityAnnotation.strategy` is the Tacit Codex entropy classification
+built directly into the AST schema.** It encodes whether each node's decisions
+should be handled by a formula, template instantiation, a tiny specialist,
+a small LLM, a large LLM, or a human. Training the `automatability_strategy`
+specialist means populating this field automatically across the entire codebase —
+every node gets its entropy tier assigned without an LLM call.
+
+`RiskAnnotation.dependentCount` is the "how many things downstream depend on this?"
+metric the system already tracks. Combined with blast radius analysis
+(`BlastRadiusAnalyzer.h`) and call hierarchy (`ContextAPI.h`), the AST already
+measures change risk. These are input features for the `change_risk_tier` specialist.
+
+## The Core Premise
+
+### What an MCP tool call currently does
+
+When you call `whetstone_generate_taskitems`, the LLM receives context and makes
+roughly these decisions:
+
+```
+How many milestones?         → integer 1-6
+Verification type?           → unit | integration | schema | smoke | docs
+Prereq ops needed?           → combo of 4 possible ops
+Worker type?                 → implementer | reviewer | architect | qa
+Confidence level?            → integer 0-100 (computed from formula)
+Should escalate?             → bool
+...and then the generative parts:
+Task title text?             → open text
+Acceptance criteria text?    → open text
+```
+
+The first group are **bounded decisions** — the output is always one choice from
+a small fixed set. These are classifier problems. The second group are **generative
+decisions** — the output is open text. These still require the LLM.
+
+### What the specialist fleet does instead
+
+Every bounded decision becomes a specialist call. All specialists fire in parallel
+before or instead of the LLM call. The LLM, if called at all, only handles the
+open-text portions — with the bounded decisions already pre-filled.
+
+```
+[specialist: verification_type]  ──► unit/integration/schema/smoke/docs   ~0.5ms
+[specialist: prereq_op_selector] ──► [validate-intake, resolve-deps]       ~0.5ms
+[specialist: confidence_tier]    ──► high/medium/low                       ~0.5ms
+[specialist: worker_type]        ──► implementer                           ~0.5ms
+                     ↓ all parallel, all in <2ms total
+[LLM: generate task title]       ──► "ResourceLocks schema extension"      ~3s
+[LLM: write acceptance criteria] ──► "..."                                 ~2s
+```
+
+### Why this scales
+
+An RTX 3060 has 12GB VRAM. One quantized 14B LLM takes ~8GB. One specialist takes
+~800KB. The same card that holds one LLM can hold **~14,000 specialists resident
+simultaneously**.
+
+50 bounded decision types × 20 domain variants each = 1,000 specialists = 800MB VRAM.
+All 1,000 can run in a single batched forward pass.
+
+---
+
+## The Philosophical Anchor (Tacit Codex)
+
+This project is an instance of the broader
+[Tacit Codex](../tacit_codex/) work.
+
+The central claim: **"LLM required" is not a permanent category. It is a label for
+decisions we haven't formalized yet.**
+
+Every decision in software generation is either:
+1. **Deterministic** — given the same inputs, always the same output (e.g., filename
+   = `FeatureName.h`, MCP tool name = `whetstone_` + snake_case verb + noun)
+2. **Statistical** — the output is one of N known options, predictable from features
+   (e.g., verification type from step description) → specialist
+3. **"Human mind magic"** — the residual category for decisions that feel non-mechanical
+
+The third category is not permanent. It marks where the decision space hasn't been
+decomposed fine enough yet. "Write a function body" feels like magic until you
+decompose it into: structural pattern (specialist) × error handling style (specialist)
+× data structure choice (specialist) × loop form (specialist) × variable naming
+pattern (specialist), all composed through deterministic template instantiation.
+
+**The LLM is the scaffolding.** It operates at high entropy while the specialists
+are being built. Every LLM decision is an implicit training example for the specialist
+that will eventually replace it. The Tacit Codex entropy ledger tracks which decisions
+are still unformalized — the specialist fleet is what the collapsed version looks like.
+
+---
+
+## Current State (as of 2026-03-28)
+
+Two specialists trained and validated:
+
+| Specialist | Classes | Accuracy | Training data | Checkpoint |
+|-----------|---------|----------|---------------|------------|
+| `verification_type` | unit/integration/schema/smoke/docs (5-way) | **83.7%** | 978 examples from 260 sprint plans | `/mnt/storage/fabricate_runs/whetstone_verification_type/` |
+| `confidence_tier` | high/medium/low (3-way) | **100%** | 1020 synthetic examples from formula | `/mnt/storage/fabricate_runs/whetstone_confidence_tier/` |
+
+The `verification_type` specialist was trained entirely from the sprint plan files
+that already exist in this repo — no external data, no hand-labeling. The sprint
+step descriptions (`### Step NNN: <description>`) are labeled by heuristic rules
+that classify each description into its verification type.
+
+The `confidence_tier` specialist was trained on synthetic data generated from the
+actual formula in `TaskitemConfidenceAmbiguity.h`. It learned the formula exactly.
+
+---
+
+## Quick Start
+
+```bash
+# From whetstone_DSL root
+
+# Generate training data (sprint plans → TSV)
+python3 specialists/scripts/gen_verification_type_data.py --sprint-dir .
+
+# Train
+bash specialists/scripts/train_verification_type.sh
+
+# Infer on a single example
+python3 specialists/scripts/infer.py \
+  --run-dir /mnt/storage/fabricate_runs/whetstone_verification_type \
+  --labels "unit,integration,schema,smoke,docs" \
+  --text "whetstone_detect_conflicts MCP tool"
+# → integration (conf=1.00)
+
+# Interactive REPL (omit --text)
+python3 specialists/scripts/infer.py \
+  --run-dir /mnt/storage/fabricate_runs/whetstone_verification_type \
+  --labels "unit,integration,schema,smoke,docs"
+
+# Evaluate accuracy on a held-out TSV
+python3 specialists/scripts/infer.py \
+  --run-dir /mnt/storage/fabricate_runs/whetstone_verification_type \
+  --labels "unit,integration,schema,smoke,docs" \
+  --eval-tsv specialists/data/generated/verification_type_eval.tsv
+```
+
+---
+
+## Technical Notes for Next Agent
+
+**Training data format** (TSV, required by the Fabricate binary):
+```
+<integer_label>\t<hops>\t<text>
+```
+- `integer_label`: index into `--labels` list (0-based)
+- `hops`: always `0` for Whetstone specialists (unused, required by format)
+- `text`: the input text to classify
+
+**The `--labels` flag** maps integer indices to token strings the model emits.
+`--labels "unit,integration,schema,smoke,docs"` means label 0 = "unit", 1 = "integration", etc.
+**The label column must be an integer, not a string.** String labels cause a crash.
+
+**Fabricate binary**: `/home/bill/Documents/WhetstoneAI_Fabricate/build/whetstone_fabricate`
+**Training harness**: `/home/bill/Documents/WhetstoneAI_Fabricate/run_grokking_until.py`
+**Eval tool**: `/home/bill/Documents/WhetstoneAI_Fabricate/tools/eval_multiclass.py`
+**Inference helpers**: `/home/bill/Documents/WhetstoneAI_Fabricate/tools/eval_with_vocab.py`
+
+**Model architecture** (fixed for all specialists):
+`--hidden-dim 128 --layers 1 --heads 1 --seq-len 96 --batch-size 32`
+This produces ~213K parameters, ~800KB checkpoint.
+
+**Grokking** is the normal training outcome. The model memorizes first (high training
+accuracy, low eval accuracy), then at some step (usually 1000-3000) the eval accuracy
+jumps sharply. The harness detects this and stops automatically. Do not be alarmed by
+"Fresh start" at the top of training logs — this means no prior checkpoint exists in
+the output dir, not that the weights are random (they're initialized from scratch as
+intended).
+
+**Known issue from RSA work**: If you ever resume training a specialist from an
+existing checkpoint while changing the `--labels` order, the token rows in the
+checkpoint will be misaligned. Use
+`/home/bill/Documents/WhetstoneAI_Fabricate/tools/swap_label_rows.py` to fix this
+before resuming. See `WhetstoneAI_Fabricate/HANDOFF-2026-03-28-RSA.md` for the
+full story of this bug.
+
+---
+
+## Files in This Directory
+
+```
+specialists/
+  README.md                    ← this file
+  ROADMAP.md                   ← phased implementation plan
+  scripts/
+    gen_verification_type_data.py   extract sprint plans → labeled TSV
+    gen_confidence_tier_data.py     generate synthetic formula data → TSV
+    train_verification_type.sh      train verification_type specialist
+    train_confidence_tier.sh        train confidence_tier specialist
+    infer.py                        single/batch/eval/REPL inference
+  data/
+    generated/                 TSVs produced by gen_* scripts (committed)
+    mined/                     data extracted from JSONL pipeline runs (future)
+  docs/
+    ARCHITECTURE.md            full decision-point inventory + design rationale
+  runs/
+    *_history.json             grokking history per specialist (committed)
+    *_training.log             full training logs (gitignore these)
+```
diff --git a/specialists/ROADMAP.md b/specialists/ROADMAP.md
new file mode 100644
index 0000000..2cad5e6
--- /dev/null
+++ b/specialists/ROADMAP.md
@@ -0,0 +1,388 @@
+# Whetstone Specialist Fleet — Roadmap
+
+**Read README.md first.** This document assumes you understand the core premise.
+
+**Guiding principle:** "LLM required" is not a permanent category. Every decision is
+either deterministic, statistical, or not yet decomposed. This roadmap is the plan
+for collapsing the "not yet decomposed" pile, one specialist at a time.
+
+**Architectural correction (important):** Whetstone is NOT a C++ code generator.
+C++, Python, Rust, and all other output languages are *projections* of an
+AST-as-source-of-truth. The AST uses language-neutral `conceptType` names
+(`Function`, `ClassDeclaration`, `IfStatement`, etc.) that map uniformly across
+19+ target languages. Specialists operate at the AST level — one specialist per
+decision type covers all projection languages. Do not train per-language variants
+of decisions that live in the AST layer.
+
+---
+
+## The AST Is Already Rich
+
+Whetstone's AST is not a minimal syntax tree. It carries annotation metadata that
+most derived ASTs never have. Key files:
+
+- `editor/src/ast/ASTNode.h` — base node with `conceptType`, `semanticHash`, role-based children
+- `editor/src/ast/Annotation.h` — 60+ annotation types (see Subject breakdown below)
+- `editor/src/SemanticCoreIR.h` — cross-language IR with `intentTags`
+- `editor/src/SemanticTags.h` — 10-tag vocabulary (`@serialize`, `@crypto`, `@io`, etc.)
+- `editor/src/ContextAPI.h` — `CallHierarchyInfo` with `callerIds`/`calleeIds`
+- `editor/src/graduation/BlastRadiusAnalyzer.h` — impact scoring
+- `editor/src/DependencyGraph.h` — dependency edges and critical path
+
+### The metadata that matters for specialists
+
+**Downstream impact** — already computed, available as input features:
+```
+RiskAnnotation.dependentCount    int    how many callers/consumers
+CallHierarchyInfo.callerIds      list   which functions call this one
+BlastRadiusReport.riskLevel      enum   none/low/medium/high
+BlastRadiusEntry.impactScore     int    numerical impact
+FailureTriageScore.blastRadius   int    propagation count
+```
+
+**Intent and complexity** — already stored on nodes:
+```
+IntentAnnotation.category        enum   validation/transformation/io/
+                                        coordination/computation/initialization
+IntentAnnotation.summary         text   1-sentence description
+ComplexityAnnotation.timeComplexity     O(1)/O(n)/O(n^2)/...
+ComplexityAnnotation.cognitiveComplexity  1-10 scale
+SemanticTagAnnotation.tags       list   @serialize/@crypto/@io/@math/etc.
+```
+
+**The Tacit Codex annotation — already in the schema:**
+```
+AutomatabilityAnnotation.strategy  enum   deterministic | template | slm | llm | human
+AutomatabilityAnnotation.confidence  float  0.0–1.0
+```
+
+This annotation encodes the entire Tacit Codex hierarchy on individual AST nodes.
+A specialist that predicts `automatability.strategy` for a given node *is* the
+entropy classifier the Tacit Codex is building toward — it just needs to be trained.
+
+---
+
+## Phase 0 — Foundation (COMPLETE)
+
+**Done:**
+- `verification_type` specialist: 5-way, 83.7% accuracy
+  - Input: sprint step description text
+  - Output: unit / integration / schema / smoke / docs
+  - Checkpoint: `/mnt/storage/fabricate_runs/whetstone_verification_type/`
+- `confidence_tier` specialist: 3-way, 100% accuracy
+  - Input: `"conflicts=N ambiguity=M deps=D prereqs=P queueready=yes/no"`
+  - Output: high / medium / low
+  - Checkpoint: `/mnt/storage/fabricate_runs/whetstone_confidence_tier/`
+- Inference wrapper: `scripts/infer.py`
+
+**Key finding from Phase 0:** Sprint plan step descriptions are the right training
+data source for text-based specialists. The 18,519 JSONL pipeline run records are
+too homogeneous (2 task patterns repeated). Do not use JSONL as the text source.
+
+---
+
+## Phase 1 — Pipeline Specialists (NEXT)
+
+**Goal:** Cover the remaining bounded decisions in `generate_taskitems`
+and `validate_taskitem`. These are at the taskitem planning layer, above the AST.
+
+### 1.1 — `prereq_op_selector` (4-way)
+
+Input: task step description text
+Output:
+```
+0: standard       → [validate-intake, resolve-dependencies]
+1: needs_review   → [validate-intake, architect-review]
+2: needs_approval → [validate-intake, manual-approval]
+3: full_gates     → [validate-intake, architect-review, manual-approval]
+```
+Training data: synthetic — generate task descriptions that signal each gate type.
+Script to write: `scripts/gen_prereq_op_data.py`
+
+### 1.2 — `worker_type` (4-way)
+
+Input: task step description text
+Output: implementer / reviewer / architect / qa
+Training data: sprint plan files (same heuristic approach as `verification_type`).
+Script to write: `scripts/gen_worker_type_data.py`
+
+### 1.3 — `gap_class` (5-way)
+
+Input: taskitem issue description text
+Output: under_constrained / environment_gap / planning_gap / cross_project_gap / generator_capability_gap
+Training data: synthetic — generate issue descriptions encoding each gap type.
+Script to write: `scripts/gen_gap_class_data.py`
+
+### 1.4 — `escalation_predictor` (binary)
+
+Input: `"confidence=N ambiguity=M"` structural encoding
+Output: escalate / proceed
+Training data: synthetic from formula (`escalate = confidence < 60 OR ambiguity > 0`).
+Expected accuracy: 100% (deterministic formula, same as `confidence_tier`).
+Script to write: `scripts/gen_escalation_data.py`
+
+---
+
+## Phase 2 — AST Annotation Specialists
+
+**Goal:** Train specialists that predict annotation fields on AST nodes.
+These operate one level deeper than the taskitem pipeline — they answer questions
+about individual AST nodes, not about sprint planning.
+
+The annotation system already defines the output schema. Specialists populate
+annotation fields that are currently filled by the LLM or left empty.
+
+### 2.1 — `automatability_strategy` specialist (THE KEY ONE) — (6-way)
+
+**This is the Tacit Codex entropy classifier applied to AST nodes.**
+
+**IMPORTANT — update `editor/src/ast/Annotation.h` first.** The current enum
+(`deterministic|template|slm|llm|human`) is missing the `specialist` tier.
+The correct 6-tier hierarchy is:
+
+```
+deterministic   same input → same output always. Formula. Zero parameters.
+template        deterministic composition of known structural patterns. No ambiguity.
+specialist      bounded output vocabulary (fixed N classes). Fuzzy/probabilistic.
+                ~800KB model, microseconds — THIS PROJECT's outputs.
+slm             open vocabulary, arbitrary token generation. ~1-7GB.
+llm             full generative, cross-context, novel reasoning. ~14B+.
+human           not yet formalizable.
+```
+
+Boundary between `specialist` and `slm`: can the output be a fixed enum?
+Yes → specialist. Requires generating novel tokens → slm or llm.
+Boundary between `template` and `specialist`: deterministic? Yes → template.
+Probabilistic with a confidence score → specialist.
+
+Add `"specialist"` to the strategy field in `Annotation.h` before training.
+
+Input features (all from existing AST/annotation metadata):
+```
+conceptType          (Function/Class/Variable/EnumDeclaration/...)
+intentCategory       (validation/transformation/io/coordination/computation/init)
+dependentCount       (0-100+ from RiskAnnotation)
+semanticTags         (space-separated: @io @serialize @math...)
+cognitiveComplexity  (1-10 from ComplexityAnnotation)
+hasBody              (yes/no — skeleton vs implemented)
+```
+Encoded as: `"type=Function intent=validation deps=12 tags=@io @serialize complexity=4 body=yes"`
+
+Output: deterministic / template / **specialist** / slm / llm / human
+
+**Why this is the most important specialist:**
+Once trained, every node in the AST can be tagged with its automatability tier.
+The pipeline routes each node's decisions to the appropriate handler:
+- `deterministic` → formula/rule, no inference at all
+- `template` → template instantiation, near-zero cost
+- `specialist` → specialist fleet dispatch (~0.5ms)
+- `slm` → small model call (~500ms)
+- `llm` → full LLM call (~2-5s)
+- `human` → escalate via ReviewAnnotation
+
+This replaces the current situation where every decision defaults to "llm" regardless.
+
+Training data: The 93 existing `editor/src/*.h` headers are labeled examples.
+Each header class can be manually categorized. Starting corpus: ~93 examples
+(thin but the patterns are strong — expect grokking with heavy oversampling).
+Future: every new header added to the codebase extends the training set.
+
+Script to write: `scripts/gen_automatability_data.py`
+
+### 2.2 — `intent_category` specialist (6-way)
+
+Input: node name + semantic tags + parent conceptType (all language-neutral)
+Output: validation / transformation / io / coordination / computation / initialization
+
+Training data: mine existing headers. The class name + role suffix strongly predicts
+the category:
+- `*Validator`, `*Scorer`, `*Auditor` → validation
+- `*Generator`, `*Builder`, `*Composer` → transformation
+- `*Loader`, `*Writer`, `*Exporter` → io
+- `*Dispatcher`, `*Router`, `*Registry` → coordination
+- `*Calculator`, `*Analyzer`, `*Solver` → computation
+- `*Initializer`, `*Factory`, `*Bootstrap` → initialization
+
+This is trainable from class name patterns alone. 93 headers = 93 examples to start.
+Script to write: `scripts/gen_intent_category_data.py`
+
+### 2.3 — `change_risk_tier` specialist (4-way)
+
+Input: `"deps=N callers=M impactScore=K blastRadius=R"` (all from existing metadata)
+Output: low / medium / high / critical
+
+This specialist uses the blast radius and downstream dependency counts as its
+primary input signal. A function with `deps=47` and `blastRadius=12` is high risk.
+A function with `deps=0` and `callers=1` is low risk.
+
+Training data: synthetic from rules:
+- critical: deps > 20 OR callerIds.count > 15
+- high: deps 10-20 OR callers 8-15
+- medium: deps 3-10 OR callers 2-8
+- low: deps 0-3, callers 0-2
+
+Script to write: `scripts/gen_change_risk_data.py`
+Expected accuracy: ~100% (formula-based, same as `confidence_tier`).
+
+### 2.4 — `semantic_tag` specialist (multi-label → primary tag, 10-way)
+
+Input: node name text (language-neutral conceptual name)
+Output: primary semantic tag from vocabulary:
+  @serialize / @crypto / @io / @network / @math / @collection /
+  @concurrency / @test / @ui / @parse
+
+The `SemanticTags.h` file already has auto-tagging rules based on library names
+and identifier keywords. This specialist learns the same rules from examples.
+
+Training data: mine existing headers using the auto-tagger rules as labels.
+Script to write: `scripts/gen_semantic_tag_data.py`
+
+### 2.5 — `component_suffix` specialist (12-way)
+
+Input: 1-sentence description of what a component does
+Output: naming suffix that fits the role:
+  Validator / Scorer / Auditor / Generator / Processor / Selector /
+  Advisor / Binder / Extractor / Composer / Builder / Registry
+
+Training data: 93 existing `editor/src/*.h` headers as labeled examples.
+Each filename encodes the ground-truth suffix. Thin data but strong patterns.
+Script to write: `scripts/gen_component_suffix_data.py`
+
+---
+
+## Phase 3 — Data Collection (Running in Background)
+
+**Goal:** Make every LLM tool call a training data generator.
+
+### 3.1 — Instrument bounded sub-decisions
+
+Add post-processing to `tools/mcp/export_taskitem_run_for_lora.sh` to extract
+per-decision labels from each pipeline run. When `generate_taskitems` emits a task
+with `verificationType: "integration"`, append `(step_description, "integration")`
+to `training_data/specialists/verification_type_examples.jsonl`.
+
+### 3.2 — Annotation capture hook
+
+When the LLM fills an `AutomatabilityAnnotation` or `IntentAnnotation` on any
+AST node, capture `(input_features, annotation_value)` as a training example.
+These accumulate automatically with normal usage.
+
+### 3.3 — Blast radius instrumentation
+
+When `BlastRadiusAnalyzer` runs, emit `(dependentCount, blastRadius, callerCount) → riskLevel`
+as a labeled example. This grows the `change_risk_tier` training set with every
+real analysis run.
+
+---
+
+## Phase 4 — Integration
+
+**Goal:** Wire specialists into the MCP tool layer.
+
+### 4.1 — Dispatch server
+
+`scripts/dispatch.py` — loads all trained specialists at startup, accepts JSON,
+returns predictions. All specialists run in a single batched forward pass.
+
+### 4.2 — Pre-annotation pass
+
+New MCP tool: `whetstone_annotate_nodes` — given a list of AST node descriptors,
+runs the specialist fleet and returns predicted annotation values before the LLM
+is called. The LLM then only handles open-text fields and `human`-tier decisions.
+
+### 4.3 — Automatability routing
+
+Once `automatability_strategy` specialist is trained and validated:
+- `deterministic` nodes: handled by formula/rule engine, no inference
+- `template` nodes: handled by template instantiation
+- `slm` nodes: handled by specialist dispatch
+- `llm` nodes: LLM call with pre-filled context from specialists
+- `human` nodes: escalated via `ReviewAnnotation.required = true`
+
+---
+
+## Phase 5 — Code Generation Specialists
+
+**Goal:** Decompose "generative" decisions into bounded sub-decisions.
+
+Once Phase 2 establishes the full annotation vocabulary, code generation becomes:
+
+```
+Node with annotations
+    → automatability_strategy = "template"         ← specialist decides
+    → intent_category = "transformation"           ← specialist decides
+    → structural_pattern = "pipeline"              ← specialist decides (new)
+    → control_flow_pattern = "early-return"        ← specialist decides (new)
+    → error_handling_style = "Result<T,E>"         ← specialist decides (new)
+    → projection language = "cpp"                  ← human or router decides
+    ↓
+Deterministic template instantiation
+    ↓
+Code skeleton
+    ↓
+LLM fills only: identifier names, literal values, comments  ← narrow residual
+```
+
+Specialists for this phase:
+- `structural_pattern` (factory/visitor/strategy/builder/observer/pipeline/scorer/registry...)
+- `control_flow_pattern` (linear/if-else-chain/early-return/loop/recursive/dispatch-table...)
+- `error_handling_style` (exception/Result/optional/expected/assert/error-code...)
+- `data_structure_choice` (vector/map/deque/set/flat-array/ring-buffer...)
+
+Each of these is a bounded classification over the AST node's role and annotation
+context. Each can be trained from the existing `editor/src/` headers as examples.
+
+---
+
+## Tracking Progress
+
+| Specialist | Phase | Status | Accuracy | Checkpoint |
+|-----------|-------|--------|----------|------------|
+| `verification_type` | 0 | ✅ DONE | 83.7% | `/mnt/storage/fabricate_runs/whetstone_verification_type/` |
+| `confidence_tier` | 0 | ✅ DONE | 100% | `/mnt/storage/fabricate_runs/whetstone_confidence_tier/` |
+| `prereq_op_selector` | 1.1 | ⬜ planned | — | — |
+| `worker_type` | 1.2 | ⬜ planned | — | — |
+| `gap_class` | 1.3 | ⬜ planned | — | — |
+| `escalation_predictor` | 1.4 | ⬜ planned | — | — |
+| `automatability_strategy` | 2.1 | ⬜ planned (6-way, needs Annotation.h update first) | — | — |
+| `intent_category` | 2.2 | ⬜ planned | — | — |
+| `change_risk_tier` | 2.3 | ⬜ planned | — | — |
+| `semantic_tag` | 2.4 | ⬜ planned | — | — |
+| `component_suffix` | 2.5 | ⬜ planned | — | — |
+| `structural_pattern` | 5 | ⬜ planned | — | — |
+| `control_flow_pattern` | 5 | ⬜ planned | — | — |
+| `error_handling_style` | 5 | ⬜ planned | — | — |
+
+---
+
+## What the Next Agent Should Do First
+
+1. **Read `README.md`** and this file completely
+2. **Verify Phase 0 specialists** by running `infer.py --eval-tsv` on both held-out sets
+3. **Start Phase 2.1** (`automatability_strategy`) — it is the most architecturally
+   significant specialist and training data already exists in `editor/src/*.h`
+4. **Then Phase 1.1** (`prereq_op_selector`) to complete the pipeline layer
+5. **Do not train per-language specialist variants** — the AST is language-neutral;
+   projection is handled downstream by deterministic grammar rules
+
+---
+
+## Connection to Other Projects
+
+**WhetstoneAI_Fabricate** (`/home/bill/Documents/WhetstoneAI_Fabricate/`):
+Training infrastructure. Binary, harness, eval tools. The RSA experiments
+(see `HANDOFF-2026-03-28-RSA.md`) validated the specialist architecture at scale.
+The label convention fix (`swap_label_rows.py`) is documented there — read it
+before resuming any checkpoint with a changed `--labels` order.
+
+**Tacit Codex** (`tacit_codex/`):
+`AutomatabilityAnnotation.strategy` in `editor/src/ast/Annotation.h` is the
+Tacit Codex entropy classification built into the AST schema. Training the
+`automatability_strategy` specialist is the concrete implementation of the
+Tacit Codex collapse-to-formal-structure program applied to individual AST nodes.
+The entropy ledger in HiveMind tracks the same quantity at the job-scheduling layer.
+
+**HiveMind** (`/home/bill/Documents/hivemind/`):
+Future state: a `generate_taskitems` job becomes a fan-out of N specialist jobs
++ 1 LLM job (only for `llm`-tier nodes), merged by coordinator. The automatability
+tier determines which job type each node gets dispatched to.
diff --git a/specialists/data/combined/prereq_op_eval.tsv b/specialists/data/combined/prereq_op_eval.tsv
new file mode 100644
index 0000000..8c76cff
--- /dev/null
+++ b/specialists/data/combined/prereq_op_eval.tsv
@@ -0,0 +1,36 @@
+2	0	`whetstone_enqueue_pair_upgrade` MCP tool
+3	0	Deprecation/succession lifecycle framework
+3	0	Security scan orchestrator for generated targets
+1	0	Phase 12c Integration Tests
+2	0	Reviewer quorum policy model for critical approvals
+1	0	Phase 38a Integration
+1	0	Phase 23b Integration + Sprint Summary
+0	0	Reprojection safety policy
+2	0	Stable corpus certification suite
+0	0	Timing regression runner integration
+0	0	`whetstone_get_tool_permissions` MCP tool
+3	0	Cross-language rollout state snapshot model
+1	0	Compatibility policy profiles by interface criticality
+3	0	`whetstone_publish_roadmap_epoch` MCP tool
+2	0	Security Report Generation
+2	0	Compatibility/deprecation report artifact
+1	0	Human Review Interface via MCP
+2	0	HiveMind job publisher from editor
+1	0	Parser/AST adapter interface schema
+0	0	Sprint 76 integration summary + regression
+1	0	Scenario — Multi-Model Orchestration
+0	0	Autonomous loop controller with “green or explicit blocked” contract
+2	0	Security-Aware Routing
+0	0	Verifier trust weighting policy engine
+0	0	Define verification and readiness surface | Build distributed rate limiter with fixed/sliding window modes and policy APIs.
+1	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+1	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+1	0	Define verification and readiness surface | Refactor billing pipeline to introduce idempotency keys and rollback checkpoints across ledger service, queue worker, re
+2	0	Plan Replace search ranking pipeline behind feature flag with | Replace search ranking pipeline behind feature flag with staged rollout, telemetry checks, and automatic abort trigger.
+0	0	Define verification and readiness surface | Build a JWT auth service with register, login, refresh token rotation, and password hashing.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
diff --git a/specialists/data/combined/prereq_op_train.tsv b/specialists/data/combined/prereq_op_train.tsv
new file mode 100644
index 0000000..83505ce
--- /dev/null
+++ b/specialists/data/combined/prereq_op_train.tsv
@@ -0,0 +1,204 @@
+2	0	`whetstone_get_constructive_rollout_status` runtime implementation
+1	0	Target projection blocker taxonomy
+2	0	Strict gate mode in production loop
+3	0	`whetstone_get_polyglot_cutover_readiness` MCP tool
+3	0	Security scan orchestrator for generated targets
+1	0	Multi-language constructive loop report artifact
+2	0	`whetstone_get_constructive_rollout_status` MCP tool
+3	0	`whetstone_get_polyglot_cutover_readiness` MCP tool
+2	0	Deprecation policy model for unstable/obsolete pairs
+0	0	Projection schema extension for run catalogs
+0	0	Panel Registration System
+0	0	Sprint 146 integration summary + regression
+3	0	`whetstone_publish_roadmap_epoch` MCP tool
+1	0	Distributed path in FeedbackLoopOrchestrator
+1	0	Continuity drill framework and scoring
+3	0	Cross-language release evidence aggregator model
+0	0	Scheme lowering/raising adapters
+3	0	Cross-language release evidence aggregator model
+3	0	`whetstone_get_polyglot_cutover_readiness` MCP tool
+1	0	Phase 13a Integration — Layout Persistence
+2	0	Security Testing Skeleton Generation
+0	0	Modifier-Edge Shortcut Notation
+1	0	Architecture Templates
+3	0	Deprecation/succession lifecycle framework
+1	0	Language-specific method body scaffolding for core patterns
+1	0	Initializer Lists + STL Patterns
+3	0	Security scan orchestrator for generated targets
+0	0	`whetstone_dry_run_patch` MCP tool
+3	0	Cross-language rollout state snapshot model
+3	0	Security scan orchestrator for generated targets
+0	0	Cost telemetry integration in metrics stack
+2	0	Constitutional invariants compliance checker
+0	0	Property-based equivalence runner
+2	0	Regional rollout profile integration
+0	0	Compile/test proof binding
+3	0	Cross-language release evidence aggregator model
+2	0	`whetstone_get_certification_status` MCP tool
+3	0	Deprecation/succession lifecycle framework
+0	0	Sprint 193 Integration Summary
+1	0	Multi-language replay corpus runtime index model
+2	0	Benchmark suite for production capability claim
+3	0	`whetstone_publish_roadmap_epoch` MCP tool
+0	0	Build output parser — GCC/Clang format
+2	0	Rollout stage model (`pilot`, `canary`, `broad`, `default`)
+2	0	`whetstone_set_language_rollout_tier` MCP tool
+0	0	Breadcrumb Navigation
+1	0	Runtime-aware lowering extension points
+1	0	Phase 15b Integration — Full Protocol Test
+3	0	`whetstone_get_polyglot_cutover_readiness` MCP tool
+0	0	Python lowering adapter v2 (dynamic shape packets)
+0	0	Canonical fixture and oracle dataset packs
+0	0	Project context auto-injection
+0	0	LTS baseline criteria schema
+3	0	Security scan orchestrator for generated targets
+0	0	`whetstone_validate_tool_contracts` MCP tool
+1	0	Phase 20a Integration
+3	0	Cross-language release evidence aggregator model
+2	0	`whetstone_publish_next_block_plan` MCP tool
+2	0	Staged cutover planner (`shadow`, `dual-run`, `cutover`)
+3	0	Security scan orchestrator for generated targets
+3	0	Security scan orchestrator for generated targets
+0	0	Locals + Watches Panel Data Model
+0	0	Context Bundle Minimizer
+2	0	`whetstone_get_compliance_evidence` MCP tool
+3	0	`whetstone_get_polyglot_cutover_readiness` MCP tool
+2	0	Security Annotation Types — Subject 10
+0	0	Benchmark scorecard generator per pair
+1	0	Runtime-specific verification plugin interface
+2	0	Certification job scheduler
+3	0	Deprecation/succession lifecycle framework
+2	0	Tiered rollout and rollback control model
+2	0	Sprint 60 integration summary + release marker
+3	0	Cross-language rollout state snapshot model
+0	0	Constraint Violation Diagnostics
+1	0	Phase 12a Integration Tests
+3	0	Cross-language rollout state snapshot model
+1	0	Deterministic blocker taxonomy extension
+0	0	`whetstone_get_regeneration_decisions` MCP tool
+1	0	Long-horizon roadmap planning engine v2
+0	0	Intake extraction confidence scoring
+1	0	Phase 18b Integration + Sprint Summary
+3	0	`whetstone_get_polyglot_cutover_readiness` MCP tool
+2	0	Certification badges for roles/workflows
+0	0	Closeout consistency gate
+3	0	Cross-language rollout state snapshot model
+1	0	Binary/interface diff engine integration
+2	0	Release Readiness Gate Pack
+0	0	Source-vs-runtime drift comparator model
+1	0	Phase 12d Integration + Sprint 12 Summary
+1	0	`whetstone_verify_architecture_consistency` MCP tool
+0	0	API/ABI preservation report artifact
+0	0	Sprint 164 Integration Summary
+3	0	`whetstone_publish_roadmap_epoch` MCP tool
+0	0	`whetstone_capture_mcp_execution_trace` MCP tool
+0	0	Template recommendation model by project fingerprint
+0	0	Kotlin Generator
+3	0	Deprecation/succession lifecycle framework
+1	0	Cross-language capability contract normalizer model
+2	0	Release certification packet model
+2	0	Rollout blast-radius estimator
+3	0	Cross-language release evidence aggregator model
+2	0	Tenant-scoped waiver/approval ledger
+1	0	Governance model refresh framework
+2	0	`whetstone_publish_next_epoch_plan` MCP tool
+3	0	Cross-language release evidence aggregator model
+0	0	Embedded migration risk report template
+1	0	Cross-language node identity preservation policy model
+0	0	Key Symbol Rendering
+1	0	Burn-down planning optimizer
+3	0	Deprecation/succession lifecycle framework
+2	0	Interop certification policy bindings
+3	0	Cross-language release evidence aggregator model
+3	0	Deprecation/succession lifecycle framework
+2	0	Production summary hardening
+3	0	`whetstone_publish_roadmap_epoch` MCP tool
+2	0	Safety-critical deployment dossier artifact
+3	0	`whetstone_publish_roadmap_epoch` MCP tool
+2	0	Security-Preserving Translation
+1	0	Swarm maintenance job decomposition model
+3	0	`whetstone_get_polyglot_cutover_readiness` MCP tool
+1	0	Polyglot project migration graph model
+1	0	Phase 10d Integration Tests
+3	0	Cross-language rollout state snapshot model
+1	0	Phase 18a Integration
+2	0	Runtime-backed constructive release certification packet model
+1	0	Phase 29a Integration
+2	0	Production loop contract preflight
+1	0	Phase 20b Integration + Sprint Summary
+2	0	Pair-upgrade queue model
+2	0	`whetstone_get_tool_deprecations` MCP tool
+0	0	Wire `ArchitectIntakeProcessor` into `whetstone_architect_intake` handler
+2	0	Contract drift classifier (`compatible`, `warning`, `breaking`)
+1	0	Phase 23c Integration (tests: integration, regression)
+1	0	`FeedbackLoopOrchestrator` core loop
+1	0	WAT Cross-Language Projection
+0	0	Sprint 187 Integration Summary
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Define verification and readiness surface | Build chat bot backend with slash commands, permission checks, and scheduled digests.
+1	0	Define verification and readiness surface | Build an order saga orchestrator with compensating actions, timeout escalation, exactly-once message handling, and deter
+2	0	Define verification and readiness surface | Build a CMS backend with posts, tags, draft/publish workflow, and role-based access control.
+0	0	Define verification and readiness surface | Build a multi-tenant limiter with token bucket + sliding window hybrid policy, burst fairness guarantees, and lock-free 
+1	0	Plan Build a CI orchestrator service triggering jobs tracking | Build a CI orchestrator service triggering jobs, tracking states, and emitting summaries.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+2	0	Define verification and readiness surface | Build a CLI password vault with encrypted local storage and master-key unlock.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Define verification and readiness surface | Build a release notes generator from commits, PR labels, and issue references.
+2	0	Define verification and readiness surface | Build a secrets rotation scheduler with lease tracking, retry plans, and notifications.
+1	0	Define verification and readiness surface | Build an order saga orchestrator with compensating actions, timeout escalation, exactly-once message handling, and deter
+2	0	Define verification and readiness surface | Build a model registry API supporting stages, approvals, rollback, and signed artifact links.
+1	0	Define verification and readiness surface | Refactor billing pipeline to introduce idempotency keys and rollback checkpoints across ledger service, queue worker, re
+1	0	Define verification and readiness surface | Refactor billing pipeline to introduce idempotency keys and rollback checkpoints across ledger service, queue worker, re
+1	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+1	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Define verification and readiness surface | Build inventory endpoints with stock reservation, release, and low-stock alerts.
+0	0	Define verification and readiness surface | Build form builder backend with schema validation, submissions, and analytics counters.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+2	0	Define verification and readiness surface | Build fraud rules engine with rule evaluation graph, score thresholds, and audit trails.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+2	0	Define verification and readiness surface | Build a PII redaction pipeline masking emails, phones, and IDs with audit logging.
+1	0	Plan Build a CI orchestrator service triggering jobs tracking | Build a CI orchestrator service triggering jobs, tracking states, and emitting summaries.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+1	0	Plan Build a CI orchestrator service triggering jobs tracking | Build a CI orchestrator service triggering jobs, tracking states, and emitting summaries.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+1	0	Define verification and readiness surface | Build an order saga orchestrator with compensating actions, timeout escalation, exactly-once message handling, and deter
+0	0	Define verification and readiness surface | Build experiment tracking backend storing params, metrics, artifacts, and compare queries.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+0	0	Define verification and readiness surface | Build a .env manager CLI with profile switch, validation, and secure redaction in output.
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Define verification and readiness surface | Build a secret rotation controller with staged rollout, canary verification, rollback triggers, and explicit blast-radiu
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+1	0	Plan Build a CI orchestrator service triggering jobs tracking | Build a CI orchestrator service triggering jobs, tracking states, and emitting summaries.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build tax report backend aggregating gains/losses and generating filing-year summaries.
+1	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build a realtime log viewer backend with tail sessions, regex filters, and backpressure.
+1	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+0	0	Define verification and readiness surface | Build job queue service with delayed jobs, retries, visibility timeout, and metrics.
+1	0	Define verification and readiness surface | Refactor billing pipeline to introduce idempotency keys and rollback checkpoints across ledger service, queue worker, re
+1	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+1	0	Define verification and readiness surface | Build an order saga orchestrator with compensating actions, timeout escalation, exactly-once message handling, and deter
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Define verification and readiness surface | Build audit log service with immutable events, filtering, export, and retention policies.
+2	0	Define verification and readiness surface | Build LLM guardrail service for policy checks, PII filtering, and response classification.
+0	0	Define verification and readiness surface | Build telemetry collector with trace/span ingestion, sampling, and export adapters.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Define verification and readiness surface | Build a secret rotation controller with staged rollout, canary verification, rollback triggers, and explicit blast-radiu
+2	0	Plan Upgrade API contract from v1 to v2 backend | Upgrade API contract from v1 to v2: backend handlers, OpenAPI schema, generated client SDK, frontend integration, and co
+2	0	Define verification and readiness surface | Build configuration service with versioned configs, staged rollout, and validation checks.
+2	0	Define verification and readiness surface | Build access control service with RBAC roles, permissions, and resource-scoped grants.
+2	0	Define verification and readiness surface | Build a feature flag API with environments, targeting rules, and rollout percentages.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+0	0	Define verification and readiness surface | Build an event deduplicator using idempotency keys, time windows, and replay-safe writes.
+1	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
diff --git a/specialists/data/generated/automatability_eval.tsv b/specialists/data/generated/automatability_eval.tsv
new file mode 100644
index 0000000..356e6e6
--- /dev/null
+++ b/specialists/data/generated/automatability_eval.tsv
@@ -0,0 +1,46 @@
+2	0	name=LibrarySymbolAdvisor
+3	0	name=SelfHostTranspileAudit
+1	0	name=TrainingDataGenerator
+3	0	name=IntentTranslator
+3	0	name=TypeSystemTranslator
+0	0	name=ContextBundleMinimizer
+3	0	name=SecurityPreservingTranslation
+3	0	name=IncrementalOptimizer
+0	0	name=BufferOpsLifecycle
+0	0	name=MigrationAcceptanceContract
+5	0	name=ApprovalEscalationPlanner
+0	0	name=Sprint163IntegrationSummary
+5	0	name=ArchitectReviewInterface
+3	0	name=MemoryModelTranslator
+4	0	name=AgentCodeGen
+0	0	name=Sprint238IntegrationSummary
+1	0	name=CppJSBindingEmitter
+2	0	name=OperationClassifier
+0	0	name=Sprint36IntegrationSummary
+0	0	name=SelfHostHarness
+2	0	name=SymbolSelectorAPI
+5	0	name=ArchitectReviewSurface
+0	0	name=Sprint111IntegrationSummary
+2	0	name=LibrarySymbolAdvisor
+2	0	name=TaskitemLibraryAnnotator
+1	0	name=SyntaxHighlighterElisp
+0	0	name=SelfHostCodebaseAudit
+2	0	name=PerTaskLibrarySelector
+1	0	name=SyntaxHighlighterGo
+4	0	name=AgentRegistry
+2	0	name=OperationClassifier
+1	0	name=WorkflowTemplates
+5	0	name=AgentMutationApproval
+5	0	name=AgentMutationPreview
+1	0	name=RustPythonBindingEmitter
+3	0	name=TransformEngine
+2	0	name=OperationClassifier
+0	0	name=CompilerPhaseSpec
+5	0	name=ArchitectReviewSurface
+2	0	name=OperationSelectorAPI
+5	0	name=AgentMutationPreview
+5	0	name=AgentMutationApproval
+4	0	name=AgentTaskStatusOverlay
+4	0	name=ScopeMilestoneDecomposer
+5	0	name=SecurityExceptionReviewBoard
+1	0	name=CompositionBuilder
diff --git a/specialists/data/generated/automatability_train.tsv b/specialists/data/generated/automatability_train.tsv
new file mode 100644
index 0000000..9555984
--- /dev/null
+++ b/specialists/data/generated/automatability_train.tsv
@@ -0,0 +1,266 @@
+4	0	name=AgentChatContextInjector
+5	0	name=ApprovalEscalationPlanner
+3	0	name=StrategyAwareOptimizer
+0	0	name=SettingsManager
+2	0	name=LibrarySymbolAdvisor
+3	0	name=TransformEngineExtended
+3	0	name=StrategyAwareOptimizer
+2	0	name=SymbolSelectorAPI
+5	0	name=ApprovalEscalationPlanner
+0	0	name=PolyApiProject
+3	0	name=MemoryModelTranslator
+4	0	name=AgentChatSessionPersistence
+4	0	name=AgentSessionRecorder
+3	0	name=IncrementalOptimizer
+3	0	name=TypeSystemTranslator
+1	0	name=MigrationTestGenerator
+1	0	name=SCIPEmitter
+5	0	name=ArchitectReviewInterface
+1	0	name=CapabilityDeclarationGenerator
+0	0	name=Sprint190IntegrationSummary
+5	0	name=SecurityExceptionReviewBoard
+1	0	name=SecurityTestSkeletonGenerator
+1	0	name=SchemaToCppGenerator
+1	0	name=JobDispatchTableGenerator
+0	0	name=Sprint85IntegrationSummary
+2	0	name=SymbolSelectorAPI
+1	0	name=ProjectAstCMakeGenerator
+0	0	name=BatchMutationAPI
+2	0	name=CanaryPromotionJudge
+5	0	name=ArchitectReviewSurface
+3	0	name=TypeAwareMappings
+3	0	name=SecurityPreservingTranslation
+3	0	name=TransformEngineExtended
+5	0	name=ArchitectReviewSurface
+0	0	name=Sprint222IntegrationSummary
+5	0	name=ArchitectReviewInterface
+5	0	name=ArchitectReviewSurface
+4	0	name=PolyglotSuiteOrchestrator
+0	0	name=Sprint265IntegrationSummary
+1	0	name=SyntaxHighlighterOrg
+3	0	name=TransformEngineExtended
+3	0	name=TypeAwareMappings
+4	0	name=AgentChatPanelModel
+4	0	name=ArchitectModuleDecomposer
+0	0	name=ContextSliceAssembler
+1	0	name=SyntaxHighlighterJavaScript
+3	0	name=TranspilationConfidence
+4	0	name=ArchitectTechStackSelector
+2	0	name=PerTaskLibrarySelector
+5	0	name=ArchitectReviewInterface
+5	0	name=SecurityExceptionReviewBoard
+0	0	name=Sprint106IntegrationSummary
+2	0	name=OperationClassifier
+1	0	name=SQLiteDataLayerGenerator
+3	0	name=SelfHostTranspileAudit
+0	0	name=Sprint27IntegrationSummary
+3	0	name=IncrementalOptimizer
+5	0	name=AgentMutationApproval
+0	0	name=Sprint98IntegrationSummary
+2	0	name=OperationSelectorAPI
+3	0	name=SelfHostTranspileAudit
+2	0	name=OperationSelectorAPI
+0	0	name=AccessibilityBaseline
+5	0	name=AgentMutationApproval
+1	0	name=TraceGenerator
+0	0	name=CostEstimator
+4	0	name=AgentPermissionPolicy
+5	0	name=AgentMutationPreview
+3	0	name=ConcurrencyTranslator
+3	0	name=SelfHostTranspileAudit
+3	0	name=ContextWindowOptimizer
+5	0	name=AgentMutationPreview
+3	0	name=IntentTranslator
+4	0	name=AgentLibraryPolicy
+1	0	name=SyntaxHighlighterPython
+1	0	name=JobDispatchTableGenerator
+0	0	name=EditorModePolicy
+2	0	name=LibrarySymbolAdvisor
+4	0	name=AgentMarketplace
+3	0	name=TranspilationConfidence
+0	0	name=SwarmStatusPanelModel
+5	0	name=ApprovalEscalationPlanner
+3	0	name=TransformEngine
+1	0	name=SyntaxHighlighter
+4	0	name=ArchitectIntakeProcessor
+4	0	name=PolyglotOrchestrator
+4	0	name=AgentTaskStatusOverlay
+5	0	name=ArchitectReviewInterface
+2	0	name=OperationSelectorAPI
+2	0	name=LibrarySymbolAdvisor
+0	0	name=Sprint184IntegrationSummary
+5	0	name=SecurityExceptionReviewBoard
+0	0	name=Sprint36aIntegration
+5	0	name=ArchitectReviewInterface
+1	0	name=SyntaxHighlighterPython
+0	0	name=SearchUtils
+0	0	name=Diagnostics
+1	0	name=TestPlanGenerator
+5	0	name=SecurityExceptionReviewBoard
+1	0	name=InferenceJobGenerator
+4	0	name=ArchitectProblemParser
+0	0	name=LSPSymbolExtractor
+2	0	name=SymbolSelectorAPI
+4	0	name=ArchitectProjectPipeline
+2	0	name=PerTaskLibrarySelector
+4	0	name=AgentToolCallVisualization
+1	0	name=ProjectSkeletonGenerator
+2	0	name=TaskitemLibraryAnnotator
+2	0	name=CanaryPromotionJudge
+5	0	name=SecurityExceptionReviewBoard
+2	0	name=OperationSelectorAPI
+1	0	name=LanguageGeneratorPluginManifest
+3	0	name=TransformEngine
+4	0	name=OnCallCoveragePlanner
+2	0	name=TaskitemLibraryAnnotator
+2	0	name=TaskitemLibraryAnnotator
+5	0	name=AgentMutationPreview
+4	0	name=AgentChatSessionPersistence
+1	0	name=CHeaderEmitter
+0	0	name=Sprint49IntegrationSummary
+3	0	name=ContextWindowOptimizer
+2	0	name=PerTaskLibrarySelector
+4	0	name=ArchitectMultiLanguageOrchestrator
+0	0	name=MCPWorkflowPrompts
+3	0	name=TranspilationRPC
+3	0	name=ContextWindowOptimizer
+2	0	name=OperationSelectorAPI
+0	0	name=Sprint198IntegrationSummary
+4	0	name=ArchitectScaffoldGenerator
+1	0	name=GoCppBindingEmitter
+0	0	name=AnimationUtils
+0	0	name=SelfHostAnnotationAudit
+4	0	name=ArchitectIntakeProcessor
+4	0	name=AgentMarketplace
+3	0	name=TranspilationConfidence
+5	0	name=ArchitectReviewSurface
+1	0	name=SecureByDefaultGenerator
+2	0	name=OperationSelectorAPI
+2	0	name=CanaryPromotionJudge
+2	0	name=TaskitemLibraryAnnotator
+4	0	name=AgentChatPanelModel
+2	0	name=LibrarySymbolAdvisor
+3	0	name=SecurityPreservingTranslation
+0	0	name=APIBoundaryPreserver
+5	0	name=ApprovalEscalationPlanner
+3	0	name=TypeSystemTranslator
+3	0	name=StrategyAwareOptimizer
+3	0	name=TransformEngineExtended
+1	0	name=SyntaxHighlighterRust
+4	0	name=AgentSessionRecorder
+0	0	name=Sprint136IntegrationSummary
+0	0	name=OperationTaxonomy
+4	0	name=ArchitectTechStackSelector
+0	0	name=AnnotationConflict
+2	0	name=CanaryPromotionJudge
+5	0	name=ApprovalEscalationPlanner
+0	0	name=Sprint286IntegrationSummary
+1	0	name=MigrationPlanGenerator
+0	0	name=Sprint261IntegrationSummary
+1	0	name=TrainingDataGenerator
+1	0	name=RustPythonBindingEmitter
+4	0	name=AgentTaskSlots
+4	0	name=AgentCodeGen
+0	0	name=Sprint74IntegrationSummary
+1	0	name=ArchitectTemplates
+2	0	name=SymbolSelectorAPI
+2	0	name=TaskitemLibraryAnnotator
+5	0	name=ApprovalEscalationPlanner
+2	0	name=CanaryPromotionJudge
+5	0	name=AgentMutationApproval
+1	0	name=SyntaxHighlighterJava
+5	0	name=AgentMutationPreview
+5	0	name=AgentMutationPreview
+1	0	name=SQLiteDataLayerGenerator
+4	0	name=AgentTaskSlots
+4	0	name=AgentRPCHandler
+3	0	name=MemoryModelTranslator
+3	0	name=ConcurrencyTranslator
+5	0	name=ArchitectReviewSurface
+4	0	name=AgentToolCallVisualization
+5	0	name=SecurityExceptionReviewBoard
+1	0	name=RustGoBindingEmitter
+0	0	name=Sprint58IntegrationSummary
+3	0	name=IncrementalOptimizer
+0	0	name=Sprint239IntegrationSummary
+1	0	name=SyntaxHighlighterElisp
+1	0	name=PlatformCMakeTargetGenerator
+5	0	name=ApprovalEscalationPlanner
+1	0	name=MqttBoilerplateGenerator
+2	0	name=PerTaskLibrarySelector
+3	0	name=IntentTranslator
+5	0	name=SecurityExceptionReviewBoard
+1	0	name=KeySymbolRenderer
+2	0	name=OperationSelectorAPI
+3	0	name=MemoryModelTranslator
+5	0	name=ArchitectReviewInterface
+1	0	name=ArchitectSkeletonGenerator
+2	0	name=OperationClassifier
+5	0	name=ApprovalEscalationPlanner
+2	0	name=OperationClassifier
+1	0	name=PackageScriptGenerator
+2	0	name=TaskitemLibraryAnnotator
+3	0	name=TypeAwareMappings
+5	0	name=ArchitectReviewInterface
+1	0	name=GoCppBindingEmitter
+5	0	name=AgentMutationApproval
+1	0	name=CHeaderEmitter
+1	0	name=SyntaxHighlighterOrg
+1	0	name=TaskitemGeneratorV2
+0	0	name=Sprint96IntegrationSummary
+0	0	name=ScoreClampUtil
+4	0	name=AgentRPCHandler
+4	0	name=ArchitectScaffoldGenerator
+4	0	name=ScopeMilestoneDecomposer
+5	0	name=AgentMutationApproval
+5	0	name=AgentMutationPreview
+4	0	name=AgentPermissionPolicy
+2	0	name=PerTaskLibrarySelector
+0	0	name=ConstrainedRoutingRulesetExtension
+1	0	name=LanguageGeneratorPluginManifest
+5	0	name=ArchitectReviewSurface
+0	0	name=WorkItem
+2	0	name=TaskitemLibraryAnnotator
+2	0	name=CanaryPromotionJudge
+5	0	name=AgentMutationApproval
+2	0	name=LibrarySymbolAdvisor
+3	0	name=ConcurrencyTranslator
+2	0	name=CanaryPromotionJudge
+4	0	name=WorkflowOrchestrator
+1	0	name=InferenceJobGenerator
+3	0	name=TranspilationRPC
+4	0	name=AgentLibraryPolicy
+0	0	name=Sprint272IntegrationSummary
+3	0	name=TransformEngine
+3	0	name=TranspilationRPC
+4	0	name=OnCallCoveragePlanner
+2	0	name=SymbolSelectorAPI
+4	0	name=PolyglotOrchestrator
+2	0	name=OperationClassifier
+0	0	name=RefactorActions
+2	0	name=LibrarySymbolAdvisor
+3	0	name=IntentTranslator
+2	0	name=PerTaskLibrarySelector
+4	0	name=WorkflowOrchestrator
+4	0	name=AgentChatContextInjector
+4	0	name=AgentAnnotationAssistant
+5	0	name=AgentMutationApproval
+2	0	name=SymbolSelectorAPI
+5	0	name=ArchitectReviewInterface
+1	0	name=SyntaxHighlighterCpp
+4	0	name=ArchitectMultiLanguageOrchestrator
+1	0	name=APIDocGenerator
+4	0	name=ArchitectProblemParser
+3	0	name=StrategyAwareOptimizer
+4	0	name=AgentAnnotationAssistant
+1	0	name=PackageScriptGenerator
+4	0	name=AgentRegistry
+4	0	name=ArchitectModuleDecomposer
+0	0	name=Sprint249IntegrationSummary
+3	0	name=SecurityPreservingTranslation
+4	0	name=ArchitectProjectPipeline
+0	0	name=WorkerDispatchValidator
+3	0	name=ContextWindowOptimizer
+4	0	name=PolyglotSuiteOrchestrator
+5	0	name=AgentMutationPreview
+3	0	name=TypeSystemTranslator
diff --git a/specialists/data/generated/confidence_tier_eval.tsv b/specialists/data/generated/confidence_tier_eval.tsv
new file mode 100644
index 0000000..9ef70ab
--- /dev/null
+++ b/specialists/data/generated/confidence_tier_eval.tsv
@@ -0,0 +1,180 @@
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=8 ambiguity=1 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=5 ambiguity=0 deps=0 prereqs=4 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=2 ambiguity=2 deps=0 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=2 ambiguity=2 deps=2 prereqs=1 queueready=no
+2	0	conflicts=6 ambiguity=1 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=4 queueready=yes
+2	0	conflicts=4 ambiguity=0 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=7 ambiguity=2 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=2 ambiguity=1 deps=0 prereqs=3 queueready=no
+2	0	conflicts=4 ambiguity=0 deps=1 prereqs=3 queueready=no
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=7 ambiguity=2 deps=2 prereqs=3 queueready=no
+1	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=4 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=4 ambiguity=2 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=3 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=7 ambiguity=1 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=5 ambiguity=2 deps=2 prereqs=4 queueready=yes
+2	0	conflicts=0 ambiguity=3 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=2 ambiguity=2 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=2 ambiguity=0 deps=0 prereqs=3 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=7 ambiguity=0 deps=0 prereqs=1 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=1 ambiguity=3 deps=0 prereqs=3 queueready=no
+2	0	conflicts=2 ambiguity=1 deps=2 prereqs=3 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=5 ambiguity=3 deps=2 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=1 prereqs=3 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=4 queueready=yes
+2	0	conflicts=7 ambiguity=3 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=4 ambiguity=3 deps=0 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=8 ambiguity=3 deps=2 prereqs=4 queueready=no
+2	0	conflicts=4 ambiguity=2 deps=2 prereqs=1 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=8 ambiguity=2 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=8 ambiguity=1 deps=0 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=5 ambiguity=2 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=1 ambiguity=1 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=3 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=no
+2	0	conflicts=1 ambiguity=2 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=2 ambiguity=2 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=3 ambiguity=1 deps=0 prereqs=3 queueready=no
+2	0	conflicts=0 ambiguity=2 deps=1 prereqs=2 queueready=no
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=7 ambiguity=1 deps=1 prereqs=2 queueready=no
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=8 ambiguity=2 deps=1 prereqs=1 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=7 ambiguity=1 deps=0 prereqs=1 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=6 ambiguity=3 deps=1 prereqs=2 queueready=no
+2	0	conflicts=9 ambiguity=3 deps=2 prereqs=3 queueready=no
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=0 ambiguity=0 deps=1 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=4 ambiguity=3 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=3 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=2 ambiguity=3 deps=1 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=0 ambiguity=1 deps=1 prereqs=4 queueready=no
+2	0	conflicts=8 ambiguity=2 deps=2 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=1 ambiguity=3 deps=0 prereqs=4 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=3 ambiguity=2 deps=1 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=no
+2	0	conflicts=8 ambiguity=2 deps=1 prereqs=4 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=7 ambiguity=3 deps=2 prereqs=2 queueready=yes
diff --git a/specialists/data/generated/confidence_tier_train.tsv b/specialists/data/generated/confidence_tier_train.tsv
new file mode 100644
index 0000000..05dea7e
--- /dev/null
+++ b/specialists/data/generated/confidence_tier_train.tsv
@@ -0,0 +1,1020 @@
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=1 ambiguity=0 deps=1 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=8 ambiguity=0 deps=0 prereqs=2 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=1 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=3 queueready=yes
+2	0	conflicts=4 ambiguity=3 deps=2 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=0 ambiguity=3 deps=1 prereqs=1 queueready=no
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=4 queueready=yes
+2	0	conflicts=8 ambiguity=3 deps=2 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=1 ambiguity=1 deps=1 prereqs=1 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=4 ambiguity=2 deps=2 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=5 ambiguity=1 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=6 ambiguity=1 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=0 ambiguity=1 deps=0 prereqs=4 queueready=no
+2	0	conflicts=4 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=9 ambiguity=1 deps=2 prereqs=4 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=3 ambiguity=1 deps=1 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=4 queueready=yes
+2	0	conflicts=9 ambiguity=3 deps=1 prereqs=1 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=1 ambiguity=1 deps=2 prereqs=1 queueready=no
+2	0	conflicts=0 ambiguity=1 deps=1 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=5 ambiguity=1 deps=1 prereqs=4 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=9 ambiguity=0 deps=0 prereqs=2 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=6 ambiguity=2 deps=2 prereqs=1 queueready=no
+2	0	conflicts=3 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=2 ambiguity=0 deps=2 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=9 ambiguity=0 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=5 ambiguity=2 deps=1 prereqs=4 queueready=yes
+2	0	conflicts=9 ambiguity=2 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=8 ambiguity=1 deps=2 prereqs=4 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=3 ambiguity=3 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=7 ambiguity=3 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=4 ambiguity=0 deps=1 prereqs=2 queueready=no
+2	0	conflicts=6 ambiguity=1 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=5 ambiguity=3 deps=0 prereqs=4 queueready=no
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=9 ambiguity=0 deps=2 prereqs=4 queueready=no
+2	0	conflicts=6 ambiguity=0 deps=1 prereqs=4 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=2 ambiguity=2 deps=1 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=4 ambiguity=1 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=9 ambiguity=0 deps=2 prereqs=4 queueready=yes
+2	0	conflicts=2 ambiguity=0 deps=1 prereqs=3 queueready=no
+2	0	conflicts=5 ambiguity=3 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=9 ambiguity=2 deps=1 prereqs=2 queueready=no
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=2 ambiguity=3 deps=2 prereqs=4 queueready=no
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=0 ambiguity=3 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=5 ambiguity=3 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=3 ambiguity=3 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=7 ambiguity=3 deps=1 prereqs=4 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=6 ambiguity=3 deps=0 prereqs=1 queueready=no
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=7 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=no
+2	0	conflicts=6 ambiguity=3 deps=0 prereqs=4 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=6 ambiguity=2 deps=2 prereqs=3 queueready=no
+2	0	conflicts=5 ambiguity=3 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=8 ambiguity=2 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=3 ambiguity=2 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=7 ambiguity=0 deps=2 prereqs=3 queueready=no
+2	0	conflicts=1 ambiguity=2 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=4 queueready=yes
+2	0	conflicts=7 ambiguity=1 deps=1 prereqs=1 queueready=no
+2	0	conflicts=8 ambiguity=3 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=7 ambiguity=1 deps=0 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=5 ambiguity=2 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=4 queueready=yes
+2	0	conflicts=0 ambiguity=3 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=9 ambiguity=1 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=7 ambiguity=2 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=5 ambiguity=3 deps=1 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=4 queueready=yes
+2	0	conflicts=6 ambiguity=2 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=2 ambiguity=0 deps=1 prereqs=4 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=5 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=4 ambiguity=2 deps=0 prereqs=3 queueready=no
+2	0	conflicts=9 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=8 ambiguity=2 deps=0 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=no
+1	0	conflicts=1 ambiguity=1 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=9 ambiguity=1 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=no
+1	0	conflicts=1 ambiguity=1 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=8 ambiguity=2 deps=2 prereqs=4 queueready=no
+2	0	conflicts=5 ambiguity=3 deps=0 prereqs=1 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=0 ambiguity=3 deps=1 prereqs=4 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=4 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=4 ambiguity=1 deps=1 prereqs=4 queueready=no
+2	0	conflicts=9 ambiguity=0 deps=2 prereqs=2 queueready=no
+2	0	conflicts=1 ambiguity=3 deps=1 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=7 ambiguity=1 deps=2 prereqs=3 queueready=no
+2	0	conflicts=0 ambiguity=3 deps=0 prereqs=1 queueready=no
+2	0	conflicts=6 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=3 ambiguity=0 deps=2 prereqs=1 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=1 ambiguity=2 deps=2 prereqs=1 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=1 ambiguity=0 deps=2 prereqs=2 queueready=no
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=6 ambiguity=3 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=1 ambiguity=0 deps=1 prereqs=1 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=1 ambiguity=2 deps=1 prereqs=2 queueready=no
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=5 ambiguity=3 deps=1 prereqs=2 queueready=no
+2	0	conflicts=2 ambiguity=2 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=6 ambiguity=3 deps=1 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=no
+2	0	conflicts=3 ambiguity=0 deps=2 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=2 ambiguity=3 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=no
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=4 ambiguity=3 deps=0 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=3 ambiguity=3 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=8 ambiguity=1 deps=0 prereqs=2 queueready=no
+2	0	conflicts=4 ambiguity=1 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=2 ambiguity=0 deps=0 prereqs=2 queueready=no
+2	0	conflicts=7 ambiguity=0 deps=0 prereqs=4 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=8 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=3 queueready=yes
+2	0	conflicts=5 ambiguity=3 deps=1 prereqs=1 queueready=no
+1	0	conflicts=0 ambiguity=0 deps=1 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=9 ambiguity=2 deps=2 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=5 ambiguity=0 deps=0 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=3 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=5 ambiguity=1 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=5 ambiguity=1 deps=2 prereqs=4 queueready=yes
+2	0	conflicts=2 ambiguity=2 deps=1 prereqs=2 queueready=no
+1	0	conflicts=3 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=9 ambiguity=3 deps=0 prereqs=4 queueready=no
+1	0	conflicts=0 ambiguity=2 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=0 ambiguity=3 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=9 ambiguity=1 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=7 ambiguity=2 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=5 ambiguity=0 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=4 ambiguity=1 deps=1 prereqs=1 queueready=no
+2	0	conflicts=9 ambiguity=3 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=3 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=7 ambiguity=0 deps=1 prereqs=4 queueready=no
+2	0	conflicts=2 ambiguity=3 deps=1 prereqs=3 queueready=no
+2	0	conflicts=9 ambiguity=1 deps=1 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=1 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=5 ambiguity=0 deps=1 prereqs=2 queueready=no
+2	0	conflicts=5 ambiguity=3 deps=2 prereqs=4 queueready=no
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=0 ambiguity=3 deps=1 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=8 ambiguity=2 deps=0 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=9 ambiguity=1 deps=0 prereqs=1 queueready=no
+2	0	conflicts=5 ambiguity=3 deps=0 prereqs=2 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=0 ambiguity=2 deps=2 prereqs=4 queueready=no
+2	0	conflicts=2 ambiguity=2 deps=1 prereqs=4 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=5 ambiguity=0 deps=1 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=1 ambiguity=1 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=3 queueready=yes
+2	0	conflicts=0 ambiguity=3 deps=0 prereqs=3 queueready=no
+2	0	conflicts=3 ambiguity=0 deps=1 prereqs=2 queueready=no
+2	0	conflicts=6 ambiguity=3 deps=2 prereqs=3 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=7 ambiguity=3 deps=0 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=9 ambiguity=2 deps=0 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=8 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=6 ambiguity=0 deps=2 prereqs=4 queueready=no
+2	0	conflicts=5 ambiguity=1 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=3 ambiguity=0 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=8 ambiguity=0 deps=2 prereqs=4 queueready=no
+2	0	conflicts=3 ambiguity=1 deps=2 prereqs=3 queueready=no
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=4 queueready=yes
+2	0	conflicts=3 ambiguity=2 deps=2 prereqs=3 queueready=no
+2	0	conflicts=8 ambiguity=0 deps=0 prereqs=3 queueready=no
+2	0	conflicts=6 ambiguity=2 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=7 ambiguity=0 deps=1 prereqs=3 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=4 ambiguity=1 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=3 ambiguity=0 deps=0 prereqs=4 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=4 ambiguity=2 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=0 ambiguity=2 deps=2 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=5 ambiguity=2 deps=0 prereqs=1 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=3 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=4 ambiguity=3 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=3 queueready=yes
+2	0	conflicts=3 ambiguity=1 deps=1 prereqs=4 queueready=yes
+2	0	conflicts=4 ambiguity=3 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=4 ambiguity=0 deps=2 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=9 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=3 ambiguity=3 deps=0 prereqs=1 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=2 prereqs=4 queueready=yes
+2	0	conflicts=0 ambiguity=2 deps=2 prereqs=3 queueready=yes
+2	0	conflicts=4 ambiguity=2 deps=1 prereqs=2 queueready=no
+2	0	conflicts=6 ambiguity=0 deps=0 prereqs=4 queueready=no
+2	0	conflicts=9 ambiguity=1 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=1 prereqs=4 queueready=yes
+2	0	conflicts=7 ambiguity=3 deps=0 prereqs=3 queueready=no
+2	0	conflicts=6 ambiguity=1 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=0 ambiguity=3 deps=2 prereqs=4 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=5 ambiguity=3 deps=2 prereqs=1 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=8 ambiguity=0 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=8 ambiguity=3 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=9 ambiguity=2 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=4 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=8 ambiguity=0 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=4 ambiguity=2 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=3 ambiguity=0 deps=1 prereqs=3 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=3 ambiguity=2 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=6 ambiguity=3 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=7 ambiguity=1 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=0 ambiguity=3 deps=0 prereqs=2 queueready=no
+2	0	conflicts=3 ambiguity=2 deps=1 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=9 ambiguity=1 deps=1 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=5 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=3 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=6 ambiguity=3 deps=2 prereqs=4 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=7 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=2 ambiguity=0 deps=2 prereqs=4 queueready=no
+2	0	conflicts=1 ambiguity=1 deps=1 prereqs=2 queueready=no
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=3 ambiguity=1 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=0 ambiguity=2 deps=0 prereqs=3 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=1 ambiguity=3 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=7 ambiguity=1 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=9 ambiguity=2 deps=1 prereqs=1 queueready=no
+2	0	conflicts=4 ambiguity=0 deps=1 prereqs=1 queueready=no
+2	0	conflicts=4 ambiguity=0 deps=0 prereqs=1 queueready=no
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=4 ambiguity=2 deps=2 prereqs=4 queueready=no
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=4 queueready=yes
+2	0	conflicts=8 ambiguity=1 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=2 ambiguity=3 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=3 ambiguity=3 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=8 ambiguity=1 deps=2 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=3 ambiguity=1 deps=2 prereqs=2 queueready=no
+2	0	conflicts=6 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=6 ambiguity=2 deps=0 prereqs=3 queueready=no
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=3 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=6 ambiguity=1 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=5 ambiguity=0 deps=2 prereqs=4 queueready=no
+2	0	conflicts=6 ambiguity=1 deps=0 prereqs=1 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=8 ambiguity=0 deps=2 prereqs=3 queueready=no
+1	0	conflicts=0 ambiguity=0 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=6 ambiguity=2 deps=0 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=4 ambiguity=3 deps=1 prereqs=1 queueready=no
+2	0	conflicts=4 ambiguity=0 deps=2 prereqs=1 queueready=no
+2	0	conflicts=7 ambiguity=2 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=8 ambiguity=1 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=7 ambiguity=2 deps=1 prereqs=4 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=3 ambiguity=1 deps=0 prereqs=4 queueready=no
+2	0	conflicts=3 ambiguity=3 deps=2 prereqs=3 queueready=yes
+2	0	conflicts=7 ambiguity=0 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=4 ambiguity=3 deps=1 prereqs=3 queueready=no
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=7 ambiguity=1 deps=1 prereqs=4 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=9 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=2 ambiguity=1 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=1 ambiguity=3 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=no
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=2 ambiguity=3 deps=1 prereqs=3 queueready=yes
+1	0	conflicts=3 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=2 prereqs=3 queueready=yes
+2	0	conflicts=0 ambiguity=2 deps=1 prereqs=4 queueready=no
+2	0	conflicts=9 ambiguity=3 deps=0 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=9 ambiguity=3 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=9 ambiguity=2 deps=2 prereqs=1 queueready=no
+1	0	conflicts=1 ambiguity=1 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=3 ambiguity=3 deps=1 prereqs=1 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=2 prereqs=4 queueready=yes
+2	0	conflicts=0 ambiguity=2 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=6 ambiguity=3 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=0 ambiguity=1 deps=2 prereqs=3 queueready=no
+1	0	conflicts=1 ambiguity=1 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=9 ambiguity=1 deps=0 prereqs=2 queueready=no
+2	0	conflicts=5 ambiguity=2 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=3 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=8 ambiguity=2 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=5 ambiguity=3 deps=1 prereqs=4 queueready=no
+2	0	conflicts=6 ambiguity=2 deps=1 prereqs=1 queueready=no
+1	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=no
+2	0	conflicts=6 ambiguity=2 deps=1 prereqs=3 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=no
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=8 ambiguity=1 deps=0 prereqs=1 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=1 ambiguity=2 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=3 ambiguity=3 deps=2 prereqs=2 queueready=no
+2	0	conflicts=4 ambiguity=1 deps=2 prereqs=3 queueready=no
+2	0	conflicts=1 ambiguity=1 deps=2 prereqs=2 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=5 ambiguity=1 deps=2 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=0 ambiguity=1 deps=0 prereqs=1 queueready=no
+2	0	conflicts=6 ambiguity=0 deps=0 prereqs=1 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=4 queueready=yes
+2	0	conflicts=4 ambiguity=3 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=9 ambiguity=0 deps=2 prereqs=1 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=4 ambiguity=3 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=0 ambiguity=2 deps=1 prereqs=1 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=6 ambiguity=2 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=0 ambiguity=2 deps=1 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=8 ambiguity=1 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=2 ambiguity=0 deps=0 prereqs=1 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=0 ambiguity=1 deps=2 prereqs=1 queueready=no
+1	0	conflicts=1 ambiguity=1 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=4 ambiguity=3 deps=2 prereqs=4 queueready=no
+2	0	conflicts=4 ambiguity=0 deps=1 prereqs=4 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=2 prereqs=4 queueready=yes
+2	0	conflicts=3 ambiguity=2 deps=1 prereqs=1 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=4 ambiguity=2 deps=0 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=5 ambiguity=2 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=4 ambiguity=1 deps=0 prereqs=3 queueready=no
+2	0	conflicts=3 ambiguity=3 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=3 ambiguity=1 deps=0 prereqs=2 queueready=no
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=8 ambiguity=3 deps=1 prereqs=1 queueready=no
+2	0	conflicts=5 ambiguity=1 deps=2 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=3 ambiguity=0 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=7 ambiguity=0 deps=0 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=4 ambiguity=2 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=3 queueready=yes
+2	0	conflicts=9 ambiguity=2 deps=1 prereqs=3 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=9 ambiguity=3 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=9 ambiguity=0 deps=2 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=0 ambiguity=3 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=1 ambiguity=3 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=7 ambiguity=3 deps=2 prereqs=2 queueready=no
+1	0	conflicts=0 ambiguity=2 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=4 ambiguity=1 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=2 ambiguity=3 deps=0 prereqs=1 queueready=no
+2	0	conflicts=8 ambiguity=3 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=9 ambiguity=0 deps=1 prereqs=1 queueready=no
+2	0	conflicts=3 ambiguity=2 deps=1 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=1 ambiguity=1 deps=0 prereqs=3 queueready=no
+2	0	conflicts=7 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=7 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=3 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=5 ambiguity=2 deps=2 prereqs=3 queueready=no
+2	0	conflicts=6 ambiguity=0 deps=1 prereqs=4 queueready=yes
+2	0	conflicts=8 ambiguity=3 deps=2 prereqs=1 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=2 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=8 ambiguity=1 deps=1 prereqs=1 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=4 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=2 ambiguity=1 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=9 ambiguity=3 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=3 ambiguity=2 deps=2 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=2 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=4 ambiguity=3 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=2 ambiguity=2 deps=2 prereqs=2 queueready=no
+2	0	conflicts=5 ambiguity=2 deps=0 prereqs=4 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=6 ambiguity=0 deps=0 prereqs=3 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=5 ambiguity=0 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=4 ambiguity=2 deps=0 prereqs=1 queueready=no
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=4 queueready=yes
+2	0	conflicts=4 ambiguity=2 deps=2 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=4 ambiguity=1 deps=1 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=0 ambiguity=2 deps=0 prereqs=2 queueready=no
+2	0	conflicts=8 ambiguity=2 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=6 ambiguity=3 deps=2 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=9 ambiguity=1 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=2 ambiguity=2 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=2 ambiguity=1 deps=0 prereqs=4 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=8 ambiguity=3 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=2 ambiguity=1 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=8 ambiguity=3 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=2 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=3 ambiguity=0 deps=1 prereqs=1 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=8 ambiguity=1 deps=2 prereqs=2 queueready=no
+2	0	conflicts=2 ambiguity=1 deps=1 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=3 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=5 ambiguity=3 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=4 queueready=yes
+2	0	conflicts=2 ambiguity=1 deps=2 prereqs=4 queueready=no
+2	0	conflicts=8 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=5 ambiguity=3 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=3 queueready=yes
+2	0	conflicts=3 ambiguity=3 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=2 ambiguity=1 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=2 ambiguity=3 deps=1 prereqs=4 queueready=no
+1	0	conflicts=3 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=9 ambiguity=3 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=1 ambiguity=0 deps=0 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=0 ambiguity=3 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=1 ambiguity=3 deps=1 prereqs=4 queueready=no
+2	0	conflicts=3 ambiguity=1 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=9 ambiguity=2 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=3 ambiguity=2 deps=2 prereqs=4 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=1 ambiguity=3 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=8 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=5 ambiguity=1 deps=1 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=7 ambiguity=2 deps=1 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=3 ambiguity=0 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=2 ambiguity=0 deps=2 prereqs=2 queueready=no
+2	0	conflicts=5 ambiguity=2 deps=2 prereqs=4 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=0 ambiguity=0 deps=1 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=9 ambiguity=3 deps=1 prereqs=3 queueready=no
+2	0	conflicts=0 ambiguity=0 deps=1 prereqs=4 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=4 ambiguity=2 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=6 ambiguity=2 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=8 ambiguity=3 deps=0 prereqs=4 queueready=yes
+2	0	conflicts=8 ambiguity=0 deps=1 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=5 ambiguity=1 deps=0 prereqs=4 queueready=no
+2	0	conflicts=5 ambiguity=2 deps=1 prereqs=2 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=2 queueready=yes
+1	0	conflicts=3 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=7 ambiguity=2 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=3 ambiguity=3 deps=2 prereqs=3 queueready=no
+2	0	conflicts=4 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=5 ambiguity=0 deps=1 prereqs=4 queueready=no
+2	0	conflicts=9 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=6 ambiguity=0 deps=2 prereqs=1 queueready=no
+1	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=2 ambiguity=3 deps=1 prereqs=1 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=9 ambiguity=1 deps=0 prereqs=4 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=3 ambiguity=3 deps=0 prereqs=4 queueready=no
+2	0	conflicts=7 ambiguity=0 deps=1 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=0 ambiguity=2 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=2 ambiguity=3 deps=0 prereqs=2 queueready=no
+2	0	conflicts=3 ambiguity=0 deps=2 prereqs=3 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=2 queueready=yes
+2	0	conflicts=2 ambiguity=3 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=9 ambiguity=3 deps=0 prereqs=1 queueready=no
+2	0	conflicts=5 ambiguity=1 deps=2 prereqs=3 queueready=yes
+2	0	conflicts=9 ambiguity=0 deps=2 prereqs=1 queueready=yes
+2	0	conflicts=8 ambiguity=0 deps=1 prereqs=3 queueready=yes
+1	0	conflicts=1 ambiguity=0 deps=0 prereqs=3 queueready=yes
+2	0	conflicts=4 ambiguity=1 deps=1 prereqs=3 queueready=no
+1	0	conflicts=0 ambiguity=1 deps=1 prereqs=3 queueready=yes
+2	0	conflicts=1 ambiguity=2 deps=0 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=1 queueready=yes
+2	0	conflicts=6 ambiguity=3 deps=1 prereqs=1 queueready=no
+1	0	conflicts=1 ambiguity=0 deps=1 prereqs=4 queueready=yes
+2	0	conflicts=1 ambiguity=3 deps=0 prereqs=1 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=1 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=6 ambiguity=1 deps=0 prereqs=4 queueready=no
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=2 queueready=yes
+2	0	conflicts=5 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=1 prereqs=1 queueready=yes
+1	0	conflicts=1 ambiguity=1 deps=2 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=1 queueready=yes
+2	0	conflicts=3 ambiguity=2 deps=1 prereqs=3 queueready=no
+1	0	conflicts=2 ambiguity=0 deps=0 prereqs=2 queueready=yes
+1	0	conflicts=0 ambiguity=1 deps=0 prereqs=4 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=0 prereqs=1 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=2 prereqs=1 queueready=yes
+1	0	conflicts=0 ambiguity=2 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=4 ambiguity=1 deps=2 prereqs=3 queueready=yes
+2	0	conflicts=2 ambiguity=2 deps=1 prereqs=2 queueready=yes
+2	0	conflicts=8 ambiguity=3 deps=2 prereqs=3 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=1 prereqs=2 queueready=yes
+0	0	conflicts=0 ambiguity=0 deps=0 prereqs=4 queueready=yes
diff --git a/specialists/data/generated/prereq_op_eval.tsv b/specialists/data/generated/prereq_op_eval.tsv
new file mode 100644
index 0000000..b4448c6
--- /dev/null
+++ b/specialists/data/generated/prereq_op_eval.tsv
@@ -0,0 +1,24 @@
+2	0	`whetstone_enqueue_pair_upgrade` MCP tool
+3	0	Deprecation/succession lifecycle framework
+3	0	Security scan orchestrator for generated targets
+1	0	Phase 12c Integration Tests
+2	0	Reviewer quorum policy model for critical approvals
+1	0	Phase 38a Integration
+1	0	Phase 23b Integration + Sprint Summary
+0	0	Reprojection safety policy
+2	0	Stable corpus certification suite
+0	0	Timing regression runner integration
+0	0	`whetstone_get_tool_permissions` MCP tool
+3	0	Cross-language rollout state snapshot model
+1	0	Compatibility policy profiles by interface criticality
+3	0	`whetstone_publish_roadmap_epoch` MCP tool
+2	0	Security Report Generation
+2	0	Compatibility/deprecation report artifact
+1	0	Human Review Interface via MCP
+2	0	HiveMind job publisher from editor
+1	0	Parser/AST adapter interface schema
+0	0	Sprint 76 integration summary + regression
+1	0	Scenario — Multi-Model Orchestration
+0	0	Autonomous loop controller with “green or explicit blocked” contract
+2	0	Security-Aware Routing
+0	0	Verifier trust weighting policy engine
diff --git a/specialists/data/generated/prereq_op_projects_eval.tsv b/specialists/data/generated/prereq_op_projects_eval.tsv
new file mode 100644
index 0000000..e3952e8
--- /dev/null
+++ b/specialists/data/generated/prereq_op_projects_eval.tsv
@@ -0,0 +1,12 @@
+0	0	Define verification and readiness surface | Build distributed rate limiter with fixed/sliding window modes and policy APIs.
+1	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+1	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+1	0	Define verification and readiness surface | Refactor billing pipeline to introduce idempotency keys and rollback checkpoints across ledger service, queue worker, re
+2	0	Plan Replace search ranking pipeline behind feature flag with | Replace search ranking pipeline behind feature flag with staged rollout, telemetry checks, and automatic abort trigger.
+0	0	Define verification and readiness surface | Build a JWT auth service with register, login, refresh token rotation, and password hashing.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
diff --git a/specialists/data/generated/prereq_op_projects_train.tsv b/specialists/data/generated/prereq_op_projects_train.tsv
new file mode 100644
index 0000000..eb717b2
--- /dev/null
+++ b/specialists/data/generated/prereq_op_projects_train.tsv
@@ -0,0 +1,68 @@
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Define verification and readiness surface | Build chat bot backend with slash commands, permission checks, and scheduled digests.
+1	0	Define verification and readiness surface | Build an order saga orchestrator with compensating actions, timeout escalation, exactly-once message handling, and deter
+2	0	Define verification and readiness surface | Build a CMS backend with posts, tags, draft/publish workflow, and role-based access control.
+0	0	Define verification and readiness surface | Build a multi-tenant limiter with token bucket + sliding window hybrid policy, burst fairness guarantees, and lock-free 
+1	0	Plan Build a CI orchestrator service triggering jobs tracking | Build a CI orchestrator service triggering jobs, tracking states, and emitting summaries.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+2	0	Define verification and readiness surface | Build a CLI password vault with encrypted local storage and master-key unlock.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Define verification and readiness surface | Build a release notes generator from commits, PR labels, and issue references.
+2	0	Define verification and readiness surface | Build a secrets rotation scheduler with lease tracking, retry plans, and notifications.
+1	0	Define verification and readiness surface | Build an order saga orchestrator with compensating actions, timeout escalation, exactly-once message handling, and deter
+2	0	Define verification and readiness surface | Build a model registry API supporting stages, approvals, rollback, and signed artifact links.
+1	0	Define verification and readiness surface | Refactor billing pipeline to introduce idempotency keys and rollback checkpoints across ledger service, queue worker, re
+1	0	Define verification and readiness surface | Refactor billing pipeline to introduce idempotency keys and rollback checkpoints across ledger service, queue worker, re
+1	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+1	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Define verification and readiness surface | Build inventory endpoints with stock reservation, release, and low-stock alerts.
+0	0	Define verification and readiness surface | Build form builder backend with schema validation, submissions, and analytics counters.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+2	0	Define verification and readiness surface | Build fraud rules engine with rule evaluation graph, score thresholds, and audit trails.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+2	0	Define verification and readiness surface | Build a PII redaction pipeline masking emails, phones, and IDs with audit logging.
+1	0	Plan Build a CI orchestrator service triggering jobs tracking | Build a CI orchestrator service triggering jobs, tracking states, and emitting summaries.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+1	0	Plan Build a CI orchestrator service triggering jobs tracking | Build a CI orchestrator service triggering jobs, tracking states, and emitting summaries.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+1	0	Define verification and readiness surface | Build an order saga orchestrator with compensating actions, timeout escalation, exactly-once message handling, and deter
+0	0	Define verification and readiness surface | Build experiment tracking backend storing params, metrics, artifacts, and compare queries.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+0	0	Define verification and readiness surface | Build a .env manager CLI with profile switch, validation, and secure redaction in output.
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Define verification and readiness surface | Build a secret rotation controller with staged rollout, canary verification, rollback triggers, and explicit blast-radiu
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+1	0	Plan Build a CI orchestrator service triggering jobs tracking | Build a CI orchestrator service triggering jobs, tracking states, and emitting summaries.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build tax report backend aggregating gains/losses and generating filing-year summaries.
+1	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build a realtime log viewer backend with tail sessions, regex filters, and backpressure.
+1	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+0	0	Define verification and readiness surface | Build job queue service with delayed jobs, retries, visibility timeout, and metrics.
+1	0	Define verification and readiness surface | Refactor billing pipeline to introduce idempotency keys and rollback checkpoints across ledger service, queue worker, re
+1	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+1	0	Define verification and readiness surface | Build an order saga orchestrator with compensating actions, timeout escalation, exactly-once message handling, and deter
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Define verification and readiness surface | Build audit log service with immutable events, filtering, export, and retention policies.
+2	0	Define verification and readiness surface | Build LLM guardrail service for policy checks, PII filtering, and response classification.
+0	0	Define verification and readiness surface | Build telemetry collector with trace/span ingestion, sampling, and export adapters.
+3	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+0	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Define verification and readiness surface | Build a secret rotation controller with staged rollout, canary verification, rollback triggers, and explicit blast-radiu
+2	0	Plan Upgrade API contract from v1 to v2 backend | Upgrade API contract from v1 to v2: backend handlers, OpenAPI schema, generated client SDK, frontend integration, and co
+2	0	Define verification and readiness surface | Build configuration service with versioned configs, staged rollout, and validation checks.
+2	0	Define verification and readiness surface | Build access control service with RBAC roles, permissions, and resource-scoped grants.
+2	0	Define verification and readiness surface | Build a feature flag API with environments, targeting rules, and rollout percentages.
+3	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+0	0	Define verification and readiness surface | Build an event deduplicator using idempotency keys, time windows, and replay-safe writes.
+1	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
diff --git a/specialists/data/generated/prereq_op_train.tsv b/specialists/data/generated/prereq_op_train.tsv
new file mode 100644
index 0000000..56e0813
--- /dev/null
+++ b/specialists/data/generated/prereq_op_train.tsv
@@ -0,0 +1,136 @@
+2	0	`whetstone_get_constructive_rollout_status` runtime implementation
+1	0	Target projection blocker taxonomy
+2	0	Strict gate mode in production loop
+3	0	`whetstone_get_polyglot_cutover_readiness` MCP tool
+3	0	Security scan orchestrator for generated targets
+1	0	Multi-language constructive loop report artifact
+2	0	`whetstone_get_constructive_rollout_status` MCP tool
+3	0	`whetstone_get_polyglot_cutover_readiness` MCP tool
+2	0	Deprecation policy model for unstable/obsolete pairs
+0	0	Projection schema extension for run catalogs
+0	0	Panel Registration System
+0	0	Sprint 146 integration summary + regression
+3	0	`whetstone_publish_roadmap_epoch` MCP tool
+1	0	Distributed path in FeedbackLoopOrchestrator
+1	0	Continuity drill framework and scoring
+3	0	Cross-language release evidence aggregator model
+0	0	Scheme lowering/raising adapters
+3	0	Cross-language release evidence aggregator model
+3	0	`whetstone_get_polyglot_cutover_readiness` MCP tool
+1	0	Phase 13a Integration — Layout Persistence
+2	0	Security Testing Skeleton Generation
+0	0	Modifier-Edge Shortcut Notation
+1	0	Architecture Templates
+3	0	Deprecation/succession lifecycle framework
+1	0	Language-specific method body scaffolding for core patterns
+1	0	Initializer Lists + STL Patterns
+3	0	Security scan orchestrator for generated targets
+0	0	`whetstone_dry_run_patch` MCP tool
+3	0	Cross-language rollout state snapshot model
+3	0	Security scan orchestrator for generated targets
+0	0	Cost telemetry integration in metrics stack
+2	0	Constitutional invariants compliance checker
+0	0	Property-based equivalence runner
+2	0	Regional rollout profile integration
+0	0	Compile/test proof binding
+3	0	Cross-language release evidence aggregator model
+2	0	`whetstone_get_certification_status` MCP tool
+3	0	Deprecation/succession lifecycle framework
+0	0	Sprint 193 Integration Summary
+1	0	Multi-language replay corpus runtime index model
+2	0	Benchmark suite for production capability claim
+3	0	`whetstone_publish_roadmap_epoch` MCP tool
+0	0	Build output parser — GCC/Clang format
+2	0	Rollout stage model (`pilot`, `canary`, `broad`, `default`)
+2	0	`whetstone_set_language_rollout_tier` MCP tool
+0	0	Breadcrumb Navigation
+1	0	Runtime-aware lowering extension points
+1	0	Phase 15b Integration — Full Protocol Test
+3	0	`whetstone_get_polyglot_cutover_readiness` MCP tool
+0	0	Python lowering adapter v2 (dynamic shape packets)
+0	0	Canonical fixture and oracle dataset packs
+0	0	Project context auto-injection
+0	0	LTS baseline criteria schema
+3	0	Security scan orchestrator for generated targets
+0	0	`whetstone_validate_tool_contracts` MCP tool
+1	0	Phase 20a Integration
+3	0	Cross-language release evidence aggregator model
+2	0	`whetstone_publish_next_block_plan` MCP tool
+2	0	Staged cutover planner (`shadow`, `dual-run`, `cutover`)
+3	0	Security scan orchestrator for generated targets
+3	0	Security scan orchestrator for generated targets
+0	0	Locals + Watches Panel Data Model
+0	0	Context Bundle Minimizer
+2	0	`whetstone_get_compliance_evidence` MCP tool
+3	0	`whetstone_get_polyglot_cutover_readiness` MCP tool
+2	0	Security Annotation Types — Subject 10
+0	0	Benchmark scorecard generator per pair
+1	0	Runtime-specific verification plugin interface
+2	0	Certification job scheduler
+3	0	Deprecation/succession lifecycle framework
+2	0	Tiered rollout and rollback control model
+2	0	Sprint 60 integration summary + release marker
+3	0	Cross-language rollout state snapshot model
+0	0	Constraint Violation Diagnostics
+1	0	Phase 12a Integration Tests
+3	0	Cross-language rollout state snapshot model
+1	0	Deterministic blocker taxonomy extension
+0	0	`whetstone_get_regeneration_decisions` MCP tool
+1	0	Long-horizon roadmap planning engine v2
+0	0	Intake extraction confidence scoring
+1	0	Phase 18b Integration + Sprint Summary
+3	0	`whetstone_get_polyglot_cutover_readiness` MCP tool
+2	0	Certification badges for roles/workflows
+0	0	Closeout consistency gate
+3	0	Cross-language rollout state snapshot model
+1	0	Binary/interface diff engine integration
+2	0	Release Readiness Gate Pack
+0	0	Source-vs-runtime drift comparator model
+1	0	Phase 12d Integration + Sprint 12 Summary
+1	0	`whetstone_verify_architecture_consistency` MCP tool
+0	0	API/ABI preservation report artifact
+0	0	Sprint 164 Integration Summary
+3	0	`whetstone_publish_roadmap_epoch` MCP tool
+0	0	`whetstone_capture_mcp_execution_trace` MCP tool
+0	0	Template recommendation model by project fingerprint
+0	0	Kotlin Generator
+3	0	Deprecation/succession lifecycle framework
+1	0	Cross-language capability contract normalizer model
+2	0	Release certification packet model
+2	0	Rollout blast-radius estimator
+3	0	Cross-language release evidence aggregator model
+2	0	Tenant-scoped waiver/approval ledger
+1	0	Governance model refresh framework
+2	0	`whetstone_publish_next_epoch_plan` MCP tool
+3	0	Cross-language release evidence aggregator model
+0	0	Embedded migration risk report template
+1	0	Cross-language node identity preservation policy model
+0	0	Key Symbol Rendering
+1	0	Burn-down planning optimizer
+3	0	Deprecation/succession lifecycle framework
+2	0	Interop certification policy bindings
+3	0	Cross-language release evidence aggregator model
+3	0	Deprecation/succession lifecycle framework
+2	0	Production summary hardening
+3	0	`whetstone_publish_roadmap_epoch` MCP tool
+2	0	Safety-critical deployment dossier artifact
+3	0	`whetstone_publish_roadmap_epoch` MCP tool
+2	0	Security-Preserving Translation
+1	0	Swarm maintenance job decomposition model
+3	0	`whetstone_get_polyglot_cutover_readiness` MCP tool
+1	0	Polyglot project migration graph model
+1	0	Phase 10d Integration Tests
+3	0	Cross-language rollout state snapshot model
+1	0	Phase 18a Integration
+2	0	Runtime-backed constructive release certification packet model
+1	0	Phase 29a Integration
+2	0	Production loop contract preflight
+1	0	Phase 20b Integration + Sprint Summary
+2	0	Pair-upgrade queue model
+2	0	`whetstone_get_tool_deprecations` MCP tool
+0	0	Wire `ArchitectIntakeProcessor` into `whetstone_architect_intake` handler
+2	0	Contract drift classifier (`compatible`, `warning`, `breaking`)
+1	0	Phase 23c Integration (tests: integration, regression)
+1	0	`FeedbackLoopOrchestrator` core loop
+1	0	WAT Cross-Language Projection
+0	0	Sprint 187 Integration Summary
diff --git a/specialists/data/generated/verification_type_eval.tsv b/specialists/data/generated/verification_type_eval.tsv
new file mode 100644
index 0000000..f7b3bc4
--- /dev/null
+++ b/specialists/data/generated/verification_type_eval.tsv
@@ -0,0 +1,172 @@
+2	0	`whetstone_apply_text_ast_merge` MCP tool
+2	0	Role-based onboarding model
+4	0	Security hardening report artifact
+3	0	Governance and audit report bundle
+0	0	VB.NET lowering/raising adapters
+4	0	Policy tuning report artifact
+1	0	`whetstone_get_learning_progress` MCP tool
+4	0	Family-level acceptance report generator
+3	0	Sprint 178 Integration Summary
+1	0	Leak/Corruption Signal Bridge
+3	0	Regression fixtures for under-constrained vs capability-limited cases
+2	0	`TestFailureRecord` schema
+2	0	MCP tool contract schema v1 model
+1	0	`whetstone_get_debug_evidence_index` MCP tool
+2	0	Error type synthesis for drone
+3	0	Sprint 154 integration summary + regression
+4	0	`whetstone_publish_next_block_plan` MCP tool
+1	0	`RegisterCodegenTools.h` — whetstone_generate_dispatch_table
+2	0	`whetstone_get_constructive_status` MCP tool
+0	0	Generator class-child traversal fix pack
+0	0	Theme System
+4	0	API/ABI preservation report artifact
+3	0	Sprint 138 integration summary + regression
+2	0	Safety case schema (`claim`, `evidence`, `assumption`, `risk`)
+3	0	Sprint 112 integration summary + regression
+3	0	Sprint 96 integration summary + regression
+1	0	`whetstone_rank_failure_triage_actions` MCP tool
+1	0	Disassembly + Register View Baseline
+3	0	Sprint 94 integration summary + regression
+4	0	Runtime selection report artifact
+2	0	Diagram/spec artifact ingestion adapters
+4	0	Program benchmarking report bundle
+2	0	Backward-compatibility serialization path
+1	0	Forbidden API policy checker
+4	0	Hybrid mode readiness report artifact
+4	0	Swarm maintenance report artifact
+0	0	`resource_conflict_unresolved` warning class
+2	0	Resilience and continuity requirements model
+1	0	`whetstone_get_region_policy_profile` MCP tool
+3	0	Sprint 117 integration summary + regression
+3	0	Sprint 79 integration summary + regression
+0	0	Staged cutover planner (`shadow`, `dual-run`, `cutover`)
+4	0	Equivalence evidence bundle exporter
+2	0	Numerical stability canonical model
+4	0	Drift outlook report artifact
+1	0	Phase 31a Integration
+3	0	API behavior regression runner integration
+2	0	Residual risk ownership model
+4	0	Documentation pack for language support + migration playbooks
+0	0	Marketplace policy bindings for tenants
+1	0	Compliance gate integration with support tiers
+2	0	Result-to-ObservationRecord mapper
+2	0	Multimodal evidence completeness scoring model
+0	0	Pattern library governance workflow
+0	0	Breadcrumb Navigation
+2	0	Conflict region detector model
+3	0	Sprint 148 integration summary + regression
+4	0	Public spec for verification evidence bundles
+3	0	Sprint 116 integration summary + regression
+1	0	Phase 11a Integration Tests
+0	0	Context slice canonicalization for handoffs
+0	0	Approval chain policy engine
+4	0	MCP closure publication bundle
+3	0	Sprint 132 integration summary + regression
+3	0	Sprint 207 Integration Summary
+3	0	Sprint 62 integration summary + regression
+0	0	Cost Estimation + Optimization
+0	0	Theme Token System v2
+2	0	GA evidence aggregation runtime model
+1	0	`whetstone_set_workstream_capacity` MCP tool
+2	0	Multi-check gate orchestrator model
+0	0	Compile/test proof binding
+0	0	Semantic pass aggregation in summaries
+1	0	`whetstone_transpile_dynamic_family` MCP tool
+3	0	Century retrospective evidence bundle composer
+2	0	`whetstone_rollback_last_patch` MCP tool
+0	0	Policy inheritance/override resolution engine
+2	0	`whetstone_get_constructive_rollout_status` MCP tool
+2	0	Polyglot project migration graph model
+1	0	Phase 12d Integration + Sprint 12 Summary
+3	0	Dependency and supply-chain audit packet
+0	0	Swarm-to-governance gate adapter
+1	0	Allocation/Ownership Trace Hooks
+2	0	Failover orchestration policy model
+4	0	Self-Hosting Metrics Report
+4	0	Multi-adapter compatibility report artifact
+4	0	Handoff integrity report artifact
+4	0	Campaign report bundle generator
+0	0	Supply-chain integrity checks for marketplace artifacts
+1	0	Real regenerate tool binding
+3	0	Sprint 66 integration summary + regression
+4	0	Lock metadata documentation packet
+3	0	Sprint 186 Integration Summary
+0	0	Kanban-Style Task Board
+4	0	Route integrity report artifact
+1	0	`whetstone_plan_rollout_stage` MCP tool
+4	0	Compatibility/deprecation report artifact
+3	0	Sprint 139 integration summary + regression
+0	0	Parse AnnotationConflictExtended.h
+3	0	Sprint 88 integration summary + regression
+2	0	Campaign checkpoint model
+4	0	Guided Architect-to-Execution Demo Mode
+2	0	`whetstone_resume_constructive_transaction` MCP tool
+0	0	SQL Annotation Mapping
+1	0	`whetstone_get_program_kpis` MCP tool
+3	0	Regression guard planner
+1	0	Tier promotion gate runner integration
+1	0	`whetstone_label_patch_risk` MCP tool
+1	0	`whetstone_parse_build_output` MCP tool
+4	0	Runtime freshness report artifact
+2	0	Campaign progress tracker model
+2	0	MCP closure gate aggregate schema
+4	0	Multi-adapter compatibility report artifact
+1	0	Snapshot persistence (sidecar integration)
+2	0	Prolog lowering adapter v1 (query/backtracking model)
+2	0	Efficiency metric model
+4	0	LTS publication bundle + operator handbook
+4	0	`whetstone_publish_next_block_plan` MCP tool
+1	0	`whetstone_verify_financial_migration` MCP tool
+0	0	Lint diagnostics normalization
+0	0	Dependency Graph Data
+4	0	MCP closure publication bundle
+1	0	Tamper detection and forensic event pipeline
+1	0	Sprint 40 Integration + Final Summary
+0	0	Async/await state intent lowering
+1	0	`whetstone_get_debt_inventory` MCP tool
+0	0	Runtime-aware raising policy selectors
+3	0	Systemic regression detector
+3	0	Phase 20b Integration + Sprint Summary
+0	0	Review-Gate Policy Refinement
+4	0	Low-level acceptance report + waiver system
+4	0	Conflict/merge report artifact
+0	0	Borrow semantics to references/views strategy
+2	0	`EnvironmentDiff` schema and computation
+0	0	Failure-class-driven patch proposer
+0	0	Internal implementation conformance runner
+2	0	Generic regeneration policy schema
+3	0	Sprint 201 Integration Summary
+3	0	Sprint 150 integration summary + full-program regression
+4	0	Enterprise rollout report template
+4	0	Federated evidence bundle composer
+4	0	Cross-language constructive execution report artifact
+1	0	Job submission via nexus HTTP API
+2	0	`resourceLocks` schema extension
+0	0	Assurance review dossier template
+4	0	Reliability and rollback report artifact
+3	0	Sprint 87 integration summary + regression
+2	0	`DistributedFailureEvidence` schema + normalizer
+3	0	Sprint 172 Integration Summary
+0	0	Exception + Stack Trace Capture
+4	0	Conflict/merge qualification report artifact
+1	0	CI integration in build/test flow
+3	0	Phase 23c Integration (tests: integration, regression)
+0	0	Next-epoch strategy generator
+0	0	`whetstone_normalize_diagnostics` runtime implementation
+4	0	Interop publication bundle
+0	0	Auto-generated mitigation suggestions for breaches
+1	0	`whetstone_propose_patch_for_failure` MCP tool
+3	0	Phase 22b Integration + Sprint Summary
+3	0	Strictness audit utility and baseline report
+2	0	Planner-executor handoff packet
+1	0	`whetstone_cluster_failures` MCP tool
+4	0	Pattern efficacy report artifact
+1	0	`whetstone_run_certification_cycle` MCP tool
+1	0	`whetstone_verify_executable_equivalence` MCP tool
+4	0	Guided transpilation exercise framework
+1	0	`whetstone_get_top_adapter_gaps` MCP tool
+2	0	Skeleton AST Support
+1	0	`whetstone_get_lts_support_matrix` MCP tool
+4	0	Documentation pack for language support + migration playbooks
+4	0	Reporting templates for C++ review teams
+2	0	Semantic diff canonical model
diff --git a/specialists/data/generated/verification_type_projects_eval.tsv b/specialists/data/generated/verification_type_projects_eval.tsv
new file mode 100644
index 0000000..2ec3b9f
--- /dev/null
+++ b/specialists/data/generated/verification_type_projects_eval.tsv
@@ -0,0 +1,9 @@
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Plan Split monolithic endpoint into read/write paths while preserving | Split monolithic endpoint into read/write paths while preserving latency and memory budgets and updating observability d
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+4	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+3	0	Define verification and readiness surface | Build an order saga orchestrator with compensating actions, timeout escalation, exactly-once message handling, and deter
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
diff --git a/specialists/data/generated/verification_type_projects_train.tsv b/specialists/data/generated/verification_type_projects_train.tsv
new file mode 100644
index 0000000..f3692e1
--- /dev/null
+++ b/specialists/data/generated/verification_type_projects_train.tsv
@@ -0,0 +1,51 @@
+3	0	Define verification and readiness surface | Build backup retention manager with policy tiers, legal hold support, and deletion safety checks.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+4	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+4	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+4	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | Build audit log service with immutable events, filtering, export, and retention policies.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+4	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+3	0	Define verification and readiness surface | Build session store service with token revocation, idle timeout, and concurrent login policy.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+4	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+4	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+4	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+4	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+3	0	Plan Build a streaming detector with keyed state windows | Build a streaming detector with keyed state windows, out-of-order event handling, watermark policy, and deterministic re
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | Build an inference API with model loading, request validation, and batch prediction endpoint.
+4	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+4	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Define verification and readiness surface | Build a whiteboard sync backend with operational ordering and conflict-safe patch application.
+3	0	Define verification and readiness surface | Build job queue service with delayed jobs, retries, visibility timeout, and metrics.
+4	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+4	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+4	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+4	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+4	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+4	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+3	0	Define verification and readiness surface | Build calendar backend with events, recurrence rules, reminders, and timezone normalization.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+4	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | Build a CDC replication worker that handles schema evolution, backfill merges, checkpoint consistency, and deterministic
+4	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+4	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+2	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
diff --git a/specialists/data/generated/verification_type_train.tsv b/specialists/data/generated/verification_type_train.tsv
new file mode 100644
index 0000000..e9bba34
--- /dev/null
+++ b/specialists/data/generated/verification_type_train.tsv
@@ -0,0 +1,978 @@
+3	0	Sprint 51 integration summary + regression
+4	0	Translation Report
+1	0	`whetstone_check_tool_compatibility` MCP tool
+2	0	C/C++ Constructive Edit Adapter
+3	0	Sprint 99 integration summary + regression
+2	0	Stage checkpoint and resume model
+4	0	Dynamic family acceptance report + review queue wiring
+3	0	Sprint 181 Integration Summary
+2	0	Semantic-noise diff suppressor model
+0	0	Invariant and intent-diff checks
+1	0	`whetstone_analyze_rust_semantics` MCP tool
+4	0	Program constitution + handbook publication bundle
+1	0	`whetstone_generate_patch_candidates` MCP tool
+0	0	Unmappable semantic blocklist + mandatory review gate
+1	0	Phase 15a Integration Tests
+2	0	Debug policy constraint model
+3	0	Sprint 177 Integration Summary
+3	0	Sprint 56 integration summary + regression
+0	0	Canonical fixture and oracle dataset packs
+2	0	Wire `ArchitectIntakeProcessor` into `whetstone_architect_intake` handler
+2	0	`whetstone_save_repro_packet` MCP tool
+4	0	Guided workflow state machine for self-service runs
+3	0	Sprint 64 integration summary + regression
+2	0	Auto Type Deduction
+1	0	`whetstone_get_governance_state` MCP tool
+1	0	`whetstone_verify_scientific_migration` MCP tool
+4	0	Replay and determinism report artifact
+4	0	Conflict/merge qualification report artifact
+4	0	Reviewer effectiveness report artifact
+3	0	Sprint 189 Integration Summary
+3	0	End-to-end constructive smoke scenarios per language family
+2	0	`whetstone_record_attempt` MCP tool
+3	0	Benchmark trend regression detector
+0	0	Context Window Optimization
+3	0	Sprint 191 Integration Summary
+2	0	Diagnostic normalization canonical model
+2	0	ABI and calling-convention canonical model
+0	0	Go lowering adapter v1
+1	0	`whetstone_get_mcp_closure_status` MCP tool
+2	0	Certification dashboard data model
+1	0	Phase 21a Integration
+3	0	Patch execution audit bundle
+4	0	Permission conformance report artifact
+4	0	API/ABI preservation report artifact
+1	0	Apiary browser panel
+3	0	Pair regression detector with severity levels
+3	0	Sprint 141 integration summary + regression
+2	0	Resource budget model (memory/CPU/IO)
+1	0	`whetstone_plan_budget_allocation` MCP tool
+4	0	Multi-language constructive loop report artifact
+1	0	`whetstone_list_toolchain_providers` runtime implementation
+2	0	`whetstone_capture_handoff_packet` MCP tool
+1	0	`whetstone_probe_toolchain_provider` MCP tool
+2	0	Orchestrator MCP handoff packet schema
+3	0	Sprint 71 integration summary + regression
+3	0	Sprint 55 integration summary + regression
+3	0	Sprint 68 integration summary + regression
+3	0	Sprint 142 integration summary + regression
+3	0	Sprint 169 Integration Summary
+1	0	Architect Tooling — RPC + MCP
+3	0	Sprint 129 integration summary + regression
+0	0	Decision consistency analyzer across reviewers
+2	0	Markdown Spec Parser
+0	0	Candidate generation engine from failure taxonomy
+1	0	`whetstone_get_representative_language_profile` MCP tool
+4	0	Web ecosystem migration report template
+1	0	`whetstone_architect_intake` MCP Tool
+2	0	Root-cause clustering model
+4	0	Data family acceptance report
+2	0	Escalation reason classifier model
+0	0	Cross-platform CMake target generator
+1	0	`whetstone_list_toolchain_providers` MCP tool
+1	0	Sprint 37 Integration + Summary
+1	0	`whetstone_select_language_adapter` MCP tool
+3	0	Sprint 195 Integration Summary
+0	0	Module Decomposition Engine
+0	0	Governance and safety invariant revalidation pass
+1	0	`whetstone_get_century_status` MCP tool
+2	0	Output payload shape profiler model
+1	0	Phase 23a Integration
+0	0	Background task slots (parallel agent work)
+3	0	Adapter health and conformance score model
+0	0	Placeholder/TODO detector and policy
+3	0	Phase 18b Integration + Sprint Summary
+1	0	Deterministic replay pack composer
+3	0	Sprint 125 integration summary + regression
+4	0	Route integrity report artifact
+0	0	Security scan orchestrator for generated targets
+0	0	ARM Assembly Parser
+3	0	Sprint 43 Integration Summary
+2	0	Feedback aggregation and prioritization model
+3	0	Sprint 210 Integration Summary
+0	0	Floating-point determinism risk classifier
+0	0	Package state probe
+0	0	Window Chrome + Hierarchy Pass
+1	0	`whetstone_list_language_adapters` MCP tool
+3	0	Rollout health scoring model
+3	0	Sprint 135 integration summary + regression
+0	0	Threshold recommendation engine
+4	0	Translation Report
+3	0	Timing regression runner integration
+1	0	`whetstone_assemble_fix_context` MCP tool
+4	0	Standards publication bundle
+2	0	Runtime selection decision packet model
+0	0	WAT Annotation Mapping
+2	0	Determinism qualification packet model
+3	0	Sprint 60 integration summary + release marker
+3	0	Sprint 128 integration summary + regression
+4	0	Permission conformance report artifact
+1	0	`whetstone_validate_patch_candidate` MCP tool
+0	0	Capability Discovery Panels
+4	0	Federated governance report generator
+3	0	Sprint 126 integration summary + regression
+4	0	Resilience operations report artifact
+3	0	Sprint 144 integration summary + regression
+1	0	Sprint 32 Integration + Summary
+3	0	Sprint 156 integration summary + regression
+0	0	Workflow Status Overlay
+2	0	Distributed evidence quality scorer model
+1	0	`whetstone_get_tool_deprecations` MCP tool
+1	0	`whetstone_execute_rollout_or_rollback` MCP tool
+4	0	Equivalence evidence bundle exporter
+2	0	`ArchitectIntakeProcessor` — merge all requirement kinds
+3	0	Sprint 115 integration summary + regression
+2	0	Cross-language merge policy engine model
+2	0	`run_pipeline` quality payload hardening
+4	0	Per-target projection validation report
+0	0	F# Parser
+3	0	Sprint 184 Integration Summary
+3	0	Sprint 208 Integration Summary
+2	0	Determinism gate policy model
+3	0	Sprint 120 integration summary + regression
+2	0	Strict contract mode for generation
+0	0	Traceability chain generator (origin->current)
+1	0	`whetstone_set_authoring_mode` MCP tool
+4	0	Cost Tracking + Reporting
+4	0	Regulated-domain migration report template
+1	0	`whetstone_reduce_repro_command` MCP tool
+1	0	`whetstone_preview_regeneration_diff_generic` MCP tool
+1	0	Phase 10d Integration Tests
+4	0	Replay/transaction reliability report artifact
+0	0	KPI delta analysis engine for block review
+3	0	Sprint 182 Integration Summary
+0	0	Swarm progress and fault recovery policy
+0	0	Process state probe
+2	0	Transpilation knowledge graph schema
+4	0	Lock validation aggregate report fields
+4	0	Weekly quality report artifact generator
+2	0	`whetstone_preview_text_ast_merge_generic` MCP tool
+3	0	Queue health scoring model
+3	0	Drift audit report artifact
+4	0	Evidence bundle store + deterministic ordering
+0	0	Pattern applicability matcher from graph context
+2	0	Conflict preview and impact estimator model
+2	0	Contract breach severity classifier
+0	0	Long-range benchmark corpus setup
+2	0	Debug hint template model
+3	0	Sprint 90 integration summary + full-program regression
+0	0	Recursive object/array rule generator
+4	0	HiveMind job publisher from editor
+0	0	Constraint encoding and strict scalar handling
+4	0	Runtime compatibility risk report generator
+4	0	`whetstone_get_reference_conformance_report` MCP tool
+1	0	`whetstone_build_debug_handoff` MCP tool
+3	0	Sprint 91 integration summary + regression
+1	0	`whetstone_list_campaign_checkpoints` MCP tool
+2	0	Cross-service contract checkpoint model
+0	0	Snapshot freeze workflow and integrity checks
+1	0	`whetstone_export_mcp_replay_pack` MCP tool
+1	0	Regional rollout profile integration
+3	0	Sprint 190 Integration Summary
+3	0	Phase 19b Integration + Sprint Summary
+0	0	Secure-by-Default Code Generation
+0	0	PostgreSQL Generator
+2	0	Debug session handoff model
+1	0	`whetstone_configure_hivemind` MCP tool
+1	0	`whetstone_generate_safe_patch_plan` MCP tool
+2	0	Build output parser — GCC/Clang format
+4	0	`whetstone_start_guided_migration` MCP tool
+2	0	Best-practice pack schema and versioning
+4	0	Self-service adoption report artifact
+1	0	`whetstone_get_language_matrix` MCP tool
+0	0	Calibration corpus loader
+1	0	`whetstone_score_failure_triage` MCP tool
+3	0	Full regression + architecture gate
+0	0	Event-order and state-sync verifier
+3	0	Sprint 127 integration summary + regression
+1	0	`whetstone_run_bisect_debug` MCP tool
+0	0	Cross-Language Consistency Gate
+1	0	`whetstone_get_execution_attestation` MCP tool
+4	0	Reviewer effectiveness report artifact
+3	0	Sprint 81 integration summary + regression
+0	0	WAT Cross-Language Projection
+0	0	Real provider probe execution paths
+0	0	Next-block priority planner and constraint binder
+0	0	Baseline-vs-target perf comparator
+2	0	Forecast confidence and uncertainty model
+4	0	Team onboarding readiness report artifact
+0	0	ARM assembly lowering adapter v1
+3	0	Class-heavy regression fixture suite
+1	0	Cost telemetry integration in metrics stack
+4	0	`whetstone_publish_next_epoch_plan` MCP tool
+1	0	`whetstone_get_polyglot_cutover_readiness` MCP tool
+1	0	`whetstone_list_representative_language_profiles` MCP tool
+2	0	`ABTestComparison` — compare two session records
+1	0	Strategy recommendation hooks from graph context
+0	0	Query behavior equivalence runner (set-based diff checks)
+3	0	Sprint 70 integration summary + regression
+2	0	Cross-language capability contract normalizer model
+1	0	`whetstone_attach_multimodal_evidence` MCP tool
+0	0	Deterministic fallback policy for unsupported operations
+3	0	Sprint 174 Integration Summary
+4	0	Drift outlook report artifact
+3	0	Sprint 85 integration summary + regression
+1	0	Tool reachability runner with deterministic probes
+3	0	Phase 11e Integration + Sprint Summary
+4	0	Constructive GA publication bundle artifact
+4	0	Regeneration stability report artifact
+4	0	Escalation report composer
+2	0	Migration template schema and catalog model
+4	0	Embedded migration risk report template
+4	0	Self-service adoption report artifact
+4	0	Compatibility/deprecation report artifact
+3	0	Sprint 157 integration summary + regression
+3	0	Dependency Audit Annotations
+2	0	Porting cost model v1
+1	0	Phase 26a Integration
+4	0	`whetstone_get_federated_verification_report` MCP tool
+2	0	Campaign resume planner model
+2	0	Toolchain provider interface schema
+0	0	Security-Preserving Translation
+3	0	Sprint 97 integration summary + regression
+3	0	Rebuild `whetstone_mcp` + smoke test
+4	0	Federated governance report generator
+0	0	Scenario — Legacy Modernization
+2	0	Constrained Routing Ruleset Extension
+0	0	Architecture claim linkage engine
+1	0	`whetstone_poll_iteration_job` MCP tool
+4	0	Practice adoption and outcome report artifact
+1	0	Phase 27a Integration
+1	0	`whetstone_validate_debug_action` MCP tool
+1	0	`whetstone_get_epoch_block_status` MCP tool
+4	0	Evidence bundle store + deterministic ordering
+2	0	Contract completeness validator
+3	0	Sprint 75 integration summary + regression
+2	0	External verifier adapter interface
+2	0	Recursion bound/termination contracts
+1	0	`whetstone_verify_architecture_consistency` MCP tool
+2	0	Data-model field materialization from source AST
+0	0	Route integrity gate bindings
+1	0	Phase 24a Integration
+2	0	Deprecation policy model for unstable/obsolete pairs
+4	0	`whetstone_publish_roadmap_epoch` MCP tool
+2	0	Capability decision packet model
+0	0	`whetstone_restore_hybrid_checkpoint` runtime implementation
+3	0	Sprint 53 integration summary + regression
+3	0	Sprint 164 Integration Summary
+0	0	Fine-grained capability grant engine
+0	0	Queue remediation call-plan emission
+4	0	Language capability conformance report artifact
+1	0	Phase 10e Integration Tests
+4	0	`whetstone_get_pair_scorecard` MCP tool
+1	0	`whetstone_get_session_state` MCP tool
+2	0	C++ constructive readiness scorer model
+0	0	Capability Vocabulary & Validation
+3	0	Sprint 95 integration summary + regression
+2	0	Region policy overlay model
+2	0	Bullet-list requirement extractor
+0	0	Async mapping strategy (coroutines/task runtime profiles)
+4	0	Efficiency diagnostics and trend report
+1	0	Semantic validator core API
+3	0	Sprint 77 integration summary + regression
+0	0	Scheme Generator
+3	0	Sprint 111 integration summary + regression
+2	0	`whetstone_replay_repro_packet` MCP tool
+4	0	Constructive GA publication bundle artifact
+2	0	Debug assist packet bundle model
+4	0	Tool-surface diagnostics report artifact
+2	0	Argument/result canonicalizer model
+4	0	Policy tuning report artifact
+3	0	Sprint 197 Integration Summary
+2	0	Root-cause ranking policy model
+1	0	`whetstone_plan_swarm_maintenance` MCP tool
+4	0	Weekly quality report artifact generator
+3	0	Sprint 104 integration summary + regression
+4	0	Dynamic family acceptance report + review queue wiring
+1	0	`whetstone_propose_adapter_improvements` MCP tool
+0	0	Blocked reason normalization
+4	0	Contract conformance report artifact
+0	0	Maintenance cadence scheduler
+1	0	`whetstone_get_debug_campaign_status` MCP tool
+3	0	Sprint 176 Integration Summary
+0	0	VB.NET Generator
+1	0	`whetstone_get_failure_trends` MCP tool
+3	0	Sprint 145 integration summary + regression
+1	0	Phase 14b Integration
+2	0	Governance model refresh framework
+3	0	Sprint 114 integration summary + regression
+2	0	Loop retry and timeout policy model
+4	0	Epoch kickoff report artifact
+3	0	Session state machine transitions
+2	0	Epoch-2 workstream schema and ownership map
+0	0	Reviewer-focused semantic diff summarizer
+3	0	Sprint 78 integration summary + regression
+0	0	Cross-team benchmark comparator
+0	0	Historical KPI aggregator
+0	0	x86 assembly lowering adapter v1
+0	0	Async, Parallelism & Error Handling Annotations
+4	0	Benchmark publication bundle
+1	0	Tool manifest build-gate integration
+0	0	Fix `visitModule` in Python, Rust, Go, Elisp generators
+1	0	Pipeline support for injected capability signals
+4	0	Requirements gap report
+4	0	Release readiness scoreboard
+2	0	Packet tagging integration across pipelines
+3	0	Sprint 105 integration summary + regression
+3	0	Sprint 108 integration summary + regression
+0	0	Java lowering adapter v1
+4	0	`whetstone_get_numerical_risk_report` MCP tool
+1	0	`whetstone_generate_project` MCP tool
+4	0	Representative language readiness report artifact
+0	0	`.whetstone.json` for the editor project itself
+3	0	Sprint 46 integration summary
+3	0	Audit trail and inspection
+4	0	Reviewer effectiveness report artifact
+4	0	Patch proposal review report artifact
+3	0	Sprint 58 integration summary + regression
+1	0	LTS risk watchlist and escalation hooks
+3	0	Sprint 196 Integration Summary
+4	0	Constructive GA publication bundle
+1	0	.NET Integration
+0	0	WorkerRegistry.h — Worker Abstractions
+3	0	Sprint 92 integration summary + regression
+1	0	`whetstone_marketplace_search` MCP tool
+3	0	Full Test Suite Regression
+2	0	Class-generation contract schema
+4	0	Program constitution + handbook publication bundle
+1	0	`whetstone_transpile_query_family` MCP tool
+3	0	Sprint 72 integration summary + regression
+1	0	`whetstone_export_repro_jsonl` MCP tool
+0	0	Production summary hardening
+0	0	Quarterly lock/review cadence policy
+2	0	Cross-language regeneration quality packet model
+2	0	Requirement Normalization and Conflict Detection
+4	0	Compatibility/deprecation report artifact
+0	0	`SelfContainmentScorer` — score a taskitem 0–100
+0	0	Ownership mapping policy engine (`unique_ptr/shared_ptr/value`)
+4	0	Incremental sync report artifact
+2	0	Wiring closure severity classifier
+4	0	Representative language readiness report artifact
+2	0	Recovery safety packet model
+2	0	`whetstone_get_constructive_rollout_status` runtime implementation
+3	0	Sprint 152 integration summary + regression
+4	0	Semantic memory report artifact
+0	0	Dependency Graph Visualization
+0	0	F# lowering/raising adapters
+4	0	`whetstone_get_numerical_risk_report` MCP tool
+2	0	Generic incremental edit packet schema
+0	0	Workflow Sidecar Persistence
+0	0	Key Symbol Rendering
+4	0	Multi-language constructive loop report artifact
+2	0	Call Stack and Frame Inspector
+1	0	Phase 20a Integration
+4	0	Constructive runtime readiness report bundle
+2	0	`whetstone_run_constructive_step`/`status`/`loop` runtime packet model
+4	0	Contract conformance report artifact
+4	0	Regeneration stability report artifact
+0	0	Safety case completeness checker
+4	0	Benchmark scorecard generator per pair
+0	0	Ruby lowering adapter v1 (meta-programming boundaries)
+3	0	Cost vs Quality Regression Suite
+2	0	Resume and rollback policy runtime model
+1	0	Priority-focused review queue shaping engine
+0	0	Self-Hosting Test Harness
+0	0	Scaffold File Generation
+1	0	Phase 18a Integration
+1	0	Build iteration session integration
+3	0	Profile compliance auditor integration
+1	0	Corpus ingestion tooling and normalization
+0	0	Safety-critical deployment dossier artifact
+4	0	Swarm maintenance report artifact
+2	0	MCP compatibility policy schema
+2	0	Stewardship and ownership policy model
+2	0	Serialization + Dispatch for New Nodes
+4	0	`whetstone_publish_roadmap_epoch` MCP tool
+3	0	Sprint 153 integration summary + regression
+3	0	Sprint 54 integration summary + regression
+2	0	Patch risk labeler model
+1	0	`whetstone_resume_debug_campaign` MCP tool
+4	0	Reliability operations report artifact
+2	0	`whetstone_rollback_constructive_transaction` MCP tool
+0	0	CMake generator from Whetstone project description
+3	0	Intake regression corpus loader
+2	0	Debug recipe step model
+4	0	Efficiency diagnostics and trend report
+0	0	Completion Overlay UX Hardening
+3	0	Sprint 183 Integration Summary
+2	0	Execution hint extraction from requirements
+2	0	Confidence synthesis model
+3	0	Sprint 161 Integration Summary
+0	0	Hint feature extraction from decision ledgers
+3	0	Sprint 171 Integration Summary
+0	0	Fuzz differential runner with seed replay
+1	0	`whetstone_generate_debug_hints` MCP tool
+3	0	Sprint 121 integration summary + regression
+1	0	`whetstone_list_plugins` MCP tool
+4	0	Benchmark publication bundle
+0	0	Scope & Namespace Annotations
+4	0	Campaign report bundle generator
+1	0	`whetstone_get_upgrade_queue` MCP tool
+3	0	Sprint 179 Integration Summary
+1	0	Cross-session context bridge
+4	0	Documentation + Operator Playbooks
+3	0	Sprint 80 integration summary + full-program regression
+2	0	Cross-tenant adapter sharing contract
+4	0	Cost-vs-quality tradeoff report generator
+3	0	Audit trail completeness checker
+2	0	Scheduler lock contract emitter
+1	0	Review SLA policy integration
+1	0	Approval workflow integration with governance ledger
+1	0	Static registry scanner for `register*Tools` blocks
+1	0	Phase 22a Integration
+1	0	`whetstone_set_runtime_profile` MCP tool
+2	0	Security Annotation Types — Subject 10
+0	0	Tier promotion engine (beta->stable)
+2	0	Locals + Watches Panel Data Model
+1	0	`whetstone_get_related_migration_patterns` MCP tool
+3	0	Phase 24b Integration + Sprint Summary
+2	0	Debug stop-reason classifier model
+1	0	Modernization RPC + MCP
+3	0	Sprint 194 Integration Summary
+2	0	Architect Review Interface
+3	0	Sprint 175 Integration Summary
+1	0	`whetstone_register_external_verifier` MCP tool
+4	0	Swarm maintenance report artifact
+1	0	`whetstone_capture_distributed_failure` MCP tool
+0	0	`TokenBudgetEnforcer` — trim assembled context to budget
+4	0	Plugin trust and risk report artifact
+3	0	Sprint 93 integration summary + regression
+1	0	New Languages Integration + MCP
+2	0	Pipeline promotion packet integration
+2	0	Real GA evidence packet generation
+0	0	Gate-driven remediation planner
+1	0	Migration Execution Integration
+3	0	Hint safety guardrails and audit fields
+4	0	Interop publication bundle
+4	0	Portfolio strategy report artifact
+1	0	`whetstone_set_language_rollout_tier` MCP tool
+2	0	KPI anomaly triage model
+0	0	Config file probe
+0	0	C++ runner adapter for harness
+1	0	`whetstone_explain_transpilation_decision` MCP tool
+4	0	Federated evidence bundle composer
+3	0	`whetstone_regression_guard` MCP tool
+1	0	Skill progression tracker integration
+0	0	Validation capability-signal scoring/classification
+4	0	`whetstone_publish_next_epoch_plan` MCP tool
+1	0	`whetstone_plan_tool_migrations` MCP tool
+1	0	`whetstone_get_debug_checklist` MCP tool
+2	0	`AgentSessionRecorder` — per-session tool call instrumentation
+2	0	Workspace-source freshness comparator model
+0	0	Queue+validation metadata enrichment
+1	0	`whetstone_get_debug_metrics` MCP tool
+0	0	Architecture Templates
+3	0	Sprint 185 Integration Summary
+4	0	Replay and determinism report artifact
+2	0	STL algorithm lifting from IR intent tags
+3	0	Epoch block health re-baseline model
+0	0	Alert policy engine for SLO breaches
+2	0	Validation lock name-format checks
+1	0	`whetstone_generate_inference_job` MCP tool
+0	0	Smart Pointer Patterns
+0	0	Legacy semantic recovery graph engine
+0	0	Context Bundle Minimizer
+3	0	Sprint 47 integration summary + regression
+2	0	Patch execution record model
+3	0	Sprint 137 integration summary + regression
+3	0	Sprint 163 Integration Summary
+1	0	Phase 10c Integration Tests
+2	0	Debugging metrics packet (`time_to_green`, `iterations`, etc.)
+1	0	`whetstone_export_tool_manifest` MCP tool
+1	0	`whetstone_run_continuity_drill` MCP tool
+1	0	`whetstone_get_debug_trace` MCP tool
+0	0	`PrerequisiteOpResolver` — resolve ops to workspace paths
+1	0	Phase 19a Integration
+3	0	Sprint 124 integration summary + regression
+2	0	`whetstone_get_safety_profile_requirements` MCP tool
+1	0	Generator Updates + Phase 11c Integration
+3	0	Sprint 118 integration summary + regression
+0	0	False-green diagnostic class
+2	0	Sidecar AST Persistence
+1	0	`whetstone_queue_ready` MCP Tool
+0	0	Conflict avoidance and lock orchestration engine
+3	0	Sprint 136 integration summary + regression
+3	0	Sprint 61 integration summary + regression
+4	0	Plugin trust and risk report artifact
+2	0	Bisect candidate selector model
+1	0	`whetstone_step_debug_campaign` MCP tool
+4	0	Runtime execution trace report artifact for sync/regeneration/merge tools
+4	0	Language capability conformance report artifact
+1	0	Phase 14d Integration + Sprint 14 Summary
+1	0	Phase 11b Integration Tests
+3	0	Transition plan model for Sprint 101+ workstreams
+4	0	Cost-vs-quality tradeoff report generator
+1	0	Sprint 26 Integration + Summary
+2	0	Swarm maintenance job decomposition model
+2	0	Fast rollback orchestration contract
+0	0	Continuity drill framework and scoring
+0	0	Multi-language minimal test harness runner
+1	0	`whetstone_list_distributed_failures` MCP tool
+0	0	Top-level strictness gate
+4	0	Portfolio strategy report artifact
+3	0	Century milestone publication package
+3	0	Sprint 203 Integration Summary
+3	0	Sprint 41 Integration Summary
+0	0	Failure fixture catalog for deterministic debug benchmarks
+4	0	Debt burn-down report artifact
+2	0	Minimal repro reducer model
+2	0	Result Acceptance Protocol
+2	0	Host boundary + FFI contract generator
+0	0	Workspace Onboarding Flow
+4	0	Generic sync determinism report artifact
+3	0	Post-incident continuity audit report model
+2	0	SLM debug readiness scoring model
+2	0	`whetstone_run_constructive_ga_gate` runtime implementation
+4	0	HiveMind job publisher from editor
+3	0	Docking Reliability Audit + Deterministic Layout Rules
+2	0	Pair-specific known-limitations generator
+0	0	Compliance policy binding engine
+0	0	Context Assembly + Budget
+0	0	Artifact fingerprinting and drift detection
+3	0	Sprint 131 integration summary + regression
+1	0	Verification gate integration for architecture artifacts
+0	0	Python + Java Parser Deepening
+3	0	Sprint 202 Integration Summary
+2	0	End-to-end validation with the failing A/B test spec
+2	0	Explainability compression model for rapid review
+0	0	Scheme lowering/raising adapters
+2	0	Failure trend tracker model
+2	0	Range-Based For + Structured Bindings
+3	0	Sprint 165 Integration Summary
+4	0	Marketplace governance report artifact
+3	0	Sprint 192 Integration Summary
+1	0	`whetstone_start_debug_campaign` MCP tool
+0	0	Property-based equivalence runner
+2	0	Rollback ledger model
+2	0	Capability negotiation and fallback policy model
+0	0	Trial evaluation rubric (quality/risk/cost)
+1	0	`whetstone_plan_deprecation_or_upgrade` MCP tool
+0	0	Constraint completeness diagnostics
+4	0	`whetstone_get_artifact_trust_report` MCP tool
+0	0	Find and Replace Improvements
+2	0	Compatibility simulation runner model
+1	0	Binary/footprint gate integration
+2	0	`whetstone_validate_tool_contracts` MCP tool
+2	0	JavaScript lowering adapter v2 (prototype/object model)
+2	0	Rust/Go Constructive Edit Adapter
+1	0	Phase 25b Integration
+0	0	C Generator
+1	0	`whetstone_propose_policy_tuning` MCP tool
+3	0	Scope and Milestone Decomposer
+1	0	Policy-compliance auto-checkpoint integration
+0	0	Incident containment automation for compromised components
+2	0	Worker Efficiency Dashboard Data Model
+4	0	`whetstone_get_artifact_trust_report` MCP tool
+3	0	Sprint 187 Integration Summary
+3	0	Sprint 123 integration summary + regression
+3	0	Sprint 146 integration summary + regression
+2	0	Constructive transaction schema
+2	0	Representative language conformance packet model
+0	0	Projection metadata summary export
+4	0	Data family acceptance report
+4	0	Documentation + Operator Playbooks
+1	0	`whetstone_validate_patch_proposal` MCP tool
+1	0	`whetstone_probe_tool_reachability` MCP tool
+2	0	Hybrid core contract schema v1
+2	0	IR lowering interface and adapter API
+1	0	Adoption telemetry integration
+1	0	`whetstone_select_mcp_runtime` MCP tool
+3	0	Sprint 113 integration summary + regression
+1	0	`whetstone_plan_debt_burndown` MCP tool
+4	0	Self-Hosting Progress Report + Phase Integration
+3	0	Tier transition guardrail and rollback runtime model
+1	0	`whetstone_save_campaign_checkpoint` MCP tool
+2	0	`whetstone_run_spec_conformance` MCP tool
+0	0	Kotlin lowering/raising adapters
+1	0	`whetstone_dry_run_patch` MCP tool
+4	0	Tool-surface diagnostics report artifact
+2	0	`whetstone_get_pair_stability_forecast` MCP tool
+4	0	Misclassification report export
+1	0	Phase 38a Integration
+3	0	Full historical regression + post-release checklist
+0	0	Actor-to-thread/async projection policy set
+2	0	Strict-mode escalation for low-specificity tasks
+0	0	Preventive mitigation planner
+2	0	Step Spec Expander (tests: unit, negative, boundary)
+0	0	Optimization Annotation Completion
+0	0	Parse AnnotationValidatorExtended.h
+4	0	`whetstone_get_sync_identity_report` MCP tool
+0	0	Value-Forward Onboarding Flow
+3	0	Runtime health probe runner
+3	0	Sprint 44 Integration Summary
+0	0	Federated reproducibility validator
+3	0	Sprint 89 integration summary + regression
+4	0	`whetstone_start_guided_migration` MCP tool
+2	0	Tool-surface checksum packet model
+3	0	Sprint 211 Integration Summary
+1	0	`whetstone_list_migration_templates` MCP tool
+0	0	Deterministic blocker taxonomy
+2	0	Language-neutral operation envelope model
+2	0	`whetstone_record_debug_trace` MCP tool
+1	0	`whetstone_export_debug_campaign_bundle` MCP tool
+1	0	`whetstone_run_test_iteration` MCP tool
+4	0	Hybrid mode readiness report artifact
+3	0	Sprint 140 integration summary + full-program regression
+0	0	Agent task status overlay
+1	0	Workstream drift detector integration
+4	0	Replay/transaction reliability report artifact
+1	0	Annotation RPC Methods
+4	0	Embedded migration risk report template
+4	0	Explainability report template pack
+1	0	`whetstone_install_or_disable_plugin` MCP tool
+0	0	Modernization Suggestions
+2	0	Rollout blast-radius estimator
+4	0	Handoff integrity report artifact
+1	0	`whetstone_get_runtime_fingerprint` MCP tool
+0	0	Inline Annotation Badges
+4	0	Web ecosystem migration report template
+0	0	T-SQL lowering/raising adapters
+1	0	Tool call visualization
+2	0	Ecosystem drift signal model
+4	0	Conflict/merge report artifact
+2	0	Maintenance TCO forecasting engine
+4	0	Epoch renewal publication bundle
+0	0	AppImage / .deb package generator
+3	0	Sprint 52 integration summary + regression
+3	0	Sprint 122 integration summary + regression
+4	0	Program benchmarking report bundle
+3	0	Sprint 110 integration summary + full-program regression
+1	0	`whetstone_start_onboarding_path` MCP tool
+1	0	`whetstone_transpile_systems_family` MCP tool
+0	0	ProjectionGenerator — Subject 5-8 + Semantic Visitors
+1	0	`whetstone_get_api_semantics` MCP tool
+1	0	`whetstone_optimize_review_queue` MCP tool
+0	0	Cross-language class emission parity tests
+4	0	Toolchain readiness report artifact
+1	0	`whetstone_validate_policy_tuning` MCP tool
+4	0	MCP Tool Documentation
+0	0	Queue/validation fingerprint convergence
+3	0	`TaskitemQualityAuditor` — batch report
+2	0	Runtime selection policy model
+3	0	Phase 23b Integration + Sprint Summary
+3	0	Strictness regression suite
+3	0	Sprint 48 integration summary + regression
+1	0	`whetstone_plan_polyglot_migration` MCP tool
+2	0	SQL canonical query IR layer
+4	0	C++ constructive loop report artifact
+3	0	Sprint 151 integration summary + regression
+1	0	`whetstone_generate_safety_case` MCP tool
+4	0	Runtime execution trace report artifact for sync/regeneration/merge tools
+4	0	Security hardening report artifact
+4	0	Public spec for verification evidence bundles
+3	0	Sprint 158 integration summary + regression
+0	0	KPI-to-workstream binding engine
+0	0	End-to-end replay + rollback scenario suite
+1	0	Transpilation RPC + MCP
+1	0	`whetstone_verify_handoff_integrity` MCP tool
+3	0	Phase 17b Integration + Sprint Summary
+4	0	Constructive runtime readiness report bundle
+0	0	Target projection blocker taxonomy
+1	0	Runtime `tools/list` snapshot normalizer
+0	0	C# Generator
+4	0	MCP Tool Documentation
+4	0	Generic sync determinism report artifact
+2	0	AST mutation preview
+0	0	Config auto-discovery (walk up from CWD)
+4	0	SemannoFormat.h — Comment Format Standard
+3	0	Sprint 193 Integration Summary
+1	0	Test iteration session integration
+3	0	Century milestone health scoreboard model
+2	0	Allowed-file scope policy model
+3	0	Sprint 59 integration summary + regression
+2	0	Debug recipe library model
+1	0	Real sync tool binding
+4	0	Guided workflow state machine for self-service runs
+2	0	Compatibility risk packet model
+1	0	Phase 28a Integration
+3	0	Sprint 57 integration summary + regression
+2	0	Candidate ranking and recommendation model
+4	0	Regulated-domain migration report template
+0	0	Union support (`oneOf`/`anyOf`/`allOf`)
+4	0	Reliability operations report artifact
+2	0	`whetstone_export_tool_contracts` MCP tool
+3	0	Sprint 49 integration summary + regression
+1	0	MCP replay diagnostics artifact
+0	0	Tenant isolation guardrails for logs/artifacts
+1	0	`whetstone_suggest_test_fix` MCP tool
+1	0	`whetstone_run_mcp_closure_gate` MCP tool
+3	0	Sprint 168 Integration Summary
+2	0	Pair stability forecasting engine
+3	0	Sprint 162 Integration Summary
+3	0	Sprint 206 Integration Summary
+2	0	Runtime-backed constructive release certification packet model
+0	0	Full loop driver (end-to-end stepping)
+1	0	Parallel dispatch optimizer with dependency constraints
+0	0	Queue lock-conflict detection
+2	0	Debug campaign spec model
+2	0	Representative language toolchain provider registry model
+1	0	`whetstone_run_interop_testbed` MCP tool
+3	0	Sprint 73 integration summary + regression
+2	0	Concurrency Annotations — Primitives & Memory Model
+2	0	Job schema → typed C++ structs generator
+4	0	LTS publication bundle + operator handbook
+2	0	Hint confidence + uncertainty packet format
+0	0	Workspace mode policy bindings
+0	0	Notification + Status Signaling Refresh
+1	0	`whetstone_get_portfolio_economics` MCP tool
+4	0	Setup verification report
+3	0	Sprint 76 integration summary + regression
+1	0	`whetstone_start_feedback_loop` MCP tool
+2	0	Non-C++ adapter conformance harness model
+1	0	`whetstone_generate_setup_script` MCP tool
+0	0	Java raising adapter v1
+2	0	EnvironmentSpec Schema & AST Node
+0	0	Multi-language compile gate executor
+1	0	`whetstone_normalize_diagnostics` MCP tool
+2	0	Stable include/order regeneration model
+0	0	Queue lock conflict summary metrics
+3	0	Sprint 199 Integration Summary
+3	0	Sprint 204 Integration Summary
+3	0	Sprint 119 integration summary + regression
+0	0	Test failure classifier
+4	0	Setup verification report
+4	0	Runtime selection report artifact
+1	0	`whetstone_run_build_iteration` MCP tool
+4	0	Reliability and rollback report artifact
+3	0	Sprint 209 Integration Summary
+3	0	Sprint 170 Integration Summary
+3	0	Regression watchlist for promoted pairs
+4	0	Documentation pack for language support + migration playbooks
+2	0	AST-to-text regeneration policy schema
+2	0	Nullability + optionality canonical model bridge
+1	0	Corpus version diff and migration tools
+0	0	Artifact quality scoring and ranking engine
+0	0	Kotlin Generator
+2	0	Autonomous loop controller with “green or explicit blocked” contract
+1	0	`whetstone_validate_patch_scope` MCP tool
+3	0	Sprint 198 Integration Summary
+3	0	Sprint 84 integration summary + regression
+2	0	Source-vs-runtime drift comparator model
+1	0	`whetstone_generate_reprojection_plan` MCP tool
+2	0	Localization-ready artifact schema
+0	0	Release Readiness Gate Pack
+1	0	`whetstone_apply_best_practice_pack` MCP tool
+0	0	Go + C++ + Elisp Parser Deepening
+4	0	Self-Hosting Progress Report + Phase Integration
+2	0	`whetstone_ingest_legacy_to_ir` MCP tool
+3	0	Sprint 74 integration summary + regression
+0	0	Annotate Whetstone's Own Code
+0	0	C interop adapter deepening
+2	0	Refactor pattern schema and metadata model
+2	0	Python/TypeScript Constructive Edit Adapter
+3	0	Sprint 159 integration summary + regression
+0	0	Scope ambiguity diagnostics
+4	0	Standards publication bundle
+1	0	`whetstone_set_zero_trust_policy` MCP tool
+0	0	Benchmark runner and metrics export
+0	0	Template parameter validation engine
+0	0	Transpile Whetstone Modules
+1	0	Phase 13c Integration — Keyboard Shortcuts Help Panel
+0	0	Deterministic plan fingerprinting
+3	0	Numerical regression gate integration
+3	0	Sprint 147 integration summary + regression
+2	0	Pair-specific hint model interface
+2	0	Parallel partitioning semantics model
+2	0	Node identity preservation policy model
+3	0	Sprint 83 integration summary + regression
+4	0	`whetstone_generate_localized_migration_report` MCP tool
+2	0	Determinism and recovery qualification packet model
+0	0	Starvation prevention mechanism
+3	0	Sprint 166 Integration Summary
+0	0	C# Parser
+4	0	Guided transpilation exercise framework
+3	0	Sprint 42 Integration Summary
+4	0	Runtime compatibility risk report generator
+2	0	Interrupt and ISR semantics packet model
+1	0	Loop integration for auto-fix attempt
+4	0	Marketplace governance report artifact
+2	0	`whetstone_get_language_contract` MCP tool
+2	0	Triage packet bundle model
+1	0	`whetstone_probe_toolchain_provider` runtime implementation
+4	0	Reporting templates for C++ review teams
+1	0	`whetstone_query_transpile_graph` MCP tool
+3	0	Sprint 86 integration summary + regression
+0	0	Preconditions/postconditions validator for patterns
+0	0	File Tree Panel
+4	0	Runtime selection report artifact
+0	0	Taskitem Generator v2
+4	0	Toolchain readiness report artifact
+0	0	Promotion policy for tuned packs
+4	0	Family acceptance and divergence report
+0	0	Edge Case Cleanup
+0	0	Promotion gate for accepted proposals
+4	0	Release readiness scoreboard
+3	0	Sprint 180 Integration Summary
+0	0	Interop certification policy bindings
+2	0	Threat Model Integration
+1	0	`whetstone_analyze_data_pipeline_semantics` MCP tool
+4	0	Low-level acceptance report + waiver system
+3	0	Sprint 101 integration summary + regression
+2	0	MCP initialize — add `instructions` field
+4	0	Regeneration stability report artifact
+0	0	Sprint 25 Summary + Post-25 Readiness
+0	0	Trend and recurrence detector
+3	0	Sprint 155 integration summary + full-program regression
+1	0	Sprint 30 Integration + Summary
+2	0	Environment-Aware Pipeline Hooks
+2	0	Benchmark ingestion support for projection schema
+0	0	Per-section recall/precision metrics
+1	0	`whetstone_assemble_context` MCP tool
+1	0	`whetstone_debug_until_green` MCP orchestrator tool
+4	0	Per-target projection validation report
+0	0	Queue blocker gap classifier
+1	0	Plugin sandbox execution policy
+0	0	Pattern recurrence detector from graph signals
+2	0	Pair selection strategy engine (hot pairs, random sampling, full sweeps)
+0	0	Fix candidate generator
+3	0	Rebuild whetstone_mcp + smoke test
+4	0	Explainability report template pack
+3	0	`whetstone_probe_mcp_runtime_health` MCP tool
+4	0	Family-level acceptance report generator
+3	0	Sprint 100 integration summary + full-program regression
+3	0	Sprint 98 integration summary + regression
+0	0	WorkflowState.h — Project-Level Workflow Tracking
+3	0	Sprint 205 Integration Summary
+1	0	`whetstone_get_epoch_workstreams` MCP tool
+2	0	Prioritization scoring model (impact x frequency x tier)
+4	0	Lock metadata documentation packet
+4	0	Escalation report composer
+3	0	Sprint 160 integration summary + full-program regression
+2	0	Capability declaration struct generator
+4	0	Lock validation aggregate report fields
+3	0	Sprint 167 Integration Summary
+1	0	Gold benchmark curation pipeline
+1	0	RoutingEngine.h — Annotation-to-Dispatch Logic
+4	0	`whetstone_get_federated_verification_report` MCP tool
+2	0	Generic/monomorphization intent model
+1	0	`whetstone_estimate_porting_cost` MCP tool
+4	0	C++ constructive loop report artifact
+4	0	Epoch kickoff report artifact
+4	0	Cost Tracking + Reporting
+0	0	Long-range blocker taxonomy
+4	0	Campaign report bundle generator
+4	0	`whetstone_get_sync_identity_report` MCP tool
+3	0	Sprint 45 Integration Summary
+0	0	Template Class Declarations
+1	0	Route-to-deterministic policy hook
+3	0	Sprint 103 integration summary + regression
+1	0	Pipeline summary propagation of call plan
+4	0	Patch proposal review report artifact
+2	0	Profile-backed readiness and qualification packet model
+4	0	`whetstone_generate_localized_migration_report` MCP tool
+2	0	Type System Translation
+1	0	`whetstone_get_regeneration_trace_generic` MCP tool
+4	0	Benchmark scorecard generator per pair
+4	0	Runtime freshness report artifact
+2	0	Merge impact and provenance model
+2	0	New AST Nodes — Class/Interface/Generic
+4	0	Family acceptance and divergence report
+3	0	Sprint 130 integration summary + regression
+2	0	Perf benchmark harness contract
+2	0	Data lineage canonical model
+3	0	Sprint 109 integration summary + regression
+4	0	Security Report Generation
+3	0	Safety Audit via Annotations
+0	0	Candidate auto-selector
+4	0	Practice adoption and outcome report artifact
+1	0	Sanitizer gate integration (ASan/UBSan profile hooks)
+4	0	Enterprise rollout report template
+3	0	Convention Extractor + Validator (tests: unit, negative, regression)
+4	0	Incremental sync report artifact
+1	0	Auto-containment strategy hooks
+0	0	Multi-Language Project Orchestration
+1	0	Phase 12a Integration Tests
+2	0	Integrity attestation packet schema
+3	0	Sprint 107 integration summary + regression
+2	0	Plugin manifest and capability declaration model
+2	0	Session-level mode override model
+3	0	Sprint 200 Integration Summary
+2	0	`whetstone_run_constructive_ga_gate` MCP tool
+4	0	Team onboarding readiness report artifact
+1	0	Orchestrator RPC + MCP
+3	0	Sprint 67 integration summary + regression
+0	0	Job status subscriber panel
+1	0	`whetstone_verify_simulation_port` MCP tool
+2	0	Zero-trust policy model for execution surfaces
+1	0	`whetstone_get_slm_debug_readiness` MCP tool
+3	0	Sprint 102 integration summary + regression
+4	0	Cross-language constructive execution report artifact
+0	0	MySQL lowering/raising adapters
+0	0	Accessibility Baseline Pass
+1	0	`whetstone_get_recovery_advice` MCP tool
+3	0	Sprint 106 integration summary + regression
+3	0	Sprint 143 integration summary + regression
+3	0	Sprint 133 integration summary + regression
+4	0	Replay/transaction reliability report artifact
+2	0	`whetstone_start_recording` + `whetstone_get_metrics` MCP tools
+0	0	C++ Generator — Inheritance + Templates
+2	0	`whetstone_preview_text_ast_merge` MCP tool
+1	0	tools.json refresh — add all Sprint 32-35 tools
+3	0	Sprint 65 integration summary + regression
+0	0	Generator Implementations — Python/C++/Rust/Go
+0	0	Multi-file conflict detection during execution
+1	0	`whetstone_start_iteration_session` MCP tool
+3	0	Sprint 50 integration summary + regression
+1	0	`whetstone_plan_preventive_maintenance` MCP tool
+3	0	Sprint 63 integration summary + regression
+0	0	Pre-Apply Validation Gate
+4	0	Epoch renewal publication bundle
+3	0	Sprint 173 Integration Summary
+4	0	Guided Architect-to-Execution Demo Mode
+1	0	Semanno Sidecar Integration
+4	0	Resilience operations report artifact
+2	0	Contract conformance score model
+4	0	Semantic memory report artifact
+3	0	Sprint 134 integration summary + regression
+4	0	Security Report Generation
+3	0	Sprint 188 Integration Summary
+4	0	Requirements gap report
+2	0	C++ formatting/styling guard model
+0	0	Security Testing Skeleton Generation
+3	0	Sprint 82 integration summary + regression
+4	0	`whetstone_get_reference_conformance_report` MCP tool
+0	0	Multi-service rollback choreography policy
+2	0	Test output parser — gtest format
+2	0	Language operation support policy model (`edit`, `sync`, `regenerate`, `merge`, `build_test`)
+1	0	`whetstone_get_certification_status` MCP tool
+0	0	Deterministic threshold profile selection
+3	0	Phase 21b Integration + Sprint Summary
+2	0	Real replay suite wiring to runtime traces
+0	0	Panel Registration System
+2	0	LTS baseline criteria schema
+3	0	Mode transition event packet model
+4	0	Pattern efficacy report artifact
+2	0	Campaign queue policy model
+3	0	Sprint 149 integration summary + regression
+1	0	Lint executor integration
+0	0	Task Detail View
+0	0	Distributed path in FeedbackLoopOrchestrator
+0	0	Recommendation engine for applicable packs
+4	0	Localized report generation pipeline
+2	0	Agent Chat Panel model
+2	0	Binary/interface diff engine integration
+4	0	Constructive GA publication bundle
+1	0	`whetstone_run_improvement_trial` MCP tool
+2	0	Determinism mode packet (`strict`, `best-effort`)
+4	0	Debt burn-down report artifact
+4	0	`whetstone_get_pair_scorecard` MCP tool
+0	0	Symbol Scope Extractor
+2	0	MCP execution event schema
+1	0	API Boundary Preservation
+1	0	`whetstone_transpile_low_level_family` MCP tool
+2	0	Constructive loop orchestration schema
+1	0	MCP Server Configuration + Discovery
+4	0	Program constitution + handbook publication bundle
+1	0	Operation Selector API
+4	0	Self-Hosting Metrics Report
+1	0	Phase 13d Integration + Sprint 13 Summary
+4	0	Localized report generation pipeline
+4	0	SemannoFormat.h — Comment Format Standard
+3	0	Sprint 69 integration summary + regression
+4	0	Misclassification report export
+0	0	Deterministic data-sample replay runner
+0	0	Analytics migration dossier template
+0	0	Checkpoint and rollback node insertion
+0	0	Method completion planner
+4	0	Regeneration stability report artifact
+0	0	Dynamic strictness policy engine (`lenient/balanced/strict`)
diff --git a/specialists/data/generated/worker_type_eval.tsv b/specialists/data/generated/worker_type_eval.tsv
new file mode 100644
index 0000000..7dd5ef3
--- /dev/null
+++ b/specialists/data/generated/worker_type_eval.tsv
@@ -0,0 +1,52 @@
+0	0	Snapshot persistence (sidecar integration)
+1	0	`whetstone_preview_regeneration_diff_generic` MCP tool
+2	0	Phase 25a Integration
+3	0	Sprint 116 integration summary + regression
+0	0	Debug assist packet bundle model
+1	0	`whetstone_score_failure_triage` MCP tool
+3	0	Regression watchlist for promoted pairs
+2	0	Architecture Templates
+3	0	Phase 10d Integration Tests
+2	0	Phase 21a Integration
+2	0	Result Acceptance Protocol
+2	0	Guided transpilation exercise framework
+0	0	Patch proposal schema and provenance model
+2	0	`whetstone_publish_roadmap_epoch` MCP tool
+0	0	Compliance policy binding engine
+2	0	Workflow Session Protocol
+3	0	Sprint 181 Integration Summary
+2	0	Language-specific method body scaffolding for core patterns
+0	0	Sprint 27 Integration + Summary
+3	0	Phase 15b Integration — Full Protocol Test
+0	0	Long-range edit graph planner
+3	0	Sprint 120 integration summary + regression
+0	0	Federated governance report generator
+3	0	Sprint 66 integration summary + regression
+3	0	Sprint 204 Integration Summary
+2	0	Phase 27a Integration
+1	0	Review Gates
+3	0	Sprint 55 integration summary + regression
+0	0	LTS baseline criteria schema
+0	0	Replay comparison and divergence classification runtime model
+3	0	Sprint 101 integration summary + regression
+2	0	Retry/Escalation Protocol for Constraint Failures
+0	0	Compatibility policy profiles by interface criticality
+0	0	Language support tiers + gates
+2	0	Phase 13d Integration + Sprint 13 Summary
+1	0	Deterministic triage queue planner model
+1	0	Unmappable semantic blocklist + mandatory review gate
+2	0	Phase 23a Integration
+1	0	`whetstone_preview_text_ast_merge_generic` MCP tool
+3	0	Sprint 132 integration summary + regression
+2	0	Phase 27a Integration
+3	0	Sprint 71 integration summary + regression
+2	0	Phase 14b Integration
+0	0	Epoch-2 workstream schema and ownership map
+3	0	Sprint 141 integration summary + regression
+3	0	AST-native projection benchmark suite
+0	0	Certification evidence archive format v2
+2	0	Phase 29a Integration
+0	0	Tier promotion engine (beta->stable)
+1	0	Quarterly lock/review cadence policy
+1	0	Diagnostic normalization runtime implementation model
+2	0	Phase 22a Integration
diff --git a/specialists/data/generated/worker_type_projects_eval.tsv b/specialists/data/generated/worker_type_projects_eval.tsv
new file mode 100644
index 0000000..f254759
--- /dev/null
+++ b/specialists/data/generated/worker_type_projects_eval.tsv
@@ -0,0 +1,63 @@
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+0	0	Plan Build a streaming join worker correlating click and | Build a streaming join worker correlating click and conversion events with watermark logic.
+0	0	Define verification and readiness surface | Build a live notifications hub with subscription filters, fanout, and reconnect replay support.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+1	0	Define verification and readiness surface | Build review moderation with profanity checks, spam heuristics, and moderation queue endpoints.
+1	0	Define verification and readiness surface | Build a PII redaction pipeline masking emails, phones, and IDs with audit logging.
+1	0	Define verification and readiness surface | Build audit log service with immutable events, filtering, export, and retention policies.
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+0	0	Define verification and readiness surface | Build a secrets rotation scheduler with lease tracking, retry plans, and notifications.
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+0	0	Define verification and readiness surface | Build kanban API with boards, columns, cards, drag reorder, and activity history.
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+0	0	Plan Build an IoT telemetry gateway supporting MQTT ingestion | Build an IoT telemetry gateway supporting MQTT ingestion, validation, and stream forwarding.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build a CLI todo manager with add/list/done/delete commands and JSON file storage.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+1	0	Define verification and readiness surface | Build a ledger reconciliation service with strict double-entry invariants, idempotent replay, rollback checkpoints, and 
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build a whiteboard sync backend with operational ordering and conflict-safe patch application.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+0	0	Define verification and readiness surface | Build an image conversion CLI with resize, format conversion, and batch mode support.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Plan Split monolithic endpoint into read/write paths while preserving | Split monolithic endpoint into read/write paths while preserving latency and memory budgets and updating observability d
+1	0	Define verification and readiness surface | Build a PII redaction pipeline masking emails, phones, and IDs with audit logging.
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+3	0	Define verification and readiness surface | Build portfolio tracker with holdings, transactions, PnL snapshots, and benchmark comparison.
+0	0	Define verification and readiness surface | Build checkout backend coordinating address, shipping options, totals, and payment intent creation.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+0	0	Define verification and readiness surface | Build chat bot backend with slash commands, permission checks, and scheduled digests.
+1	0	Define verification and readiness surface | Build review moderation with profanity checks, spam heuristics, and moderation queue endpoints.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build review moderation with profanity checks, spam heuristics, and moderation queue endpoints.
+1	0	Define verification and readiness surface | Build review moderation with profanity checks, spam heuristics, and moderation queue endpoints.
+1	0	Define verification and readiness surface | Build fraud rules engine with rule evaluation graph, score thresholds, and audit trails.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+0	0	Define verification and readiness surface | Build a contacts service with import/export, deduplication, and fuzzy name search.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+0	0	Define verification and readiness surface | Build realtime polling backend with vote dedupe, room join, and result broadcast.
+1	0	Define verification and readiness surface | Build fraud rules engine with rule evaluation graph, score thresholds, and audit trails.
+0	0	Plan Upgrade API contract from v1 to v2 backend | Upgrade API contract from v1 to v2: backend handlers, OpenAPI schema, generated client SDK, frontend integration, and co
+1	0	Define verification and readiness surface | Build fraud rules engine with rule evaluation graph, score thresholds, and audit trails.
+0	0	Plan Build a streaming detector with keyed state windows | Build a streaming detector with keyed state windows, out-of-order event handling, watermark policy, and deterministic re
+0	0	Define verification and readiness surface | Build a release notes generator from commits, PR labels, and issue references.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+0	0	Define verification and readiness surface | Build a change-data-capture processor that applies upserts/deletes and checkpoints offsets.
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+0	0	Define verification and readiness surface | Build a model registry API supporting stages, approvals, rollback, and signed artifact links.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+1	0	Define verification and readiness surface | Build review moderation with profanity checks, spam heuristics, and moderation queue endpoints.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+1	0	Define verification and readiness surface | Build a ledger reconciliation service with strict double-entry invariants, idempotent replay, rollback checkpoints, and 
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
diff --git a/specialists/data/generated/worker_type_projects_train.tsv b/specialists/data/generated/worker_type_projects_train.tsv
new file mode 100644
index 0000000..eab20e0
--- /dev/null
+++ b/specialists/data/generated/worker_type_projects_train.tsv
@@ -0,0 +1,361 @@
+1	0	Define verification and readiness surface | Build review moderation with profanity checks, spam heuristics, and moderation queue endpoints.
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+0	0	Define verification and readiness surface | Build job queue service with delayed jobs, retries, visibility timeout, and metrics.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+1	0	Define verification and readiness surface | Build localization service with key bundles, fallback locales, and missing-key diagnostics.
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+1	0	Define verification and readiness surface | Build review moderation with profanity checks, spam heuristics, and moderation queue endpoints.
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+0	0	Define verification and readiness surface | Build a CLI password vault with encrypted local storage and master-key unlock.
+0	0	Define verification and readiness surface | Build order management with status machine, cancellation rules, and shipment events.
+1	0	Define verification and readiness surface | Build localization service with key bundles, fallback locales, and missing-key diagnostics.
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+1	0	Define verification and readiness surface | Build audit log service with immutable events, filtering, export, and retention policies.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+1	0	Define verification and readiness surface | Build a PII redaction pipeline masking emails, phones, and IDs with audit logging.
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+0	0	Define verification and readiness surface | Build a registry mirror service with digest verification, cache pruning, and pull-through.
+1	0	Define verification and readiness surface | Build fraud rules engine with rule evaluation graph, score thresholds, and audit trails.
+0	0	Define verification and readiness surface | Build a websocket chat server with rooms, presence, typing indicators, and message history.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+0	0	Define verification and readiness surface | Build subscription billing engine with proration, trials, coupons, and failed-payment retries.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+1	0	Define verification and readiness surface | Build review moderation with profanity checks, spam heuristics, and moderation queue endpoints.
+1	0	Define verification and readiness surface | Build fraud rules engine with rule evaluation graph, score thresholds, and audit trails.
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+1	0	Define verification and readiness surface | Build a PII redaction pipeline masking emails, phones, and IDs with audit logging.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build localization service with key bundles, fallback locales, and missing-key diagnostics.
+0	0	Define verification and readiness surface | Build invoice generator with tax calculation, due schedules, and PDF export metadata.
+0	0	Define verification and readiness surface | Build a webhook dispatcher with retries, signature verification, dead-letter queue, and backoff.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Plan Build a ticker stream service with topic subscriptions | Build a ticker stream service with topic subscriptions and throttled update delivery.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build a ledger reconciliation service with strict double-entry invariants, idempotent replay, rollback checkpoints, and 
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build a CLI log tailer with multi-file follow, regex filters, and output highlighting.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build a CSV analyzer CLI with schema inference, summary stats, and filter expressions.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build returns API with RMA creation, restock policy, and refund eligibility checks.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build a service monitor that runs health checks, tracks SLO windows, and pages on breaches.
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build expense backend with receipt OCR placeholders, category budgets, and recurring expenses.
+0	0	Define verification and readiness surface | Build a harness that transpiles class-heavy source modules across multiple target languages and validates class/method p
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+1	0	Define verification and readiness surface | Build audit log service with immutable events, filtering, export, and retention policies.
+0	0	Define verification and readiness surface | Build matchmaking service with queue buckets, skill-based pairing, and timeout fallback.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+0	0	Plan Build an incremental snapshot builder that compacts event | Build an incremental snapshot builder that compacts event logs into query-optimized state.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build localization service with key bundles, fallback locales, and missing-key diagnostics.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+0	0	Define verification and readiness surface | Build a REST Todo API with CRUD, pagination, filtering by status, and SQLite persistence.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+0	0	Plan Add normalized address schema migrate legacy rows with | Add normalized address schema, migrate legacy rows with backfill job, dual-read/write transition, and rollback script.
+0	0	Define verification and readiness surface | Build a log enrichment batch job joining geo data and user agents with partitioned output.
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+0	0	Define verification and readiness surface | Build a URL shortener API with custom aliases, redirect tracking, and expiration support.
+0	0	Define verification and readiness surface | Build meeting notes API with templates, attendee action items, and follow-up reminders.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build localization service with key bundles, fallback locales, and missing-key diagnostics.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build localization service with key bundles, fallback locales, and missing-key diagnostics.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build cart service with item merge, discount application, and inventory reservation integration.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build session store service with token revocation, idle timeout, and concurrent login policy.
+0	0	Define verification and readiness surface | Build a secret rotation controller with staged rollout, canary verification, rollback triggers, and explicit blast-radiu
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build fraud rules engine with rule evaluation graph, score thresholds, and audit trails.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+1	0	Define verification and readiness surface | Build audit log service with immutable events, filtering, export, and retention policies.
+1	0	Define verification and readiness surface | Build audit log service with immutable events, filtering, export, and retention policies.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+0	0	Plan shared sync core plus Android/Desktop/Web projections with platform-specific | Implement shared sync core plus Android/Desktop/Web projections with platform-specific persistence and transport adapter
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+1	0	Define verification and readiness surface | Build a PII redaction pipeline masking emails, phones, and IDs with audit logging.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build fraud rules engine with rule evaluation graph, score thresholds, and audit trails.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+1	0	Define verification and readiness surface | Build a ledger reconciliation service with strict double-entry invariants, idempotent replay, rollback checkpoints, and 
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+0	0	Define verification and readiness surface | Build LLM guardrail service for policy checks, PII filtering, and response classification.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build a multi-tenant limiter with token bucket + sliding window hybrid policy, burst fairness guarantees, and lock-free 
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build a PII redaction pipeline masking emails, phones, and IDs with audit logging.
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build a Kubernetes manifest linter enforcing probes, limits, and namespace policy.
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+0	0	Define verification and readiness surface | Build an API for file metadata indexing, search by attributes, and soft delete recovery.
+1	0	Define verification and readiness surface | Build fraud rules engine with rule evaluation graph, score thresholds, and audit trails.
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+0	0	Define verification and readiness surface | Build experiment tracking backend storing params, metrics, artifacts, and compare queries.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build user profile service with avatars, privacy settings, and profile completeness scoring.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build localization service with key bundles, fallback locales, and missing-key diagnostics.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+0	0	Define verification and readiness surface | Build batch inference pipeline reading partitions, loading model, and writing scored outputs.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build a PII redaction pipeline masking emails, phones, and IDs with audit logging.
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+0	0	Define verification and readiness surface | Build a JWT auth service with register, login, refresh token rotation, and password hashing.
+0	0	Define verification and readiness surface | Build notification preferences API with channel rules, quiet hours, and topic subscriptions.
+0	0	Plan Build a CI orchestrator service triggering jobs tracking | Build a CI orchestrator service triggering jobs, tracking states, and emitting summaries.
+1	0	Define verification and readiness surface | Build fraud rules engine with rule evaluation graph, score thresholds, and audit trails.
+1	0	Define verification and readiness surface | Build a ledger reconciliation service with strict double-entry invariants, idempotent replay, rollback checkpoints, and 
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+0	0	Define verification and readiness surface | Build a harness that transpiles class-heavy source modules across multiple target languages and validates class/method p
+0	0	Plan telemetry agent for linux arduino targets using shared | Implement telemetry agent for linux + arduino targets using shared serialization core and target-specific runtime wrappe
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+1	0	Define verification and readiness surface | Build a PII redaction pipeline masking emails, phones, and IDs with audit logging.
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build prompt evaluation runner with rubric scoring, aggregate metrics, and reproducible reports.
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build a feature store service with online/offline reads and point-in-time correctness.
+1	0	Define verification and readiness surface | Build a ledger reconciliation service with strict double-entry invariants, idempotent replay, rollback checkpoints, and 
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build wiki backend with page versions, backlinks, and markdown rendering pipeline hooks.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build drift detection worker comparing live feature distributions against reference baselines.
+0	0	Define verification and readiness surface | Build form builder backend with schema validation, submissions, and analytics counters.
+0	0	Define verification and readiness surface | Build email campaign manager with lists, segmentation, scheduling, and bounce tracking.
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+0	0	Define verification and readiness surface | Build an access gateway that evaluates hierarchical allow/deny policies with shadow rules, policy versioning, and explai
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build a presence tracker with heartbeat timeout, status transitions, and shard-safe updates.
+0	0	Define verification and readiness surface | Build payments API with idempotency keys, ledger-safe transfers, and webhook callbacks.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Plan Build a task runner CLI that reads YAML | Build a task runner CLI that reads YAML tasks and executes dependency-aware command graphs.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build configuration service with versioned configs, staged rollout, and validation checks.
+1	0	Define verification and readiness surface | Build a ledger reconciliation service with strict double-entry invariants, idempotent replay, rollback checkpoints, and 
+0	0	Define verification and readiness surface | Build warehouse picking service with wave assignment, pick confirmation, and shortage handling.
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build shipping label worker integrating carrier rates, label generation, and tracking updates.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build a metrics aggregation worker with tumbling windows, late data handling, and rollups.
+1	0	Define verification and readiness surface | Build a PII redaction pipeline masking emails, phones, and IDs with audit logging.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build audit log service with immutable events, filtering, export, and retention policies.
+1	0	Define verification and readiness surface | Build fraud rules engine with rule evaluation graph, score thresholds, and audit trails.
+1	0	Define verification and readiness surface | Build a ledger reconciliation service with strict double-entry invariants, idempotent replay, rollback checkpoints, and 
+1	0	Define verification and readiness surface | Build localization service with key bundles, fallback locales, and missing-key diagnostics.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build backup retention manager with policy tiers, legal hold support, and deletion safety checks.
+0	0	Plan Build reconciliation service matching ledger entries to bank | Build reconciliation service matching ledger entries to bank statements with mismatch reporting.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build an ETL pipeline that ingests orders CSV, validates schema, and loads curated tables.
+0	0	Define verification and readiness surface | Build search suggest API with prefix index, typo tolerance, and popularity boosting.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+1	0	Define verification and readiness surface | Build a ledger reconciliation service with strict double-entry invariants, idempotent replay, rollback checkpoints, and 
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build audit log service with immutable events, filtering, export, and retention policies.
+1	0	Define verification and readiness surface | Build localization service with key bundles, fallback locales, and missing-key diagnostics.
+1	0	Define verification and readiness surface | Build audit log service with immutable events, filtering, export, and retention policies.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build a PII redaction pipeline masking emails, phones, and IDs with audit logging.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+0	0	Define verification and readiness surface | Build a feature flag API with environments, targeting rules, and rollout percentages.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+0	0	Plan Replace search ranking pipeline behind feature flag with | Replace search ranking pipeline behind feature flag with staged rollout, telemetry checks, and automatic abort trigger.
+0	0	Define verification and readiness surface | Build a training scheduler with queue priorities, resource quotas, and retry policies.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+0	0	Define verification and readiness surface | Build a gateway that validates incoming payloads against versioned contracts, performs semantic constraint checks, and e
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build a backup CLI with include/exclude rules, incremental snapshots, and restore command.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+0	0	Define verification and readiness surface | Build pricing engine with currency handling, promotions, and time-bound campaign rules.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build review moderation with profanity checks, spam heuristics, and moderation queue endpoints.
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+0	0	Define verification and readiness surface | Build an event deduplicator using idempotency keys, time windows, and replay-safe writes.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+0	0	Define verification and readiness surface | Build dataset version management with lineage metadata, immutable snapshots, and diff operations.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+0	0	Define verification and readiness surface | Build tax report backend aggregating gains/losses and generating filing-year summaries.
+0	0	Define verification and readiness surface | Build distributed rate limiter with fixed/sliding window modes and policy APIs.
+0	0	Define verification and readiness surface | Build telemetry collector with trace/span ingestion, sampling, and export adapters.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Define verification and readiness surface | Build a CRDT-based collaborative notes backend with causal metadata, merge convergence proofs, tombstone compaction, and
+0	0	Define verification and readiness surface | Build a loader from object storage to warehouse with schema evolution and retry checkpoints.
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+0	0	Define verification and readiness surface | Build access control service with RBAC roles, permissions, and resource-scoped grants.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+3	0	Define verification and readiness surface | Refactor billing pipeline to introduce idempotency keys and rollback checkpoints across ledger service, queue worker, re
+0	0	Define verification and readiness surface | Build a CDC replication worker that handles schema evolution, backfill merges, checkpoint consistency, and deterministic
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build review moderation with profanity checks, spam heuristics, and moderation queue endpoints.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+1	0	Define verification and readiness surface | Build audit log service with immutable events, filtering, export, and retention policies.
+1	0	Define verification and readiness surface | Build localization service with key bundles, fallback locales, and missing-key diagnostics.
+0	0	Plan Build drift detection comparing desired and observed infra | Build drift detection comparing desired and observed infra states with actionable diff reports.
+1	0	Define verification and readiness surface | Build a ledger reconciliation service with strict double-entry invariants, idempotent replay, rollback checkpoints, and 
+0	0	Define verification and readiness surface | Build product catalog API with variants, facets, search indexing hooks, and soft delete.
+0	0	Define verification and readiness surface | Build a realtime log viewer backend with tail sessions, regex filters, and backpressure.
+1	0	Define verification and readiness surface | Build audit log service with immutable events, filtering, export, and retention policies.
+1	0	Define verification and readiness surface | Build localization service with key bundles, fallback locales, and missing-key diagnostics.
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build a realtime cursor service broadcasting positions with rate limiting and stale cleanup.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+0	0	Define verification and readiness surface | Build a text diff CLI with unified output, word-level mode, and ignore-whitespace option.
+0	0	Define verification and readiness surface | Build calendar backend with events, recurrence rules, reminders, and timezone normalization.
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build inventory endpoints with stock reservation, release, and low-stock alerts.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+0	0	Define verification and readiness surface | Build an inference API with model loading, request validation, and batch prediction endpoint.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+0	0	Define verification and readiness surface | Build recommendation API combining collaborative and rule-based fallback candidate generation.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build an order saga orchestrator with compensating actions, timeout escalation, exactly-once message handling, and deter
+2	0	Plan Introduce deny-by-default RBAC policy and propagate through backend | Introduce deny-by-default RBAC policy and propagate through backend auth middleware, endpoint guards, frontend route gua
+2	0	Plan Build a generator that scaffolds Terraform modules with | Build a generator that scaffolds Terraform modules with inputs, outputs, and policy checks.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build a .env manager CLI with profile switch, validation, and secure redaction in output.
+1	0	Define verification and readiness surface | Build fraud rules engine with rule evaluation graph, score thresholds, and audit trails.
+0	0	Define verification and readiness surface | Build a CMS backend with posts, tags, draft/publish workflow, and role-based access control.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+1	0	Define verification and readiness surface | Build review moderation with profanity checks, spam heuristics, and moderation queue endpoints.
+0	0	Define verification and readiness surface | Build a log shipping agent with offset persistence, multiline parsing, and retries.
+0	0	Define verification and readiness surface | Build a data quality checker with null, range, uniqueness, and referential integrity rules.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+0	0	Define verification and readiness surface | Build payout scheduler with settlement windows, minimum thresholds, and retry-safe processing.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
+2	0	Align docs with bootstrap architecture | Build KYC workflow with document checks, status transitions, and reviewer assignment queues.
+1	0	Define verification and readiness surface | Build a distributed scheduler with per-resource mutex locks, lease renewal, deadlock avoidance ordering, and determinist
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+1	0	Define verification and readiness surface | Build an incremental build executor over a DAG with content-hash invalidation, topological scheduling, cycle diagnostics
+2	0	Align docs with bootstrap architecture | Build document search service with indexing, permissions filtering, and ranking features.
+3	0	Define verification and readiness surface | Build a CRDT-based collaborative notes backend with causal metadata, merge convergence proofs, tombstone compaction, and
+0	0	Define verification and readiness surface | Build a notes sync API with optimistic concurrency, conflict detection, and change feed endpoints.
+0	0	Define verification and readiness surface | Build OCR queue processor with job retries, dead-letter handling, and extraction status API.
+0	0	Define verification and readiness surface | Build shared core and projections but omit projection contract fields intentionally.
+1	0	Define verification and readiness surface | Build a PII redaction pipeline masking emails, phones, and IDs with audit logging.
+3	0	Define verification and readiness surface | All core features implemented Tests passing
diff --git a/specialists/data/generated/worker_type_train.tsv b/specialists/data/generated/worker_type_train.tsv
new file mode 100644
index 0000000..c91dc0b
--- /dev/null
+++ b/specialists/data/generated/worker_type_train.tsv
@@ -0,0 +1,300 @@
+1	0	`whetstone_get_sync_diagnostics` MCP tool
+3	0	Regression fixtures for under-constrained vs capability-limited cases
+3	0	Sprint 83 integration summary + regression
+1	0	Audit trail and inspection
+3	0	Sprint 99 integration summary + regression
+3	0	Security Testing Skeleton Generation
+2	0	Phase 25b Integration
+3	0	Sprint 115 integration summary + regression
+0	0	Feedback Loop — Rejection Re-Routing
+2	0	Long-horizon roadmap planning engine v2
+2	0	Retry/Escalation Protocol for Constraint Failures
+0	0	`BuildErrorRecord` schema
+2	0	Phase 26a Integration
+3	0	Sprint 179 Integration Summary
+0	0	Policy-compliance auto-checkpoint integration
+2	0	Refactor grammar generator to use normalized schema IR
+3	0	Intake-to-Queue Simulation Harness
+3	0	Sprint 50 integration summary + regression
+1	0	Human review queue generator from ambiguities
+2	0	Scaffold File Generation
+3	0	Sprint 85 integration summary + regression
+0	0	`whetstone_get_swarm_maintenance_status` MCP tool
+3	0	Sprint 157 integration summary + regression
+3	0	Full historical regression + post-release checklist
+1	0	`whetstone_normalize_diagnostics` runtime implementation
+3	0	End-to-end — Python PriorityQueue → C++, Rust, Go
+2	0	Phase 12d Integration + Sprint 12 Summary
+1	0	False-green diagnostic class
+3	0	Sprint 86 integration summary + regression
+3	0	Sprint 64 integration summary + regression
+3	0	Sprint 142 integration summary + regression
+1	0	Priority-focused review queue shaping engine
+1	0	Efficiency diagnostics and trend report
+0	0	Context slice canonicalization for handoffs
+0	0	Context Bundle Format
+3	0	Reference implementation harness for IR/evidence specs
+1	0	Constraint completeness diagnostics
+1	0	Porting review board model
+0	0	Maintenance demand estimator
+0	0	`whetstone_get_program_kpis` MCP tool
+1	0	Tool-surface diagnostics report artifact
+3	0	Rust runner adapter for harness
+2	0	`whetstone_verify_architecture_consistency` MCP tool
+1	0	Security scan orchestrator for generated targets
+1	0	Strictness audit utility and baseline report
+0	0	Sprint pipeline summary enrichment
+0	0	C Parser — Functions, Structs, Enums
+1	0	Call Stack and Frame Inspector
+3	0	Sprint 74 integration summary + regression
+2	0	Phase 25b Integration
+0	0	Binary/interface diff engine integration
+3	0	Sprint 125 integration summary + regression
+1	0	Reviewer quorum policy model for critical approvals
+1	0	Conflict preview and impact estimator model
+3	0	Test failure classifier
+3	0	Benchmark ingestion support for projection schema
+0	0	Artifact quality scoring and ranking engine
+3	0	Sprint 47 integration summary + regression
+0	0	`whetstone_assemble_context` MCP tool
+0	0	Legacy modernization dossier generator
+0	0	Step Engine
+1	0	Agent role contract schema (`analyzer`, `patcher`, `verifier`, `reviewer`)
+0	0	`whetstone_trigger_porting_incident_drill` MCP tool
+1	0	Lint diagnostics normalization
+1	0	AST mutation preview
+2	0	Burn-down planning optimizer
+1	0	KPI delta analysis engine for block review
+2	0	Long-horizon roadmap planning engine v2
+0	0	`whetstone_rollback_constructive_transaction` MCP tool
+2	0	Deterministic blocker taxonomy
+1	0	Docking Reliability Audit + Deterministic Layout Rules
+0	0	`whetstone_run_build_iteration` MCP tool
+1	0	Divergence triage model for interop failures
+2	0	Phase 36a Integration
+0	0	Language capability conformance report artifact
+3	0	`whetstone_run_pair_benchmark` MCP tool
+1	0	`whetstone_optimize_review_queue` MCP tool
+3	0	Sprint 205 Integration Summary
+1	0	Loop summary contract diagnostics
+2	0	Phase 13a Integration — Layout Persistence
+3	0	Sprint 189 Integration Summary
+2	0	Continuity drill framework and scoring
+2	0	Deterministic blocker taxonomy extension
+3	0	Cross-team benchmark comparator
+0	0	VB.NET lowering/raising adapters
+1	0	Review-Gate Policy Refinement
+2	0	Phase 14d Integration + Sprint 14 Summary
+0	0	Multi-language compile gate executor
+1	0	Dependency and supply-chain audit packet
+1	0	Reviewer-focused semantic diff summarizer
+1	0	Architect Review Surface for Intake
+2	0	Phase 24a Integration
+2	0	Long-range blocker taxonomy
+2	0	Result Acceptance Protocol
+1	0	Patch proposal review report artifact
+1	0	Parity diagnostics export
+3	0	Debug benchmark harness for fixture suites
+1	0	Triage packet bundle model
+1	0	Incremental sync diagnostics model
+2	0	Swarm maintenance job decomposition model
+0	0	Epoch kickoff report artifact
+3	0	Sprint 96 integration summary + regression
+2	0	Phase 32a Integration
+2	0	Self-Hosting Progress Report + Phase Integration
+2	0	Phase 13a Integration — Layout Persistence
+0	0	Sprint 28 Integration + Summary
+2	0	Migration Plan Generator
+2	0	Phase 20a Integration
+3	0	End-to-end replay + rollback scenario suite
+3	0	C++ runner adapter for harness
+1	0	Safety Audit via Annotations
+2	0	Phase 33a Integration
+3	0	Sprint 79 integration summary + regression
+0	0	Lisp lowering/raising adapters
+0	0	`whetstone_run_porting_gates` MCP tool
+1	0	Explainability compression model for rapid review
+2	0	Phase 32a Integration
+3	0	Sprint 51 integration summary + regression
+3	0	Sprint 175 Integration Summary
+3	0	Sprint 146 integration summary + regression
+2	0	Deterministic blocker taxonomy extension
+3	0	Sprint 139 integration summary + regression
+3	0	Sprint 90 integration summary + full-program regression
+3	0	Benchmark publication bundle
+2	0	Transpilation failure taxonomy schema
+3	0	Sprint 138 integration summary + regression
+1	0	Architect Review Interface
+1	0	Real adapter diagnostics in status tool
+0	0	Kotlin Parser
+1	0	Remediation router from gate diagnostics to tool actions
+0	0	Family promotion checks (experimental->beta)
+2	0	Phase 16a Integration
+3	0	Sprint 73 integration summary + regression
+2	0	Phase 13d Integration + Sprint 13 Summary
+0	0	AnnotationValidator — Subject 5-8 Rules
+0	0	Deterministic replay pack composer
+1	0	Assurance review dossier template
+3	0	Auto-tuning proposal generator from benchmark deltas
+1	0	`whetstone_get_distributed_triage_queue` MCP tool
+1	0	KPI anomaly triage model
+2	0	Phase 36a Integration
+2	0	Module Decomposition Engine
+1	0	Scope ambiguity diagnostics
+2	0	Phase 38a Integration
+1	0	Governance and audit report bundle
+2	0	Phase 13b Integration — Visual Consistency
+3	0	CI integration in build/test flow
+0	0	Zero-trust policy model for execution surfaces
+1	0	Patch execution audit bundle
+2	0	Phase 30a Integration
+1	0	Drift audit report artifact
+0	0	`whetstone_marketplace_search` MCP tool
+3	0	Cross-language class emission parity tests
+3	0	Test output parser — gtest format
+0	0	`whetstone_get_debug_campaign_status` MCP tool
+1	0	Dynamic family acceptance report + review queue wiring
+3	0	Sprint 191 Integration Summary
+2	0	Generator Updates + Phase 11c Integration
+1	0	Review effort estimator from ambiguity/risk packets
+3	0	Sprint 152 integration summary + regression
+2	0	Candidate generation engine from failure taxonomy
+0	0	Ambiguity packet model (`assumed`, `unknown`, `conflict`)
+3	0	Sprint 198 Integration Summary
+1	0	Memory Inspector UI Model
+0	0	Argument/result canonicalizer model
+3	0	Sprint 89 integration summary + regression
+1	0	Divergence classification and triage model
+0	0	`whetstone_run_constructive_loop` MCP tool
+1	0	Dependency Audit Annotations
+3	0	Benchmark runner and metrics export
+3	0	Sprint 62 integration summary + regression
+1	0	Review Comparison View
+1	0	Review SLA policy integration
+0	0	Failure injection and safety validation model
+1	0	Intent drift diagnostics export
+3	0	Sprint 186 Integration Summary
+0	0	Auto Type Deduction
+2	0	Scope and Milestone Decomposer
+2	0	Phase 25a Integration
+1	0	Constraint Violation Diagnostics
+2	0	Refactor pattern schema and metadata model
+1	0	`whetstone_normalize_diagnostics` MCP tool
+0	0	Replay/transaction reliability report artifact
+1	0	Reviewer effectiveness report artifact
+1	0	Post-incident continuity audit report model
+1	0	Entropy scanner — editor-side
+2	0	Phase 13c Integration — Keyboard Shortcuts Help Panel
+3	0	Test gate runner and minimal harness injection
+2	0	Backend framework adapter hooks
+0	0	Enhanced evidence requirements engine
+1	0	`whetstone_preview_regenerated_diff` MCP tool
+1	0	Hint safety guardrails and audit fields
+2	0	Refactor intent annotation model
+3	0	Sprint 137 integration summary + regression
+1	0	Reporting templates for C++ review teams
+0	0	Generic/monomorphization intent model
+0	0	`whetstone_compare_mcp_replays` MCP tool
+0	0	Memory Model Translation
+0	0	Schema evolution compatibility checker
+0	0	Closeout consistency gate
+0	0	Real-time constraint model (`deadline`, `jitter`, `period`)
+2	0	Workflow Session Protocol
+0	0	Include/import ordering normalization model
+2	0	Deprecation/succession lifecycle framework
+0	0	Trace Timeline Model
+1	0	Decision consistency analyzer across reviewers
+1	0	Profile compliance auditor integration
+0	0	`whetstone_plan_rollout_stage` MCP tool
+0	0	Top-level strictness gate
+0	0	Strict queue gate for execution contracts
+3	0	Sprint 58 integration summary + regression
+3	0	Sprint 48 integration summary + regression
+0	0	`whetstone_detect_text_ast_conflicts_generic` MCP tool
+2	0	Phase 19a Integration
+1	0	`whetstone_get_review_load_status` MCP tool
+3	0	Phase 19b Integration + Sprint Summary
+3	0	Sprint 200 Integration Summary
+0	0	Async/await state intent lowering
+1	0	Failure triage score model
+0	0	Plan fingerprinting and replay checks
+1	0	MCP replay diagnostics artifact
+0	0	Plugin compatibility resolver (core/API/version)
+1	0	Diagnostic normalization canonical model
+3	0	Test-to-source locator
+0	0	Debug action budget model
+2	0	Scaffold File Generation
+1	0	Canonical diagnostic normalization from provider/LSP output
+2	0	Backend framework adapter hooks
+3	0	Sprint 162 Integration Summary
+2	0	Phase 39a Integration
+0	0	ADT + pattern matching canonical lowering
+2	0	Verification gate integration for architecture artifacts
+0	0	Modernization RPC + MCP
+2	0	Phase 19a Integration
+3	0	Convention Extractor + Validator (tests: unit, negative, regression)
+1	0	`whetstone_preview_text_ast_merge` MCP tool
+0	0	ABI gate bindings to tier promotions
+2	0	Architecture claim linkage engine
+0	0	Permission conformance report artifact
+2	0	Phase 20a Integration
+2	0	Governance model refresh framework
+1	0	Sync diagnostics normalization model
+3	0	Sprint 63 integration summary + regression
+0	0	`whetstone_suggest_build_fix` MCP tool
+3	0	A/B + benchmark validation artifact update
+0	0	`whetstone_plan_tool_migrations` MCP tool
+3	0	Full regression + architecture gate
+0	0	C interop adapter deepening
+3	0	Phase 10e Integration Tests
+2	0	Phase 29a Integration
+2	0	Scope and Milestone Decomposer
+0	0	Multi-check gate orchestrator model
+1	0	Lint executor integration
+3	0	Sprint 118 integration summary + regression
+1	0	Dropped-content diagnostics packet
+0	0	Execution trace and recovery packet export
+2	0	Phase 18a Integration
+3	0	Sprint 45 Integration Summary
+2	0	Deterministic blocker taxonomy
+2	0	Refactor pattern schema and metadata model
+1	0	Reviewer decision ledger integration
+2	0	Phase 28a Integration
+2	0	Phase 37a Integration
+3	0	Sprint 114 integration summary + regression
+2	0	`whetstone_publish_roadmap_epoch` MCP tool
+1	0	`whetstone_review_porting_decision` MCP tool
+2	0	Language-specific method body scaffolding for core patterns
+1	0	Audit trail completeness checker
+1	0	C++ include fixer from gate diagnostics
+3	0	End-to-end capability inventory and gap scan
+1	0	Static registry scanner for `register*Tools` blocks
+2	0	Phase 31a Integration
+1	0	`whetstone_rank_failure_triage_actions` MCP tool
+3	0	Timing regression runner integration
+2	0	Phase 40a Integration
+1	0	Reviewer load and fatigue model
+2	0	Swarm maintenance job decomposition model
+2	0	Target projection blocker taxonomy
+0	0	Annotate Whetstone's Own Code
+2	0	Phase 15c Integration + Sprint 15 Summary
+0	0	Epoch block health re-baseline model
+3	0	Sprint 159 integration summary + regression
+1	0	Human Review Interface via MCP
+0	0	Transpile Whetstone Modules
+0	0	Capability matrix model
+2	0	Phase 14d Integration + Sprint 14 Summary
+3	0	Prior-Step Context Injector (tests: unit, integration)
+0	0	Unsafe-practice detection and quarantine
+3	0	Sprint 127 integration summary + regression
+1	0	`TaskitemQualityAuditor` — batch report
+3	0	Systemic regression detector
+0	0	Debug campaign spec model
+3	0	`whetstone_get_execution_attestation` MCP tool
+3	0	Sprint 72 integration summary + regression
+2	0	Self-Hosting Progress Report + Phase Integration
+0	0	C# Parser
+1	0	Ambiguity triage UI model and API
+0	0	F# Parser
+0	0	Result-to-ObservationRecord mapper
+0	0	Self-service escalation handoff hooks
diff --git a/specialists/docs/ARCHITECTURE.md b/specialists/docs/ARCHITECTURE.md
new file mode 100644
index 0000000..b94821c
--- /dev/null
+++ b/specialists/docs/ARCHITECTURE.md
@@ -0,0 +1,197 @@
+# Whetstone Specialist Fleet — Architecture
+
+**Date:** 2026-03-28
+**Status:** Phase 1 — first specialist training
+
+---
+
+## The Core Idea
+
+The current Whetstone pipeline makes every bounded decision with one large LLM call.
+Each call is ~seconds, sequential, and expensive. Many of these decisions are
+bounded-output classifications — the exact problem that WhetstoneAI Fabricate's
+~213K-parameter (~800KB) specialists solve in microseconds.
+
+**Hypothesis:** Replace bounded-output LLM decisions in the Whetstone taskitem
+pipeline with a fleet of micro-specialists, each ~800KB, running in parallel.
+
+### Why It Works
+
+The WhetstoneAI RSA experiment demonstrated:
+- 96.3% in-distribution accuracy on domain classification
+- 76.7% OOD accuracy after replay hardening (up from 44.8%)
+- Inference time: microseconds per query on GPU, 1–5ms on CPU (numpy)
+- Model size: ~213K params, ~800KB per checkpoint
+- RTX 3060 (12GB VRAM): can hold ~14,000 specialists resident simultaneously
+
+The Whetstone taskitem pipeline already has:
+- Bounded enum decision points (verification type, prerequisite ops, worker type...)
+- Recorded task completions (18,519 pipeline runs)
+- Concurrency/dependency architecture (lock-based pool with parallel workers)
+
+This means the specialists can map 1:1 onto the pipeline's decision points and
+run in parallel with the existing task dependency graph.
+
+---
+
+## Decision Points Amenable to Specialists
+
+### Tier 1: Text-Input Classifiers (primary target)
+
+These decisions receive a task description and output a bounded choice.
+They currently run inside the LLM and are the highest-value targets.
+
+| Specialist | Input | Output | Classes |
+|-----------|-------|--------|---------|
+| `verification_type` | task step description | verification type | unit / integration / schema / smoke / docs |
+| `prereq_op_selector` | task description | prerequisite op set | validate-intake / resolve-dependencies / architect-review / manual-approval |
+| `worker_type` | task description | worker role | implementer / reviewer / architect / qa |
+| `gap_class` | taskitem issues list | gap classification | under_constrained / env_gap / planning_gap / cross_project / generator_cap |
+
+### Tier 2: Structural Specialists (formula approximators)
+
+These learn the deterministic scoring formulas — useful for ultra-fast pre-screening
+without spinning up the full validator.
+
+| Specialist | Input | Output | Notes |
+|-----------|-------|--------|-------|
+| `confidence_tier` | (conflict_count, ambiguity_count, deps, prereqs) | high / medium / low | Learns the deduction formula |
+| `escalation_predictor` | (confidence, ambiguity_count) | escalate: bool | Binary: confidence < 60 OR ambiguity > 0 |
+| `quality_score_tier` | task features | pass / warn / fail | Learns SelfContainmentScorer cutoffs |
+
+### Tier 3: Future Specialists (need data collection)
+
+| Specialist | Input | Output | Blocker |
+|-----------|-------|--------|---------|
+| `library_selector` | task intent + available libs | selected library | No labeled data yet |
+| `milestone_count` | requirement count + conflicts | 1 / 2 / 3 / 4+ milestones | Need more diverse pipeline runs |
+| `task_position_to_dependency` | task index + total count | has_dep: bool | Imbalanced data (97% no-dep) |
+
+---
+
+## Architecture: Fleet Model
+
+```
+┌─────────────────────────────────────────────────────────────────┐
+│ Whetstone MCP Request                                           │
+│ (generate_taskitems, validate_taskitem, ...)                    │
+└────────────────────────┬────────────────────────────────────────┘
+                         │
+                         ▼
+┌─────────────────────────────────────────────────────────────────┐
+│ Specialist Dispatcher (Python: infer.py)                        │
+│                                                                 │
+│  Task text ──► verification_type_specialist ──► unit            │
+│  Task text ──► prereq_op_specialist         ──► [validate, ...]  │
+│  Task text ──► worker_type_specialist       ──► implementer     │
+│  Features  ──► confidence_tier_specialist   ──► high            │
+│                                                                 │
+│  All specialists run in parallel via thread pool                │
+└──────────────┬──────────────────────────────────────────────────┘
+               │ aggregated results
+               ▼
+┌─────────────────────────────────────────────────────────────────┐
+│ Whetstone C++ Runtime                                           │
+│ (enriches taskitem fields with specialist predictions)          │
+└─────────────────────────────────────────────────────────────────┘
+```
+
+### Specialist Model Spec (all identical architecture)
+
+```
+Architecture: Transformer (D=128, L=1, heads=1, seq_len=96)
+Parameters:   ~213K
+Size:         ~832KB per checkpoint
+Inference:    <1ms CPU (numpy), ~50µs GPU
+Training:     1K–5K steps, grokking detection, weight decay
+```
+
+Reuses the WhetstoneAI Fabricate binary (`whetstone_fabricate`) and training harness
+(`run_grokking_until.py`) without modification.
+
+---
+
+## Training Pipeline
+
+```
+sprint_plans/*.md
+        │
+        ▼
+scripts/gen_verification_type_data.py   ← TSV extractor (one per specialist)
+        │
+        ▼
+data/generated/verification_type_train.tsv   (label<TAB>0<TAB>text)
+data/generated/verification_type_eval.tsv
+        │
+        ▼
+scripts/train_verification_type.sh           ← wraps run_grokking_until.py
+        │
+        ▼
+/mnt/storage/fabricate_runs/whetstone_verification_type/
+  checkpoint.bin   (weights + optional Adam state)
+  fabricate.db     (vocabulary)
+        │
+        ▼
+specialists/scripts/infer.py                 ← Python inference wrapper
+        │
+        ▼
+Whetstone MCP tool integration (future)
+```
+
+---
+
+## Data Sources
+
+### Current (18,519 pipeline run records)
+`training_data/lora/taskitem_pipeline_runs.jsonl`
+
+**Limitation:** Almost entirely homogeneous (2 task types repeated, no text diversity).
+Useful for structural/formula specialists (Tier 2). NOT useful for text-based classifiers.
+
+### Primary for Text Specialists
+`sprint*_plan.md` files (260 files, ~2,600 labeled step descriptions)
+
+Each step description has an implicit verification type based on its content:
+- `### Step NNN: whetstone_foo MCP tool` → integration
+- `### Step NNN: Schema for X` → schema
+- `### Step NNN: Sprint N integration summary + regression` → smoke
+- `### Step NNN: X documentation packet` → docs
+- Everything else → unit
+
+### Future Data Collection
+Add richer recording to `run_sprint_taskitem_pipeline.sh`:
+- Capture per-step verification type decisions
+- Record library selection choices
+- Record milestone count decisions
+
+---
+
+## Current Status
+
+| Specialist | Data Source | Training Status | Accuracy |
+|-----------|------------|-----------------|---------|
+| `verification_type` | sprint plans (978 train, 172 eval) | **TRAINING** | TBD |
+| All others | — | planned | — |
+
+---
+
+## Integration Plan (Future)
+
+1. **Phase 1 (now):** Train verification_type specialist, validate accuracy
+2. **Phase 2:** Wire into `whetstone_validate_taskitem` as a pre-prediction step
+3. **Phase 3:** Train prereq_op and confidence_tier specialists
+4. **Phase 4:** Build Python dispatcher that runs all specialists in parallel
+5. **Phase 5:** Expose as new MCP tool `whetstone_specialist_predict`
+6. **Phase 6:** Replace LLM calls for bounded decisions with specialist fleet
+
+---
+
+## Connection to WhetstoneAI Fabricate
+
+This project reuses the Fabricate infrastructure directly:
+- Binary: `/home/bill/Documents/WhetstoneAI_Fabricate/build/whetstone_fabricate`
+- Harness: `/home/bill/Documents/WhetstoneAI_Fabricate/run_grokking_until.py`
+- Eval tools: `/home/bill/Documents/WhetstoneAI_Fabricate/tools/eval_multiclass.py`
+- Inference: `/home/bill/Documents/WhetstoneAI_Fabricate/tools/eval_with_vocab.py`
+
+No code changes to Fabricate are needed. It is consumed as a training and inference substrate.
diff --git a/specialists/eval/eval_specialist.py b/specialists/eval/eval_specialist.py
new file mode 100644
index 0000000..82978eb
--- /dev/null
+++ b/specialists/eval/eval_specialist.py
@@ -0,0 +1,282 @@
+#!/usr/bin/env python3
+"""
+eval_specialist.py
+
+Extended evaluation for whetstone specialist checkpoints.
+Produces confusion matrix, per-class accuracy, confidence distribution,
+and threshold analysis (accepted precision + abstain rate).
+
+Output: JSON to stdout or --out file.
+
+Usage:
+    python3 specialists/eval/eval_specialist.py \
+        --checkpoint /mnt/storage/fabricate_runs/whetstone_worker_type/checkpoint.bin \
+        --db         /mnt/storage/fabricate_runs/whetstone_worker_type/fabricate.db \
+        --dataset    specialists/data/generated/worker_type_eval.tsv \
+        --labels     "implementer,reviewer,architect,qa" \
+        [--out       specialists/eval/results/worker_type_eval.json]
+        [--thresholds 0.5,0.7,0.8,0.9]
+        [--seq-len   96]
+"""
+
+import argparse
+import json
+import sys
+import time
+from collections import defaultdict
+from pathlib import Path
+
+import numpy as np
+
+FABRICATE_TOOLS = Path("/home/bill/Documents/WhetstoneAI_Fabricate/tools")
+sys.path.insert(0, str(FABRICATE_TOOLS))
+
+from eval_with_vocab import (
+    load_checkpoint, load_vocab, load_layers, build_runtime,
+    tokenize_batch, forward_batch,
+)
+
+
+def softmax(x):
+    e = np.exp(x - x.max(axis=-1, keepdims=True))
+    return e / e.sum(axis=-1, keepdims=True)
+
+
+def load_labels(s):
+    p = Path(s)
+    if p.exists():
+        return [t.strip() for t in p.read_text().strip().split(",") if t.strip()]
+    return [t.strip() for t in s.split(",") if t.strip()]
+
+
+def evaluate(checkpoint_path, db_path, dataset_path, label_tokens,
+             seq_len=96, batch_size=256):
+    weights, _ = load_checkpoint(checkpoint_path)
+    _, str_to_id = load_vocab(db_path)
+    layers = load_layers(db_path)
+
+    vocab_size = len(str_to_id)
+    _, emb_cnt = layers["token_embedding"]
+    D = emb_cnt // vocab_size
+    _, ff1_cnt = layers["block0.ff1_weight"]
+    F = ff1_cnt // D
+    L = sum(1 for k in layers if k.startswith("block") and k.endswith(".qkv_weight"))
+    cfg = dict(seq_len=seq_len, D=D, F=F, V=vocab_size, L=L)
+    runtime = build_runtime(weights, layers, cfg)
+
+    label_vocab_ids = []
+    missing = []
+    for tok in label_tokens:
+        vid = str_to_id.get(tok)
+        if vid is None:
+            missing.append(tok)
+            vid = 1
+        label_vocab_ids.append(vid)
+    label_vocab_ids_arr = np.array(label_vocab_ids, dtype=np.int32)
+
+    labels_gt, texts = [], []
+    with open(dataset_path) as f:
+        for line in f:
+            parts = line.strip().split("\t", 2)
+            if len(parts) < 3:
+                continue
+            labels_gt.append(int(parts[0]))
+            texts.append(parts[2])
+
+    labels_gt = np.array(labels_gt, dtype=np.int32)
+    n = len(labels_gt)
+    n_labels = len(label_tokens)
+
+    all_preds = []
+    all_confs = []  # max probability for predicted class
+
+    t0 = time.perf_counter()
+    for start in range(0, n, batch_size):
+        stop = min(start + batch_size, n)
+        token_ids = tokenize_batch(texts[start:stop], str_to_id, seq_len)
+        logits = forward_batch(token_ids, weights, layers, cfg, runtime=runtime)
+        label_logits = logits[:, label_vocab_ids_arr]
+        probs = softmax(label_logits.astype(np.float64))
+        pred_idx = np.argmax(probs, axis=1)
+        conf = probs[np.arange(len(pred_idx)), pred_idx]
+        all_preds.append(pred_idx)
+        all_confs.append(conf)
+
+    eval_latency_ms = (time.perf_counter() - t0) * 1000 / max(n, 1)
+
+    preds = np.concatenate(all_preds)
+    confs = np.concatenate(all_confs)
+    correct = (preds == labels_gt)
+
+    # --- Overall accuracy ---
+    overall_acc = float(correct.mean() * 100)
+
+    # --- Per-class accuracy ---
+    per_class = {}
+    for i, name in enumerate(label_tokens):
+        mask = labels_gt == i
+        if mask.sum() == 0:
+            per_class[name] = {"acc": None, "n": 0}
+        else:
+            per_class[name] = {
+                "acc": float(correct[mask].mean() * 100),
+                "n": int(mask.sum()),
+            }
+
+    # --- Confusion matrix ---
+    confusion = np.zeros((n_labels, n_labels), dtype=int)
+    for gt, pred in zip(labels_gt, preds):
+        confusion[gt, pred] += 1
+
+    # Adjacent confusion (common mistake pairs)
+    confusion_summary = []
+    for i in range(n_labels):
+        for j in range(n_labels):
+            if i != j and confusion[i, j] > 0:
+                confusion_summary.append({
+                    "gt": label_tokens[i],
+                    "pred": label_tokens[j],
+                    "count": int(confusion[i, j]),
+                    "gt_total": int((labels_gt == i).sum()),
+                })
+    confusion_summary.sort(key=lambda x: -x["count"])
+
+    # --- Confidence distribution ---
+    conf_bins = [0.0, 0.5, 0.6, 0.7, 0.8, 0.9, 0.95, 1.0]
+    conf_hist = {}
+    for lo, hi in zip(conf_bins[:-1], conf_bins[1:]):
+        mask = (confs >= lo) & (confs < hi)
+        n_in_bin = int(mask.sum())
+        if n_in_bin == 0:
+            continue
+        acc_in_bin = float(correct[mask].mean() * 100) if n_in_bin > 0 else None
+        conf_hist[f"{lo:.2f}-{hi:.2f}"] = {"n": n_in_bin, "acc": acc_in_bin}
+
+    # --- Threshold analysis ---
+    threshold_results = []
+    for threshold in [0.5, 0.6, 0.7, 0.8, 0.9]:
+        accepted_mask = confs >= threshold
+        n_accepted = int(accepted_mask.sum())
+        n_abstained = n - n_accepted
+        if n_accepted == 0:
+            threshold_results.append({
+                "threshold": threshold,
+                "accepted": 0,
+                "abstain_rate": 1.0,
+                "accepted_accuracy": None,
+            })
+        else:
+            acc = float(correct[accepted_mask].mean() * 100)
+            threshold_results.append({
+                "threshold": threshold,
+                "accepted": n_accepted,
+                "abstain_rate": round(n_abstained / n, 3),
+                "accepted_accuracy": round(acc, 1),
+            })
+
+    # --- Guardrail assessment ---
+    # Find lowest threshold where accepted_accuracy >= 95%
+    guardrail_threshold = None
+    guardrail_abstain_rate = None
+    for tr in threshold_results:
+        if tr["accepted_accuracy"] is not None and tr["accepted_accuracy"] >= 95.0:
+            guardrail_threshold = tr["threshold"]
+            guardrail_abstain_rate = tr["abstain_rate"]
+            break
+
+    # --- Deployment verdict ---
+    if overall_acc >= 97:
+        verdict = "deploy_direct"
+    elif guardrail_threshold is not None and guardrail_abstain_rate is not None and guardrail_abstain_rate <= 0.30:
+        verdict = "deploy_with_guardrails"
+    elif overall_acc >= 80:
+        verdict = "scale_more"
+    elif len(confusion_summary) > 0 and confusion_summary[0]["count"] / n > 0.10:
+        verdict = "decompose_or_relabel"
+    else:
+        verdict = "needs_investigation"
+
+    return {
+        "overall_accuracy": round(overall_acc, 1),
+        "n_eval": n,
+        "per_class": per_class,
+        "confusion_summary": confusion_summary[:10],
+        "confidence_histogram": conf_hist,
+        "threshold_analysis": threshold_results,
+        "guardrail_threshold": guardrail_threshold,
+        "guardrail_abstain_rate": guardrail_abstain_rate,
+        "eval_latency_ms_per_example": round(eval_latency_ms, 3),
+        "missing_label_tokens": missing,
+        "model_config": {"D": D, "F": F, "L": L, "vocab": vocab_size},
+        "verdict": verdict,
+    }
+
+
+def print_report(result, labels):
+    print(f"\n{'='*60}")
+    print(f"  Overall accuracy: {result['overall_accuracy']:.1f}%  (n={result['n_eval']})")
+    print(f"  Verdict: {result['verdict']}")
+    print(f"{'='*60}")
+
+    print("\nPer-class accuracy:")
+    for name, stat in result["per_class"].items():
+        if stat["acc"] is None:
+            print(f"  {name:20s}  (no examples)")
+        else:
+            print(f"  {name:20s}  {stat['acc']:5.1f}%  (n={stat['n']})")
+
+    if result["confusion_summary"]:
+        print("\nTop confusions (gt → pred):")
+        for c in result["confusion_summary"][:6]:
+            pct = c["count"] / c["gt_total"] * 100
+            print(f"  {c['gt']:20s} → {c['pred']:20s}  {c['count']}x  ({pct:.0f}% of gt class)")
+
+    print("\nThreshold analysis:")
+    print(f"  {'thresh':>6}  {'accepted':>8}  {'abstain%':>8}  {'acc@thresh':>10}")
+    for tr in result["threshold_analysis"]:
+        abstain_pct = tr["abstain_rate"] * 100
+        acc = f"{tr['accepted_accuracy']:.1f}%" if tr["accepted_accuracy"] is not None else "  n/a"
+        print(f"  {tr['threshold']:>6.2f}  {tr['accepted']:>8}  {abstain_pct:>7.1f}%  {acc:>10}")
+
+    if result["guardrail_threshold"]:
+        print(f"\n  ✓ Guarded deployment viable at threshold={result['guardrail_threshold']:.2f}"
+              f"  (abstain {result['guardrail_abstain_rate']*100:.0f}%)")
+    else:
+        print("\n  ✗ No threshold achieves 95% accepted accuracy")
+
+    print(f"\nEval latency: {result['eval_latency_ms_per_example']:.3f} ms/example")
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--checkpoint", required=True)
+    ap.add_argument("--db",         required=True)
+    ap.add_argument("--dataset",    required=True)
+    ap.add_argument("--labels",     required=True)
+    ap.add_argument("--seq-len",    type=int, default=96)
+    ap.add_argument("--batch-size", type=int, default=256)
+    ap.add_argument("--out",        default=None, help="Write JSON result to file")
+    ap.add_argument("--quiet",      action="store_true", help="Suppress human-readable output")
+    args = ap.parse_args()
+
+    label_tokens = load_labels(args.labels)
+
+    result = evaluate(
+        args.checkpoint, args.db, args.dataset,
+        label_tokens, seq_len=args.seq_len, batch_size=args.batch_size,
+    )
+
+    if not args.quiet:
+        print_report(result, label_tokens)
+
+    if args.out:
+        out_path = Path(args.out)
+        out_path.parent.mkdir(parents=True, exist_ok=True)
+        out_path.write_text(json.dumps(result, indent=2))
+        print(f"\nJSON → {args.out}")
+
+    print(json.dumps(result, indent=2))
+
+
+if __name__ == "__main__":
+    main()
diff --git a/specialists/eval/eval_specialist_pt.py b/specialists/eval/eval_specialist_pt.py
new file mode 100644
index 0000000..7a0dba2
--- /dev/null
+++ b/specialists/eval/eval_specialist_pt.py
@@ -0,0 +1,296 @@
+#!/usr/bin/env python3
+"""
+eval_specialist_pt.py
+
+Extended evaluation for PyTorch specialist checkpoints.
+Drop-in replacement for eval_specialist.py, but loads checkpoint.pt instead
+of checkpoint.bin + fabricate.db.
+
+Output: JSON to stdout (and optionally --out file).
+
+Usage:
+    python3 specialists/eval/eval_specialist_pt.py \
+        --checkpoint /tmp/runs/worker_type/checkpoint.pt \
+        --dataset    specialists/data/generated/worker_type_eval.tsv \
+        --labels     "implementer,reviewer,architect,qa" \
+        [--out       specialists/eval/results/worker_type_eval.json] \
+        [--quiet]
+"""
+
+import argparse
+import json
+import time
+from pathlib import Path
+
+import numpy as np
+import torch
+import torch.nn as nn
+from torch.utils.data import DataLoader, Dataset
+
+
+# ── Copied from train_specialist_pt.py ───────────────────────────────────────
+
+def _tokenize(text):
+    return text.lower().split()
+
+
+class Vocab:
+    PAD, UNK = 0, 1
+
+    def __init__(self, str_to_id=None):
+        if str_to_id is not None:
+            self.str_to_id = str_to_id
+            self.id_to_str = [None] * len(str_to_id)
+            for s, i in str_to_id.items():
+                self.id_to_str[i] = s
+        else:
+            self.str_to_id = {"<pad>": 0, "<unk>": 1}
+            self.id_to_str = ["<pad>", "<unk>"]
+
+    def encode(self, text, seq_len):
+        ids = [self.str_to_id.get(t, self.UNK) for t in _tokenize(text)]
+        ids = ids[:seq_len]
+        ids += [self.PAD] * (seq_len - len(ids))
+        return ids
+
+    def __len__(self):
+        return len(self.id_to_str)
+
+
+class SpecialistModel(nn.Module):
+    def __init__(self, vocab_size, hidden_dim, layers, heads, n_labels, seq_len):
+        super().__init__()
+        self.seq_len = seq_len
+        self.tok_emb = nn.Embedding(vocab_size, hidden_dim, padding_idx=0)
+        self.pos_emb = nn.Embedding(seq_len, hidden_dim)
+        encoder_layer = nn.TransformerEncoderLayer(
+            d_model=hidden_dim,
+            nhead=heads,
+            dim_feedforward=hidden_dim * 4,
+            dropout=0.0,
+            batch_first=True,
+        )
+        self.encoder = nn.TransformerEncoder(encoder_layer, num_layers=layers)
+        self.head = nn.Linear(hidden_dim, n_labels)
+
+    def forward(self, x):
+        pad_mask = (x == 0)
+        pos = torch.arange(x.size(1), device=x.device).unsqueeze(0)
+        h = self.tok_emb(x) + self.pos_emb(pos)
+        h = self.encoder(h, src_key_padding_mask=pad_mask)
+        lengths = (~pad_mask).float().sum(dim=1, keepdim=True).clamp(min=1)
+        h = (h * (~pad_mask).unsqueeze(-1).float()).sum(dim=1) / lengths
+        return self.head(h)
+
+
+# ── Eval ──────────────────────────────────────────────────────────────────────
+
+def softmax(x):
+    e = np.exp(x - x.max(axis=-1, keepdims=True))
+    return e / e.sum(axis=-1, keepdims=True)
+
+
+def evaluate(checkpoint_path, dataset_path, label_tokens, seq_len=96, batch_size=256):
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+    ckpt = torch.load(checkpoint_path, map_location=device)
+    cfg  = ckpt["config"]
+
+    # Reconstruct vocab from checkpoint
+    vocab_obj = ckpt.get("vocab_obj")
+    if vocab_obj is None:
+        vocab_obj = Vocab(str_to_id=ckpt["vocab"])
+
+    model = SpecialistModel(
+        vocab_size=cfg["vocab_size"],
+        hidden_dim=cfg["hidden_dim"],
+        layers=cfg["layers"],
+        heads=cfg["heads"],
+        n_labels=cfg["n_labels"],
+        seq_len=cfg["seq_len"],
+    ).to(device)
+    model.load_state_dict(ckpt["model"])
+    model.eval()
+
+    # Load eval data
+    labels_gt, texts = [], []
+    with open(dataset_path) as f:
+        for line in f:
+            parts = line.strip().split("\t", 2)
+            if len(parts) < 3:
+                continue
+            labels_gt.append(int(parts[0]))
+            texts.append(parts[2])
+
+    n = len(labels_gt)
+    labels_gt_arr = np.array(labels_gt, dtype=np.int32)
+
+    # Run inference
+    all_preds = []
+    all_confs = []
+    t0 = time.perf_counter()
+
+    with torch.no_grad():
+        for start in range(0, n, batch_size):
+            batch_texts = texts[start:start + batch_size]
+            ids = [vocab_obj.encode(t, cfg["seq_len"]) for t in batch_texts]
+            x = torch.tensor(ids, dtype=torch.long, device=device)
+            logits = model(x).cpu().numpy().astype(np.float64)
+            probs = softmax(logits)
+            pred = np.argmax(probs, axis=1)
+            conf = probs[np.arange(len(pred)), pred]
+            all_preds.append(pred)
+            all_confs.append(conf)
+
+    eval_latency_ms = (time.perf_counter() - t0) * 1000 / max(n, 1)
+    preds = np.concatenate(all_preds)
+    confs = np.concatenate(all_confs)
+    correct = (preds == labels_gt_arr)
+    overall_acc = float(correct.mean() * 100)
+
+    # Per-class accuracy
+    per_class = {}
+    for i, name in enumerate(label_tokens):
+        mask = labels_gt_arr == i
+        if mask.sum() == 0:
+            per_class[name] = {"acc": None, "n": 0}
+        else:
+            per_class[name] = {
+                "acc": float(correct[mask].mean() * 100),
+                "n": int(mask.sum()),
+            }
+
+    # Confusion matrix
+    n_labels = len(label_tokens)
+    confusion = np.zeros((n_labels, n_labels), dtype=int)
+    for gt, pred in zip(labels_gt_arr, preds):
+        confusion[gt, pred] += 1
+
+    confusion_summary = []
+    for i in range(n_labels):
+        for j in range(n_labels):
+            if i != j and confusion[i, j] > 0:
+                confusion_summary.append({
+                    "gt": label_tokens[i],
+                    "pred": label_tokens[j],
+                    "count": int(confusion[i, j]),
+                    "gt_total": int((labels_gt_arr == i).sum()),
+                })
+    confusion_summary.sort(key=lambda x: -x["count"])
+
+    # Threshold analysis
+    threshold_results = []
+    for threshold in [0.5, 0.6, 0.7, 0.8, 0.9]:
+        accepted_mask = confs >= threshold
+        n_accepted = int(accepted_mask.sum())
+        n_abstained = n - n_accepted
+        if n_accepted == 0:
+            threshold_results.append({
+                "threshold": threshold, "accepted": 0,
+                "abstain_rate": 1.0, "accepted_accuracy": None,
+            })
+        else:
+            acc = float(correct[accepted_mask].mean() * 100)
+            threshold_results.append({
+                "threshold": threshold,
+                "accepted": n_accepted,
+                "abstain_rate": round(n_abstained / n, 3),
+                "accepted_accuracy": round(acc, 1),
+            })
+
+    # Guardrail assessment (lowest threshold with ≥95% accepted accuracy)
+    guardrail_threshold = None
+    guardrail_abstain_rate = None
+    for tr in threshold_results:
+        if tr["accepted_accuracy"] is not None and tr["accepted_accuracy"] >= 95.0:
+            guardrail_threshold = tr["threshold"]
+            guardrail_abstain_rate = tr["abstain_rate"]
+            break
+
+    # Deployment verdict
+    if overall_acc >= 97:
+        verdict = "deploy_direct"
+    elif guardrail_threshold is not None and guardrail_abstain_rate is not None and guardrail_abstain_rate <= 0.30:
+        verdict = "deploy_with_guardrails"
+    elif overall_acc >= 80:
+        verdict = "scale_more"
+    elif confusion_summary and confusion_summary[0]["count"] / n > 0.10:
+        verdict = "decompose_or_relabel"
+    else:
+        verdict = "needs_investigation"
+
+    return {
+        "overall_accuracy": round(overall_acc, 1),
+        "n_eval": n,
+        "per_class": per_class,
+        "confusion_summary": confusion_summary[:10],
+        "threshold_analysis": threshold_results,
+        "guardrail_threshold": guardrail_threshold,
+        "guardrail_abstain_rate": guardrail_abstain_rate,
+        "eval_latency_ms_per_example": round(eval_latency_ms, 3),
+        "model_config": cfg,
+        "verdict": verdict,
+    }
+
+
+def print_report(result, labels):
+    print(f"\n{'='*60}")
+    print(f"  Overall accuracy: {result['overall_accuracy']:.1f}%  (n={result['n_eval']})")
+    print(f"  Verdict: {result['verdict']}")
+    print(f"{'='*60}")
+    print("\nPer-class accuracy:")
+    for name, stat in result["per_class"].items():
+        if stat["acc"] is None:
+            print(f"  {name:20s}  (no examples)")
+        else:
+            print(f"  {name:20s}  {stat['acc']:5.1f}%  (n={stat['n']})")
+    if result["confusion_summary"]:
+        print("\nTop confusions (gt → pred):")
+        for c in result["confusion_summary"][:6]:
+            pct = c["count"] / c["gt_total"] * 100
+            print(f"  {c['gt']:20s} → {c['pred']:20s}  {c['count']}x  ({pct:.0f}% of gt class)")
+    print("\nThreshold analysis:")
+    print(f"  {'thresh':>6}  {'accepted':>8}  {'abstain%':>8}  {'acc@thresh':>10}")
+    for tr in result["threshold_analysis"]:
+        abstain_pct = tr["abstain_rate"] * 100
+        acc = f"{tr['accepted_accuracy']:.1f}%" if tr["accepted_accuracy"] is not None else "  n/a"
+        print(f"  {tr['threshold']:>6.2f}  {tr['accepted']:>8}  {abstain_pct:>7.1f}%  {acc:>10}")
+    if result["guardrail_threshold"]:
+        print(f"\n  Guarded deployment viable at threshold={result['guardrail_threshold']:.2f}"
+              f"  (abstain {result['guardrail_abstain_rate']*100:.0f}%)")
+    else:
+        print("\n  No threshold achieves 95% accepted accuracy")
+    print(f"\nEval latency: {result['eval_latency_ms_per_example']:.3f} ms/example")
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--checkpoint", required=True)
+    ap.add_argument("--dataset",    required=True)
+    ap.add_argument("--labels",     required=True)
+    ap.add_argument("--seq-len",    type=int, default=96)
+    ap.add_argument("--batch-size", type=int, default=256)
+    ap.add_argument("--out",        default=None)
+    ap.add_argument("--quiet",      action="store_true")
+    # Accept --db for compatibility with old callers (ignored)
+    ap.add_argument("--db",         default=None)
+    args = ap.parse_args()
+
+    label_tokens = [l.strip() for l in args.labels.split(",")]
+    result = evaluate(args.checkpoint, args.dataset, label_tokens,
+                      seq_len=args.seq_len, batch_size=args.batch_size)
+
+    if not args.quiet:
+        print_report(result, label_tokens)
+
+    if args.out:
+        out_path = Path(args.out)
+        out_path.parent.mkdir(parents=True, exist_ok=True)
+        out_path.write_text(json.dumps(result, indent=2))
+        print(f"\nJSON → {args.out}")
+
+    print(json.dumps(result, indent=2))
+
+
+if __name__ == "__main__":
+    main()
diff --git a/specialists/eval/results/verification_type_sweep.log b/specialists/eval/results/verification_type_sweep.log
new file mode 100644
index 0000000..0ba85e6
--- /dev/null
+++ b/specialists/eval/results/verification_type_sweep.log
@@ -0,0 +1,10 @@
+
+============================================================
+  Tier: small_plus  hidden=256 layers=2
+  Run dir: /mnt/storage/fabricate_runs/whetstone_verification_type_small_plus
+============================================================
+
+  [pilot] 3000 steps...
+=== run_grokking_until: whetstone_verification_type_small_plus ===
+dataset=verification_type_train.tsv  lr=0.001  wd=0.01  max_steps=3000
+Fresh start
diff --git a/specialists/runs/automatability.log b/specialists/runs/automatability.log
new file mode 100644
index 0000000..8a0cc4f
--- /dev/null
+++ b/specialists/runs/automatability.log
@@ -0,0 +1,18 @@
+Training automatability_strategy specialist...
+  train: /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/automatability_train.tsv
+  eval:  /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/automatability_eval.tsv
+  out:   /mnt/storage/fabricate_runs/whetstone_automatability
+  log:   /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/runs/automatability_training.log
+
+=== run_grokking_until: whetstone_automatability ===
+dataset=automatability_train.tsv  lr=0.001  wd=0.01  max_steps=10000
+Resuming from step 3000
+Resuming: grokking already detected at step 1000, 3 blocks since grokking.
+step=  4000  overall=69.6%  T=69.6%  F=69.6%  loss=0.0000  min=0.0000  delta=+0.0%  free=904GB <<< GROKKING (loss)
+Stopping 4 blocks after grokking (step 4000).
+
+=== DONE: whetstone_automatability ===
+Final: step=4000  overall=69.6%
+Grokking at step 1000
+
+Done. Checkpoint at: /mnt/storage/fabricate_runs/whetstone_automatability/checkpoint.bin
diff --git a/specialists/runs/automatability_history.json b/specialists/runs/automatability_history.json
new file mode 100644
index 0000000..f98b9a7
--- /dev/null
+++ b/specialists/runs/automatability_history.json
@@ -0,0 +1,48 @@
+{
+  "dataset": "/home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/automatability_train.tsv",
+  "lr": 0.001,
+  "weight_decay": 0.01,
+  "seed": "1234ABCD",
+  "history": [
+    {
+      "step": 1000,
+      "overall": 69.56521739130434,
+      "true_pct": 69.56521739130434,
+      "false_pct": 69.56521739130434,
+      "delta": 19.565217391304344,
+      "end_loss": 0.000246,
+      "min_loss": 0.000246,
+      "grokking": true
+    },
+    {
+      "step": 2000,
+      "overall": 69.56521739130434,
+      "true_pct": 69.56521739130434,
+      "false_pct": 69.56521739130434,
+      "delta": 0.0,
+      "end_loss": 9.2e-05,
+      "min_loss": 9.2e-05,
+      "grokking": true
+    },
+    {
+      "step": 3000,
+      "overall": 69.56521739130434,
+      "true_pct": 69.56521739130434,
+      "false_pct": 69.56521739130434,
+      "delta": 0.0,
+      "end_loss": 4.8e-05,
+      "min_loss": 4.6e-05,
+      "grokking": true
+    },
+    {
+      "step": 4000,
+      "overall": 69.56521739130434,
+      "true_pct": 69.56521739130434,
+      "false_pct": 69.56521739130434,
+      "delta": 0.0,
+      "end_loss": 2.7e-05,
+      "min_loss": 2.6e-05,
+      "grokking": true
+    }
+  ]
+}
diff --git a/specialists/runs/automatability_training.log b/specialists/runs/automatability_training.log
new file mode 100644
index 0000000..23bf41d
--- /dev/null
+++ b/specialists/runs/automatability_training.log
@@ -0,0 +1,10 @@
+=== run_grokking_until: whetstone_automatability ===
+dataset=automatability_train.tsv  lr=0.001  wd=0.01  max_steps=10000
+Resuming from step 3000
+Resuming: grokking already detected at step 1000, 3 blocks since grokking.
+step=  4000  overall=69.6%  T=69.6%  F=69.6%  loss=0.0000  min=0.0000  delta=+0.0%  free=904GB <<< GROKKING (loss)
+Stopping 4 blocks after grokking (step 4000).
+
+=== DONE: whetstone_automatability ===
+Final: step=4000  overall=69.6%
+Grokking at step 1000
diff --git a/specialists/runs/confidence_tier_history.json b/specialists/runs/confidence_tier_history.json
new file mode 100644
index 0000000..93cc82b
--- /dev/null
+++ b/specialists/runs/confidence_tier_history.json
@@ -0,0 +1,38 @@
+{
+  "dataset": "/home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/confidence_tier_train.tsv",
+  "lr": 0.001,
+  "weight_decay": 0.01,
+  "seed": "1234ABCD",
+  "history": [
+    {
+      "step": 1000,
+      "overall": 100.0,
+      "true_pct": 100.0,
+      "false_pct": 100.0,
+      "delta": 50.0,
+      "end_loss": 0.000587,
+      "min_loss": 0.000427,
+      "grokking": true
+    },
+    {
+      "step": 2000,
+      "overall": 100.0,
+      "true_pct": 100.0,
+      "false_pct": 100.0,
+      "delta": 0.0,
+      "end_loss": 0.000178,
+      "min_loss": 9.1e-05,
+      "grokking": true
+    },
+    {
+      "step": 3000,
+      "overall": 100.0,
+      "true_pct": 100.0,
+      "false_pct": 100.0,
+      "delta": 0.0,
+      "end_loss": 5.1e-05,
+      "min_loss": 3e-05,
+      "grokking": true
+    }
+  ]
+}
diff --git a/specialists/runs/confidence_tier_training.log b/specialists/runs/confidence_tier_training.log
new file mode 100644
index 0000000..f358a9d
--- /dev/null
+++ b/specialists/runs/confidence_tier_training.log
@@ -0,0 +1,16 @@
+=== run_grokking_until: whetstone_confidence_tier ===
+dataset=confidence_tier_train.tsv  lr=0.001  wd=0.01  max_steps=5000
+Fresh start
+step=  1000  overall=100.0%  T=100.0%  F=100.0%  loss=0.0006  min=0.0004  delta=+50.0%  free=836GB <<< GROKKING (accuracy jump)
+
+============================================================
+GROKKING at step 1000! (stopping after 3 more blocks)
+============================================================
+
+step=  2000  overall=100.0%  T=100.0%  F=100.0%  loss=0.0002  min=0.0001  delta=+0.0%  free=834GB <<< GROKKING (loss)
+step=  3000  overall=100.0%  T=100.0%  F=100.0%  loss=0.0001  min=0.0000  delta=+0.0%  free=832GB <<< GROKKING (loss)
+Stopping 3 blocks after grokking (step 3000).
+
+=== DONE: whetstone_confidence_tier ===
+Final: step=3000  overall=100.0%
+Grokking at step 1000
diff --git a/specialists/runs/prereq_op.log b/specialists/runs/prereq_op.log
new file mode 100644
index 0000000..c448ede
--- /dev/null
+++ b/specialists/runs/prereq_op.log
@@ -0,0 +1,18 @@
+Training prereq_op_selector specialist...
+  train: /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/prereq_op_train.tsv
+  eval:  /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/prereq_op_eval.tsv
+  out:   /mnt/storage/fabricate_runs/whetstone_prereq_op
+  log:   /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/runs/prereq_op_training.log
+
+=== run_grokking_until: whetstone_prereq_op ===
+dataset=prereq_op_train.tsv  lr=0.001  wd=0.01  max_steps=5000
+Resuming from step 3000
+Resuming: grokking already detected at step 1000, 3 blocks since grokking.
+step=  4000  overall=62.5%  T=62.5%  F=62.5%  loss=0.0000  min=0.0000  delta=+0.0%  free=904GB <<< GROKKING (loss)
+Stopping 4 blocks after grokking (step 4000).
+
+=== DONE: whetstone_prereq_op ===
+Final: step=4000  overall=62.5%
+Grokking at step 1000
+
+Done. Checkpoint at: /mnt/storage/fabricate_runs/whetstone_prereq_op/checkpoint.bin
diff --git a/specialists/runs/prereq_op_combined.log b/specialists/runs/prereq_op_combined.log
new file mode 100644
index 0000000..81a9c64
--- /dev/null
+++ b/specialists/runs/prereq_op_combined.log
@@ -0,0 +1,17 @@
+=== run_grokking_until: whetstone_prereq_op ===
+dataset=prereq_op_train.tsv  lr=0.001  wd=0.01  max_steps=8000
+Fresh start
+step=  1000  overall=69.4%  T=69.4%  F=69.4%  loss=0.0004  min=0.0004  delta=+19.4%  free=901GB <<< GROKKING (accuracy jump)
+
+============================================================
+GROKKING at step 1000! (stopping after 4 more blocks)
+============================================================
+
+step=  2000  overall=69.4%  T=69.4%  F=69.4%  loss=0.0001  min=0.0001  delta=+0.0%  free=897GB <<< GROKKING (loss)
+step=  3000  overall=69.4%  T=69.4%  F=69.4%  loss=0.0001  min=0.0001  delta=+0.0%  free=893GB <<< GROKKING (loss)
+step=  4000  overall=69.4%  T=69.4%  F=69.4%  loss=0.0000  min=0.0000  delta=+0.0%  free=890GB <<< GROKKING (loss)
+Stopping 4 blocks after grokking (step 4000).
+
+=== DONE: whetstone_prereq_op ===
+Final: step=4000  overall=69.4%
+Grokking at step 1000
diff --git a/specialists/runs/prereq_op_history.json b/specialists/runs/prereq_op_history.json
new file mode 100644
index 0000000..7b72ee8
--- /dev/null
+++ b/specialists/runs/prereq_op_history.json
@@ -0,0 +1,48 @@
+{
+  "dataset": "/home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/combined/prereq_op_train.tsv",
+  "lr": 0.001,
+  "weight_decay": 0.01,
+  "seed": "1234ABCD",
+  "history": [
+    {
+      "step": 1000,
+      "overall": 69.44444444444444,
+      "true_pct": 69.44444444444444,
+      "false_pct": 69.44444444444444,
+      "delta": 19.444444444444443,
+      "end_loss": 0.000413,
+      "min_loss": 0.0004,
+      "grokking": true
+    },
+    {
+      "step": 2000,
+      "overall": 69.44444444444444,
+      "true_pct": 69.44444444444444,
+      "false_pct": 69.44444444444444,
+      "delta": 0.0,
+      "end_loss": 0.000127,
+      "min_loss": 0.000127,
+      "grokking": true
+    },
+    {
+      "step": 3000,
+      "overall": 69.44444444444444,
+      "true_pct": 69.44444444444444,
+      "false_pct": 69.44444444444444,
+      "delta": 0.0,
+      "end_loss": 5.5e-05,
+      "min_loss": 5.4e-05,
+      "grokking": true
+    },
+    {
+      "step": 4000,
+      "overall": 69.44444444444444,
+      "true_pct": 69.44444444444444,
+      "false_pct": 69.44444444444444,
+      "delta": 0.0,
+      "end_loss": 2.2e-05,
+      "min_loss": 2e-05,
+      "grokking": true
+    }
+  ]
+}
diff --git a/specialists/runs/prereq_op_training.log b/specialists/runs/prereq_op_training.log
new file mode 100644
index 0000000..872602a
--- /dev/null
+++ b/specialists/runs/prereq_op_training.log
@@ -0,0 +1,10 @@
+=== run_grokking_until: whetstone_prereq_op ===
+dataset=prereq_op_train.tsv  lr=0.001  wd=0.01  max_steps=5000
+Resuming from step 3000
+Resuming: grokking already detected at step 1000, 3 blocks since grokking.
+step=  4000  overall=62.5%  T=62.5%  F=62.5%  loss=0.0000  min=0.0000  delta=+0.0%  free=904GB <<< GROKKING (loss)
+Stopping 4 blocks after grokking (step 4000).
+
+=== DONE: whetstone_prereq_op ===
+Final: step=4000  overall=62.5%
+Grokking at step 1000
diff --git a/specialists/runs/verif_type.log b/specialists/runs/verif_type.log
new file mode 100644
index 0000000..7a3b9ac
--- /dev/null
+++ b/specialists/runs/verif_type.log
@@ -0,0 +1,18 @@
+Training verification_type specialist...
+  train: /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/verification_type_train.tsv
+  eval:  /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/verification_type_eval.tsv
+  out:   /mnt/storage/fabricate_runs/whetstone_verification_type
+  log:   /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/runs/verification_type_training.log
+
+=== run_grokking_until: whetstone_verification_type ===
+dataset=verification_type_train.tsv  lr=0.001  wd=0.01  max_steps=5000
+Resuming from step 3000
+Resuming: grokking already detected at step 1000, 3 blocks since grokking.
+step=  4000  overall=84.9%  T=84.9%  F=84.9%  loss=0.0000  min=0.0000  delta=+1.2%  free=904GB <<< GROKKING (loss)
+Stopping 4 blocks after grokking (step 4000).
+
+=== DONE: whetstone_verification_type ===
+Final: step=4000  overall=84.9%
+Grokking at step 1000
+
+Done. Checkpoint at: /mnt/storage/fabricate_runs/whetstone_verification_type/checkpoint.bin
diff --git a/specialists/runs/verification_type_history.json b/specialists/runs/verification_type_history.json
new file mode 100644
index 0000000..a52ac40
--- /dev/null
+++ b/specialists/runs/verification_type_history.json
@@ -0,0 +1,48 @@
+{
+  "dataset": "/home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/verification_type_train.tsv",
+  "lr": 0.001,
+  "weight_decay": 0.01,
+  "seed": "1234ABCD",
+  "history": [
+    {
+      "step": 1000,
+      "overall": 81.97674418604652,
+      "true_pct": 81.97674418604652,
+      "false_pct": 81.97674418604652,
+      "delta": 31.976744186046517,
+      "end_loss": 0.000621,
+      "min_loss": 0.0006,
+      "grokking": true
+    },
+    {
+      "step": 2000,
+      "overall": 83.13953488372093,
+      "true_pct": 83.13953488372093,
+      "false_pct": 83.13953488372093,
+      "delta": 1.16279069767441,
+      "end_loss": 0.000206,
+      "min_loss": 0.000175,
+      "grokking": true
+    },
+    {
+      "step": 3000,
+      "overall": 83.72093023255815,
+      "true_pct": 83.72093023255815,
+      "false_pct": 83.72093023255815,
+      "delta": 0.5813953488372192,
+      "end_loss": 5.1e-05,
+      "min_loss": 5.1e-05,
+      "grokking": true
+    },
+    {
+      "step": 4000,
+      "overall": 84.88372093023256,
+      "true_pct": 84.88372093023256,
+      "false_pct": 84.88372093023256,
+      "delta": 1.16279069767441,
+      "end_loss": 3.1e-05,
+      "min_loss": 2.5e-05,
+      "grokking": true
+    }
+  ]
+}
diff --git a/specialists/runs/verification_type_training.log b/specialists/runs/verification_type_training.log
new file mode 100644
index 0000000..5d8fe41
--- /dev/null
+++ b/specialists/runs/verification_type_training.log
@@ -0,0 +1,10 @@
+=== run_grokking_until: whetstone_verification_type ===
+dataset=verification_type_train.tsv  lr=0.001  wd=0.01  max_steps=5000
+Resuming from step 3000
+Resuming: grokking already detected at step 1000, 3 blocks since grokking.
+step=  4000  overall=84.9%  T=84.9%  F=84.9%  loss=0.0000  min=0.0000  delta=+1.2%  free=904GB <<< GROKKING (loss)
+Stopping 4 blocks after grokking (step 4000).
+
+=== DONE: whetstone_verification_type ===
+Final: step=4000  overall=84.9%
+Grokking at step 1000
diff --git a/specialists/runs/whetstone_automatability_large.log b/specialists/runs/whetstone_automatability_large.log
new file mode 100644
index 0000000..6075014
--- /dev/null
+++ b/specialists/runs/whetstone_automatability_large.log
@@ -0,0 +1,44 @@
+=== whetstone_automatability (large) ===
+  train:  /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/automatability_train.tsv
+  eval:   /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/automatability_eval.tsv
+  model:  hidden=512 layers=4 heads=1
+
+=== run_grokking_until: whetstone_automatability_large ===
+dataset=automatability_train.tsv  lr=0.0005  wd=0.01  max_steps=20000
+Fresh start
+Traceback (most recent call last):
+  File "/home/bill/Documents/WhetstoneAI_Fabricate/run_grokking_until.py", line 299, in <module>
+    main()
+  File "/home/bill/Documents/WhetstoneAI_Fabricate/run_grokking_until.py", line 220, in main
+    block_losses = train_block(
+                   ^^^^^^^^^^^^
+  File "/home/bill/Documents/WhetstoneAI_Fabricate/run_grokking_until.py", line 94, in train_block
+    raise RuntimeError(proc.stderr[-2000:] or proc.stdout[-2000:])
+RuntimeError: === Whetstone Fabricate ===
+steps=1000  batch_size=32  lr=0.000500  weight_decay=0.010000  hidden_dim=512  model=gpu_transformer  morphisms=binary .morf (fast)
+
+[1] Loading dataset...
+ProntoQADataset: 266 examples, vocab_size=140, seq_len=96
+
+[2] ModelConfig
+    vocab_size=140  embedding_dim=512  hidden_dim=512  layers=4  heads=1  ff_hidden_dim=2048  seq_len=96
+
+[3] Constructing model engine...
+    num_params=12795020
+    morphism binary: 97.6 MB/step  ->  97618.3 MB for 1000 steps
+
+[4] Output directory: /mnt/storage/fabricate_runs/whetstone_automatability_large
+[5] --reset: removing checkpoint and output files.
+[6] TrajectoryWriter: /mnt/storage/fabricate_runs/whetstone_automatability_large/train.fabr  (fresh)
+    Initial weights written.
+[7] MorphismWriter: /mnt/storage/fabricate_runs/whetstone_automatability_large/morphisms.morf  (fresh)
+    Building fabricate.db (static tables only)...
+Initialising weight_metadata (12795020 params)...
+  weight_metadata populated.
+    Static tables ready.
+
+[8] Training (1000 update steps from step 0, batch_size=32, lr=0.000500)
+    step    loss          task
+    ------  ------------  ----
+    0       5.978329      0
+
diff --git a/specialists/runs/whetstone_verification_type_large.log b/specialists/runs/whetstone_verification_type_large.log
new file mode 100644
index 0000000..9bcb32c
--- /dev/null
+++ b/specialists/runs/whetstone_verification_type_large.log
@@ -0,0 +1,44 @@
+=== whetstone_verification_type (large) ===
+  train:  /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/verification_type_train.tsv
+  eval:   /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/verification_type_eval.tsv
+  model:  hidden=512 layers=4 heads=1
+
+=== run_grokking_until: whetstone_verification_type_large ===
+dataset=verification_type_train.tsv  lr=0.0005  wd=0.01  max_steps=20000
+Fresh start
+Traceback (most recent call last):
+  File "/home/bill/Documents/WhetstoneAI_Fabricate/run_grokking_until.py", line 299, in <module>
+    main()
+  File "/home/bill/Documents/WhetstoneAI_Fabricate/run_grokking_until.py", line 220, in main
+    block_losses = train_block(
+                   ^^^^^^^^^^^^
+  File "/home/bill/Documents/WhetstoneAI_Fabricate/run_grokking_until.py", line 94, in train_block
+    raise RuntimeError(proc.stderr[-2000:] or proc.stdout[-2000:])
+RuntimeError: === Whetstone Fabricate ===
+steps=1000  batch_size=32  lr=0.000500  weight_decay=0.010000  hidden_dim=512  model=gpu_transformer  morphisms=binary .morf (fast)
+
+[1] Loading dataset...
+ProntoQADataset: 978 examples, vocab_size=1251, seq_len=96
+
+[2] ModelConfig
+    vocab_size=1251  embedding_dim=512  hidden_dim=512  layers=4  heads=1  ff_hidden_dim=2048  seq_len=96
+
+[3] Constructing model engine...
+    num_params=13933795
+    morphism binary: 106.3 MB/step  ->  106306.4 MB for 1000 steps
+
+[4] Output directory: /mnt/storage/fabricate_runs/whetstone_verification_type_large
+[5] --reset: removing checkpoint and output files.
+[6] TrajectoryWriter: /mnt/storage/fabricate_runs/whetstone_verification_type_large/train.fabr  (fresh)
+    Initial weights written.
+[7] MorphismWriter: /mnt/storage/fabricate_runs/whetstone_verification_type_large/morphisms.morf  (fresh)
+    Building fabricate.db (static tables only)...
+Initialising weight_metadata (13933795 params)...
+  weight_metadata populated.
+    Static tables ready.
+
+[8] Training (1000 update steps from step 0, batch_size=32, lr=0.000500)
+    step    loss          task
+    ------  ------------  ----
+    0       7.076128      0
+
diff --git a/specialists/runs/whetstone_worker_type_large.log b/specialists/runs/whetstone_worker_type_large.log
new file mode 100644
index 0000000..2b98b3a
--- /dev/null
+++ b/specialists/runs/whetstone_worker_type_large.log
@@ -0,0 +1,44 @@
+=== whetstone_worker_type (large) ===
+  train:  /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/worker_type_train.tsv
+  eval:   /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/worker_type_eval.tsv
+  model:  hidden=512 layers=4 heads=1
+
+=== run_grokking_until: whetstone_worker_type_large ===
+dataset=worker_type_train.tsv  lr=0.0005  wd=0.01  max_steps=20000
+Fresh start
+Traceback (most recent call last):
+  File "/home/bill/Documents/WhetstoneAI_Fabricate/run_grokking_until.py", line 299, in <module>
+    main()
+  File "/home/bill/Documents/WhetstoneAI_Fabricate/run_grokking_until.py", line 220, in main
+    block_losses = train_block(
+                   ^^^^^^^^^^^^
+  File "/home/bill/Documents/WhetstoneAI_Fabricate/run_grokking_until.py", line 94, in train_block
+    raise RuntimeError(proc.stderr[-2000:] or proc.stdout[-2000:])
+RuntimeError: === Whetstone Fabricate ===
+steps=1000  batch_size=32  lr=0.000500  weight_decay=0.010000  hidden_dim=512  model=gpu_transformer  morphisms=binary .morf (fast)
+
+[1] Loading dataset...
+ProntoQADataset: 300 examples, vocab_size=539, seq_len=96
+
+[2] ModelConfig
+    vocab_size=539  embedding_dim=512  hidden_dim=512  layers=4  heads=1  ff_hidden_dim=2048  seq_len=96
+
+[3] Constructing model engine...
+    num_params=13203995
+    morphism binary: 100.7 MB/step  ->  100738.5 MB for 1000 steps
+
+[4] Output directory: /mnt/storage/fabricate_runs/whetstone_worker_type_large
+[5] --reset: removing checkpoint and output files.
+[6] TrajectoryWriter: /mnt/storage/fabricate_runs/whetstone_worker_type_large/train.fabr  (fresh)
+    Initial weights written.
+[7] MorphismWriter: /mnt/storage/fabricate_runs/whetstone_worker_type_large/morphisms.morf  (fresh)
+    Building fabricate.db (static tables only)...
+Initialising weight_metadata (13203995 params)...
+  weight_metadata populated.
+    Static tables ready.
+
+[8] Training (1000 update steps from step 0, batch_size=32, lr=0.000500)
+    step    loss          task
+    ------  ------------  ----
+    0       6.984013      0
+
diff --git a/specialists/runs/worker_type.log b/specialists/runs/worker_type.log
new file mode 100644
index 0000000..d4124f6
--- /dev/null
+++ b/specialists/runs/worker_type.log
@@ -0,0 +1,18 @@
+Training worker_type specialist...
+  train: /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/worker_type_train.tsv
+  eval:  /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/worker_type_eval.tsv
+  out:   /mnt/storage/fabricate_runs/whetstone_worker_type
+  log:   /home/bill/Documents/CLionProjects/whetstone_DSL/specialists/runs/worker_type_training.log
+
+=== run_grokking_until: whetstone_worker_type ===
+dataset=worker_type_train.tsv  lr=0.001  wd=0.01  max_steps=5000
+Resuming from step 3000
+Resuming: grokking already detected at step 1000, 3 blocks since grokking.
+step=  4000  overall=88.5%  T=88.5%  F=88.5%  loss=0.0000  min=0.0000  delta=+7.7%  free=904GB <<< GROKKING (loss)
+Stopping 4 blocks after grokking (step 4000).
+
+=== DONE: whetstone_worker_type ===
+Final: step=4000  overall=88.5%
+Grokking at step 1000
+
+Done. Checkpoint at: /mnt/storage/fabricate_runs/whetstone_worker_type/checkpoint.bin
diff --git a/specialists/runs/worker_type_history.json b/specialists/runs/worker_type_history.json
new file mode 100644
index 0000000..a329bcc
--- /dev/null
+++ b/specialists/runs/worker_type_history.json
@@ -0,0 +1,48 @@
+{
+  "dataset": "/home/bill/Documents/CLionProjects/whetstone_DSL/specialists/data/generated/worker_type_train.tsv",
+  "lr": 0.001,
+  "weight_decay": 0.01,
+  "seed": "1234ABCD",
+  "history": [
+    {
+      "step": 1000,
+      "overall": 80.76923076923077,
+      "true_pct": 80.76923076923077,
+      "false_pct": 80.76923076923077,
+      "delta": 30.769230769230774,
+      "end_loss": 0.0004,
+      "min_loss": 0.000391,
+      "grokking": true
+    },
+    {
+      "step": 2000,
+      "overall": 80.76923076923077,
+      "true_pct": 80.76923076923077,
+      "false_pct": 80.76923076923077,
+      "delta": 0.0,
+      "end_loss": 0.000126,
+      "min_loss": 0.000124,
+      "grokking": true
+    },
+    {
+      "step": 3000,
+      "overall": 80.76923076923077,
+      "true_pct": 80.76923076923077,
+      "false_pct": 80.76923076923077,
+      "delta": 0.0,
+      "end_loss": 4.7e-05,
+      "min_loss": 4.7e-05,
+      "grokking": true
+    },
+    {
+      "step": 4000,
+      "overall": 88.46153846153845,
+      "true_pct": 88.46153846153845,
+      "false_pct": 88.46153846153845,
+      "delta": 7.692307692307679,
+      "end_loss": 2.1e-05,
+      "min_loss": 2e-05,
+      "grokking": true
+    }
+  ]
+}
diff --git a/specialists/runs/worker_type_training.log b/specialists/runs/worker_type_training.log
new file mode 100644
index 0000000..2c18ebb
--- /dev/null
+++ b/specialists/runs/worker_type_training.log
@@ -0,0 +1,10 @@
+=== run_grokking_until: whetstone_worker_type ===
+dataset=worker_type_train.tsv  lr=0.001  wd=0.01  max_steps=5000
+Resuming from step 3000
+Resuming: grokking already detected at step 1000, 3 blocks since grokking.
+step=  4000  overall=88.5%  T=88.5%  F=88.5%  loss=0.0000  min=0.0000  delta=+7.7%  free=904GB <<< GROKKING (loss)
+Stopping 4 blocks after grokking (step 4000).
+
+=== DONE: whetstone_worker_type ===
+Final: step=4000  overall=88.5%
+Grokking at step 1000
diff --git a/specialists/scripts/capacity_sweep.py b/specialists/scripts/capacity_sweep.py
new file mode 100644
index 0000000..85ba3c4
--- /dev/null
+++ b/specialists/scripts/capacity_sweep.py
@@ -0,0 +1,288 @@
+#!/usr/bin/env python3
+"""
+capacity_sweep.py
+
+Runs a staged capacity sweep for one specialist gate.
+Tiers: small+ (256/2) → medium (384/3) → large (512/4).
+
+For each tier:
+  1. Run a short pilot (3000 steps). If accuracy doesn't improve over
+     the previous tier's result, skip the full run.
+  2. Run the full training (up to --max-steps) only if pilot improved.
+  3. Evaluate with eval_specialist.py (confusion, confidence, thresholds).
+  4. Append result row to the sweep table.
+
+Usage:
+    python3 specialists/scripts/capacity_sweep.py \\
+        --name        worker_type \\
+        --train       specialists/data/generated/worker_type_train.tsv \\
+        --eval        specialists/data/generated/worker_type_eval.tsv \\
+        --labels      "implementer,reviewer,architect,qa" \\
+        --runs-dir    /mnt/storage/fabricate_runs \\
+        --results-out specialists/eval/results/worker_type_sweep.json
+
+    # Run from whetstone_DSL root.
+"""
+
+import argparse
+import json
+import os
+import shutil
+import subprocess
+import sys
+import time
+from pathlib import Path
+
+ROOT        = Path(__file__).parent.parent.parent   # whetstone_DSL root
+FABRICATE   = Path("/home/bill/Documents/WhetstoneAI_Fabricate")
+VENV_PYTHON   = FABRICATE / ".venv" / "bin" / "python3"
+TRAIN_SCRIPT  = ROOT / "specialists" / "scripts" / "train_specialist_pt.py"
+EVAL_SCRIPT   = ROOT / "specialists" / "eval" / "eval_specialist_pt.py"
+
+TIERS = [
+    {"name": "small_plus", "hidden": 256, "layers": 2, "heads": 4},
+    {"name": "medium",     "hidden": 384, "layers": 3, "heads": 6},
+    {"name": "large",      "hidden": 512, "layers": 4, "heads": 8},
+]
+
+PILOT_STEPS   = 3000   # steps for pilot run
+FULL_STEPS    = 15000  # max steps for full run
+BLOCK         = 1000   # steps per eval block
+BATCH_SIZE    = 256
+LR            = 0.001
+WD            = 0.01
+
+
+def log(msg):
+    print(msg, flush=True)
+
+
+def run_training(run_dir: Path, train_data: Path, eval_data: Path,
+                 labels: str, hidden: int, layers: int, heads: int,
+                 max_steps: int, reset: bool, history_out: Path) -> bool:
+    """Invoke train_specialist_pt.py. Returns True on success."""
+    run_dir.mkdir(parents=True, exist_ok=True)
+    if reset:
+        for f in run_dir.glob("*"):
+            f.unlink()
+        history_out.unlink(missing_ok=True)
+
+    cmd = [
+        str(VENV_PYTHON), str(TRAIN_SCRIPT),
+        "--dataset",               str(train_data),
+        "--eval-dataset",          str(eval_data),
+        "--out-dir",               str(run_dir),
+        "--labels",                labels,
+        "--lr",                    str(LR),
+        "--weight-decay",          str(WD),
+        "--max-steps",             str(max_steps),
+        "--hidden-dim",            str(hidden),
+        "--layers",                str(layers),
+        "--heads",                 str(heads),
+        "--batch-size",            str(BATCH_SIZE),
+        "--grok-loss-threshold",   "0.05",
+        "--grok-acc-jump",         "10.0",
+        "--stop-after-grokking-blocks", "4",
+        "--history-out",           str(history_out),
+        "--reset",
+    ]
+
+    result = subprocess.run(cmd, capture_output=False)
+    return result.returncode == 0
+
+
+def run_eval(run_dir: Path, eval_data: Path, labels: str) -> dict | None:
+    """Run eval_specialist_pt.py on the checkpoint. Returns parsed JSON or None."""
+    ck = run_dir / "checkpoint.pt"
+    if not ck.exists():
+        return None
+
+    result = subprocess.run(
+        [str(VENV_PYTHON), str(EVAL_SCRIPT),
+         "--checkpoint", str(ck),
+         "--dataset",    str(eval_data),
+         "--labels",     labels,
+         "--quiet"],
+        capture_output=True, text=True,
+    )
+    if result.returncode != 0:
+        log(f"  [WARN] eval failed: {result.stderr[-500:]}")
+        return None
+    try:
+        # Last line should be the JSON dump
+        last_line = result.stdout.strip().splitlines()[-1]
+        return json.loads(last_line)
+    except (json.JSONDecodeError, IndexError):
+        log(f"  [WARN] could not parse eval JSON")
+        return None
+
+
+def read_history(history_path: Path) -> dict:
+    if not history_path.exists():
+        return {}
+    return json.loads(history_path.read_text())
+
+
+def get_best_accuracy(history: dict) -> float:
+    entries = history.get("history", [])
+    if not entries:
+        return 0.0
+    return max(e.get("overall", 0.0) for e in entries)
+
+
+def suggest_action(eval_result: dict, prev_accuracy: float) -> str:
+    if eval_result is None:
+        return "training_failed"
+    acc = eval_result.get("overall_accuracy", 0.0)
+    verdict = eval_result.get("verdict", "unknown")
+
+    delta = acc - prev_accuracy
+    if delta < 1.0:
+        return "no_gain_stop_scaling"
+    return verdict
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--name",        required=True, help="Gate name, e.g. worker_type")
+    ap.add_argument("--train",       required=True, help="Training TSV path")
+    ap.add_argument("--eval",        required=True, help="Eval TSV path")
+    ap.add_argument("--labels",      required=True, help="Comma-sep label tokens")
+    ap.add_argument("--runs-dir",    default="/mnt/storage/fabricate_runs")
+    ap.add_argument("--results-out", default=None,  help="JSON output path")
+    ap.add_argument("--tiers",       default="small_plus,medium,large",
+                    help="Comma-sep tier names to run (subset of small_plus,medium,large)")
+    ap.add_argument("--pilot-only",  action="store_true",
+                    help="Only run pilots, no full training runs")
+    args = ap.parse_args()
+
+    train_data  = Path(args.train)
+    eval_data   = Path(args.eval)
+    runs_base   = Path(args.runs_dir)
+    active_tiers = [t for t in TIERS if t["name"] in args.tiers.split(",")]
+
+    sweep_results = {
+        "gate": args.name,
+        "labels": args.labels,
+        "train": str(train_data),
+        "eval": str(eval_data),
+        "tiers": [],
+    }
+
+    prev_accuracy = 0.0  # baseline for gain comparison
+    results_dir = ROOT / "specialists" / "eval" / "results"
+    results_dir.mkdir(parents=True, exist_ok=True)
+
+    for tier in active_tiers:
+        tname   = tier["name"]
+        hidden  = tier["hidden"]
+        layers  = tier["layers"]
+        heads   = tier.get("heads", 4)
+        run_id  = f"whetstone_{args.name}_{tname}"
+        run_dir = runs_base / run_id
+        history = run_dir / "history.json"
+        tier_result_path = results_dir / f"{args.name}_{tname}.json"
+
+        log(f"\n{'='*60}")
+        log(f"  Tier: {tname}  hidden={hidden} layers={layers} heads={heads}")
+        log(f"  Run dir: {run_dir}")
+        log(f"{'='*60}")
+
+        tier_row = {
+            "tier": tname,
+            "hidden": hidden,
+            "layers": layers,
+            "heads": heads,
+            "model_size_approx": f"~{hidden*hidden*layers*12 // 1_000_000}M params",
+        }
+
+        # --- Pilot run ---
+        log(f"\n  [pilot] {PILOT_STEPS} steps...")
+        t0 = time.perf_counter()
+        ok = run_training(run_dir, train_data, eval_data, args.labels,
+                          hidden, layers, heads, PILOT_STEPS, reset=True, history_out=history)
+        pilot_time = time.perf_counter() - t0
+        pilot_history = read_history(history)
+        pilot_acc = get_best_accuracy(pilot_history)
+
+        log(f"  [pilot] done in {pilot_time:.0f}s  best={pilot_acc:.1f}%")
+
+        if not ok or pilot_acc == 0.0:
+            log(f"  [pilot] FAILED or no accuracy recorded — skipping tier")
+            tier_row["pilot_accuracy"] = None
+            tier_row["action"] = "pilot_failed"
+            sweep_results["tiers"].append(tier_row)
+            continue
+
+        tier_row["pilot_accuracy"] = round(pilot_acc, 1)
+        tier_row["pilot_time_s"] = round(pilot_time, 0)
+
+        gain = pilot_acc - prev_accuracy
+        log(f"  [pilot] gain vs prev tier: {gain:+.1f}pp")
+
+        if gain < 1.0 and prev_accuracy > 0:
+            log(f"  [pilot] <1pp gain — skipping full run for this tier")
+            tier_row["action"] = "no_gain_stop_scaling"
+            sweep_results["tiers"].append(tier_row)
+            break
+
+        if args.pilot_only:
+            log(f"  [pilot-only mode] evaluating pilot checkpoint...")
+        else:
+            # --- Full run ---
+            log(f"\n  [full] continuing to {FULL_STEPS} steps...")
+            t1 = time.perf_counter()
+            run_training(run_dir, train_data, eval_data, args.labels,
+                         hidden, layers, heads, FULL_STEPS, reset=False, history_out=history)
+            full_time = time.perf_counter() - t1
+            full_history = read_history(history)
+            full_acc = get_best_accuracy(full_history)
+            log(f"  [full] done in {full_time:.0f}s  best={full_acc:.1f}%")
+            tier_row["full_accuracy"] = round(full_acc, 1)
+            tier_row["full_time_s"] = round(full_time, 0)
+
+        # --- Eval ---
+        log(f"\n  [eval] running extended evaluation...")
+        eval_result = run_eval(run_dir, eval_data, args.labels)
+        if eval_result:
+            # Save per-tier eval result
+            tier_result_path.write_text(json.dumps(eval_result, indent=2))
+            log(f"  [eval] accuracy={eval_result['overall_accuracy']}%  "
+                f"verdict={eval_result['verdict']}")
+
+            tier_row["eval_accuracy"] = eval_result["overall_accuracy"]
+            tier_row["per_class"] = eval_result.get("per_class", {})
+            tier_row["confusion_summary"] = eval_result.get("confusion_summary", [])[:5]
+            tier_row["threshold_analysis"] = eval_result.get("threshold_analysis", [])
+            tier_row["guardrail_threshold"] = eval_result.get("guardrail_threshold")
+            tier_row["guardrail_abstain_rate"] = eval_result.get("guardrail_abstain_rate")
+            tier_row["eval_latency_ms"] = eval_result.get("eval_latency_ms_per_example")
+            tier_row["verdict"] = eval_result["verdict"]
+            tier_row["action"] = suggest_action(eval_result, prev_accuracy)
+            tier_row["eval_result_path"] = str(tier_result_path)
+
+            prev_accuracy = eval_result["overall_accuracy"]
+        else:
+            tier_row["action"] = "eval_failed"
+
+        sweep_results["tiers"].append(tier_row)
+
+    # --- Final table ---
+    log(f"\n{'='*60}")
+    log(f"  SWEEP RESULTS: {args.name}")
+    log(f"{'='*60}")
+    header = f"  {'tier':12s}  {'hidden/L':8s}  {'pilot':7s}  {'full':7s}  {'verdict'}"
+    log(header)
+    for row in sweep_results["tiers"]:
+        pilot = f"{row.get('pilot_accuracy','—'):>5}" if row.get('pilot_accuracy') else "  —  "
+        full  = f"{row.get('eval_accuracy','—'):>5}" if row.get('eval_accuracy') else "  —  "
+        hl    = f"{row['hidden']}/{row['layers']}"
+        log(f"  {row['tier']:12s}  {hl:8s}  {pilot}%  {full}%  {row.get('verdict','—')}")
+
+    out = args.results_out or str(results_dir / f"{args.name}_sweep.json")
+    Path(out).write_text(json.dumps(sweep_results, indent=2))
+    log(f"\nFull results → {out}")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/specialists/scripts/gen_automatability_data.py b/specialists/scripts/gen_automatability_data.py
new file mode 100644
index 0000000..91aa048
--- /dev/null
+++ b/specialists/scripts/gen_automatability_data.py
@@ -0,0 +1,293 @@
+#!/usr/bin/env python3
+"""
+gen_automatability_data.py
+
+Labels the editor/src/*.h headers to produce training data for the
+automatability_strategy specialist.
+
+The six tiers (matching AutomatabilityAnnotation.strategy in Annotation.h):
+
+  deterministic  — Same input always produces same output. Pure formula,
+                   rule engine, or algorithmic transform. Zero parameters.
+                   Examples: validators, scorers, parsers, extractors, trackers.
+
+  template       — Deterministic composition of known structural patterns.
+                   Code generators driven by fixed templates; no ambiguity.
+                   Examples: code generators, builders, highlighters, register tools.
+
+  specialist     — Bounded output vocabulary (fixed N classes). Probabilistic
+                   but constrained. ~800KB model, microseconds.
+                   Examples: selectors with uncertainty, judges, advisors,
+                   annotators that make probabilistic choices.
+
+  slm            — Open vocabulary but local-scale. Complex transformation
+                   or interpretation where output tokens vary but scope is
+                   bounded. ~1-7GB.
+                   Examples: transpilers, type translators, optimizers.
+
+  llm            — Full generative reasoning. Cross-context, novel output,
+                   planning-level decisions.
+                   Examples: architect intake, decomposers, agent code gen,
+                   high-level orchestrators.
+
+  human          — Not yet formalizable. Requires manual judgment, approval,
+                   or editorial oversight.
+                   Examples: mutation approval, review surfaces, escalation.
+
+Input text format (what the specialist sees at inference time):
+  "name=ClassName"
+
+This mirrors what Claude provides when calling add_skeleton_node — the
+class name is the primary signal. The Whetstone naming convention
+enforces semantic suffixes, making the name a strong predictor.
+
+Output: TSV with columns: label<TAB>hops<TAB>text
+  label: integer index 0–5
+         0=deterministic  1=template  2=specialist  3=slm  4=llm  5=human
+  hops:  always 0
+  text:  "name=ClassName"
+
+Usage:
+    python3 specialists/scripts/gen_automatability_data.py \\
+        --src-dir editor/src \\
+        --train-out specialists/data/generated/automatability_train.tsv \\
+        --eval-out  specialists/data/generated/automatability_eval.tsv
+"""
+import re
+import sys
+import random
+import argparse
+from pathlib import Path
+from collections import Counter
+
+# ---------------------------------------------------------------------------
+# Classification heuristics — suffix and prefix patterns on class names
+# ---------------------------------------------------------------------------
+#
+# Priority order: human > llm > slm > specialist > template > deterministic
+# (more specific / rarer tiers checked first to avoid being swallowed by
+# the large deterministic default)
+
+# human: requires manual judgment, approval workflows
+HUMAN_SUFFIXES = (
+    "Approval", "ReviewInterface", "ReviewSurface", "MutationPreview",
+    "EscalationPlanner", "ApprovalGate",
+)
+HUMAN_SUBSTRINGS = (
+    "Approval", "HumanReview", "ManualGate", "ReviewBoard",
+)
+
+# llm: high-level reasoning, planning, novel generation
+LLM_PREFIXES = (
+    "Agent",          # AgentCodeGen, AgentMutationApproval (covered by human above)
+)
+LLM_SUFFIXES = (
+    "IntakeProcessor", "Decomposer", "ProjectPipeline", "ModuleDecomposer",
+    "ProblemParser", "MultiLanguageOrchestrator", "WorkflowOrchestrator",
+    "PolyglotOrchestrator", "PolyglotSuiteOrchestrator",
+    "TechStackSelector",   # open-ended tech selection = llm (vs bounded = specialist)
+    "ScaffoldGenerator",   # spec-to-scaffold = llm-driven
+)
+LLM_SUBSTRINGS = (
+    "Planner", "Strategist",
+)
+
+# slm: complex transformation, open output vocab but local scope
+SLM_SUFFIXES = (
+    "Transpiler", "TypeSystemTranslator", "TypeAwareMappings",
+    "TransformEngine", "TransformEngineExtended",
+    "Optimizer", "Lowering",
+)
+SLM_SUBSTRINGS = (
+    "Transpil", "Transform",
+    "Translat",   # TranslationReport is excluded below via DETERMINISTIC_OVERRIDES
+)
+
+# Names that would otherwise match a higher tier but are actually deterministic.
+# Checked before any tier classification.
+DETERMINISTIC_OVERRIDES = {
+    "TranslationReport",      # report doc, not a transformer
+    "DWARFBoundaryAnnotator", # rule-based DWARF annotation
+    "IntakeTextUtil",         # text utility, not an intake processor
+}
+
+# specialist: bounded probabilistic — selectors, judges, advisors, annotators
+SPECIALIST_SUFFIXES = (
+    "Selector", "Judge", "Advisor", "Recommender",
+    "Annotator", "Predictor", "Classifier",
+)
+SPECIALIST_SUBSTRINGS = (
+    "Selector", "Judge", "Advisor",
+)
+
+# template: code/output generation from fixed patterns
+TEMPLATE_PREFIXES = (
+    "Register",        # RegisterXxxTools.h — tool registration templates
+    "SyntaxHighlighter",
+)
+TEMPLATE_SUFFIXES = (
+    "Generator", "Builder", "Factory", "Templates", "Highlighter",
+    "Emitter", "Renderer", "Serializer", "Encoder", "Composer",
+    "Formatter",       # output formatters = deterministic, but code formatters = template
+    "CodeGen",
+)
+TEMPLATE_SUBSTRINGS = (
+    "Generator", "Builder", "Templates",
+)
+
+# deterministic: pure computation — everything else
+# (validators, scorers, parsers, extractors, trackers, monitors, etc.)
+
+
+LABEL_ORDER = ["deterministic", "template", "specialist", "slm", "llm", "human"]
+LABEL_INDEX  = {lbl: i for i, lbl in enumerate(LABEL_ORDER)}
+
+
+def classify(class_name: str) -> str:
+    n = class_name  # keep original case for prefix/suffix matching
+
+    # Hard overrides — names that pattern-match a higher tier but are deterministic
+    if n in DETERMINISTIC_OVERRIDES:
+        return "deterministic"
+
+    # human — must check before llm (some overlap on Agent* names)
+    for suf in HUMAN_SUFFIXES:
+        if n.endswith(suf):
+            return "human"
+    for sub in HUMAN_SUBSTRINGS:
+        if sub in n:
+            return "human"
+
+    # llm — planning-level, open-ended
+    for pre in LLM_PREFIXES:
+        if n.startswith(pre):
+            return "llm"
+    for suf in LLM_SUFFIXES:
+        if n.endswith(suf):
+            return "llm"
+    for sub in LLM_SUBSTRINGS:
+        if sub in n:
+            return "llm"
+
+    # slm — complex transformation
+    for suf in SLM_SUFFIXES:
+        if n.endswith(suf):
+            return "slm"
+    for sub in SLM_SUBSTRINGS:
+        if sub in n:
+            return "slm"
+
+    # specialist — bounded probabilistic decision
+    for suf in SPECIALIST_SUFFIXES:
+        if n.endswith(suf):
+            return "specialist"
+    for sub in SPECIALIST_SUBSTRINGS:
+        if sub in n:
+            return "specialist"
+
+    # template — code/output generation
+    for pre in TEMPLATE_PREFIXES:
+        if n.startswith(pre):
+            return "template"
+    for suf in TEMPLATE_SUFFIXES:
+        if n.endswith(suf):
+            return "template"
+    for sub in TEMPLATE_SUBSTRINGS:
+        if sub in n:
+            return "template"
+
+    # deterministic — default
+    return "deterministic"
+
+
+def header_to_class_name(path: Path) -> str:
+    """Convert FileName.h → FileName (strip extension)."""
+    return path.stem
+
+
+# ---------------------------------------------------------------------------
+# Main
+# ---------------------------------------------------------------------------
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--src-dir", default="editor/src",
+                    help="Directory containing *.h headers (non-recursive)")
+    ap.add_argument("--train-out", default="specialists/data/generated/automatability_train.tsv")
+    ap.add_argument("--eval-out",  default="specialists/data/generated/automatability_eval.tsv")
+    ap.add_argument("--eval-frac", type=float, default=0.15)
+    ap.add_argument("--seed", type=int, default=42)
+    ap.add_argument("--show-labels", action="store_true",
+                    help="Print each header and its assigned label (for review)")
+    args = ap.parse_args()
+
+    src_dir = Path(args.src_dir)
+    headers = sorted(src_dir.glob("*.h"))
+    if not headers:
+        print(f"No .h files found in {src_dir}", file=sys.stderr)
+        sys.exit(1)
+    print(f"Found {len(headers)} header files")
+
+    all_examples: list[tuple[str, str]] = []
+    for h in headers:
+        class_name = header_to_class_name(h)
+        label = classify(class_name)
+        all_examples.append((label, class_name))
+        if args.show_labels:
+            print(f"  {label:15s}  {class_name}")
+
+    counts = Counter(lbl for lbl, _ in all_examples)
+    print("Raw class distribution:", dict(counts))
+
+    # Balance: use the median class size as target so we keep more deterministic
+    # examples (the dominant real-world case) while still oversampling sparse tiers.
+    sorted_counts = sorted(counts.values())
+    median_count = sorted_counts[len(sorted_counts) // 2]
+    target = min(median_count * 2, max(sorted_counts))
+    target = max(target, 20)
+    print(f"Balancing to {target} examples per class")
+
+    rng = random.Random(args.seed)
+    by_class: dict[str, list[str]] = {}
+    for lbl, name in all_examples:
+        by_class.setdefault(lbl, []).append(name)
+
+    balanced: list[tuple[str, str]] = []
+    for lbl in LABEL_ORDER:
+        names = by_class.get(lbl, [])
+        if not names:
+            print(f"WARNING: no examples for class '{lbl}'", file=sys.stderr)
+            continue
+        rng.shuffle(names)
+        chosen = names[:target]
+        while len(chosen) < target:
+            chosen += names
+        balanced.extend((lbl, name) for name in chosen[:target])
+
+    rng.shuffle(balanced)
+    print(f"Balanced total: {len(balanced)} examples")
+
+    n_eval = max(1, int(len(balanced) * args.eval_frac))
+    eval_set  = balanced[:n_eval]
+    train_set = balanced[n_eval:]
+
+    Path(args.train_out).parent.mkdir(parents=True, exist_ok=True)
+    Path(args.eval_out).parent.mkdir(parents=True, exist_ok=True)
+
+    def write_tsv(path, rows):
+        with open(path, "w") as f:
+            for lbl, name in rows:
+                idx = LABEL_INDEX[lbl]
+                f.write(f"{idx}\t0\tname={name}\n")
+        print(f"Wrote {len(rows)} rows → {path}")
+
+    write_tsv(args.train_out, train_set)
+    write_tsv(args.eval_out,  eval_set)
+
+    for split_name, split in [("train", train_set), ("eval", eval_set)]:
+        c = Counter(lbl for lbl, _ in split)
+        print(f"  {split_name}: {dict(sorted(c.items()))}")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/specialists/scripts/gen_confidence_tier_data.py b/specialists/scripts/gen_confidence_tier_data.py
new file mode 100644
index 0000000..7053459
--- /dev/null
+++ b/specialists/scripts/gen_confidence_tier_data.py
@@ -0,0 +1,130 @@
+#!/usr/bin/env python3
+"""
+gen_confidence_tier_data.py
+
+Generate synthetic training data for the confidence_tier specialist.
+
+The Whetstone confidence formula (from TaskitemConfidenceAmbiguity.h):
+  base = 85
+  - 25 if not queueReady
+  - 10 per ambiguous requirement (up to 3)
+  - 8 per conflict detected
+  - 5 if prereq_count > 2
+  - 4 if has_dependencies
+  clamped to [0, 100]
+
+Tier labels:
+  0 = high    (confidence >= 80)
+  1 = medium  (confidence 60–79)
+  2 = low     (confidence < 60)
+
+The specialist encodes features as a short token sequence:
+  "conflicts=N ambiguity=M deps=D prereqs=P queueready=B"
+
+Usage:
+    python3 specialists/scripts/gen_confidence_tier_data.py \
+        --train-out specialists/data/generated/confidence_tier_train.tsv \
+        --eval-out  specialists/data/generated/confidence_tier_eval.tsv
+"""
+import random
+import argparse
+from pathlib import Path
+from collections import Counter
+
+LABEL_ORDER = ["high", "medium", "low"]
+LABEL_INDEX = {lbl: i for i, lbl in enumerate(LABEL_ORDER)}
+
+
+def compute_confidence(conflicts, ambiguity, deps, prereqs, queue_ready):
+    conf = 85
+    if not queue_ready:
+        conf -= 25
+    conf -= min(ambiguity, 3) * 10
+    conf -= conflicts * 8
+    if prereqs > 2:
+        conf -= 5
+    if deps > 0:
+        conf -= 4
+    return max(0, min(100, conf))
+
+
+def tier(conf):
+    if conf >= 80:
+        return "high"
+    elif conf >= 60:
+        return "medium"
+    else:
+        return "low"
+
+
+def encode_input(conflicts, ambiguity, deps, prereqs, queue_ready):
+    qr = "yes" if queue_ready else "no"
+    return f"conflicts={conflicts} ambiguity={ambiguity} deps={deps} prereqs={prereqs} queueready={qr}"
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--train-out", default="specialists/data/generated/confidence_tier_train.tsv")
+    ap.add_argument("--eval-out",  default="specialists/data/generated/confidence_tier_eval.tsv")
+    ap.add_argument("--n-per-class", type=int, default=400)
+    ap.add_argument("--seed", type=int, default=42)
+    args = ap.parse_args()
+
+    rng = random.Random(args.seed)
+
+    # Generate all unique input combinations and label them
+    seen = set()
+    by_tier: dict[str, list[str]] = {"high": [], "medium": [], "low": []}
+
+    for conflicts in range(0, 10):
+        for ambiguity in range(0, 4):
+            for deps in range(0, 3):
+                for prereqs in range(1, 5):
+                    for queue_ready in (True, False):
+                        key = (conflicts, ambiguity, deps, prereqs, queue_ready)
+                        if key in seen:
+                            continue
+                        seen.add(key)
+                        conf = compute_confidence(*key)
+                        t = tier(conf)
+                        text = encode_input(*key)
+                        by_tier[t].append(text)
+
+    print("Raw tier distribution:", {k: len(v) for k, v in by_tier.items()})
+
+    # Balance to n_per_class (oversample if needed)
+    target = args.n_per_class
+    balanced = []
+    for lbl, examples in by_tier.items():
+        rng.shuffle(examples)
+        chosen = examples[:]
+        while len(chosen) < target:
+            chosen.extend(examples)
+        chosen = chosen[:target]
+        balanced.extend((lbl, ex) for ex in chosen)
+
+    rng.shuffle(balanced)
+
+    n_eval = max(1, int(len(balanced) * 0.15))
+    eval_set  = balanced[:n_eval]
+    train_set = balanced[n_eval:]
+
+    Path(args.train_out).parent.mkdir(parents=True, exist_ok=True)
+    Path(args.eval_out).parent.mkdir(parents=True, exist_ok=True)
+
+    def write_tsv(path, rows):
+        with open(path, "w") as f:
+            for lbl, text in rows:
+                f.write(f"{LABEL_INDEX[lbl]}\t0\t{text}\n")
+        print(f"Wrote {len(rows)} rows → {path}")
+
+    write_tsv(args.train_out, train_set)
+    write_tsv(args.eval_out,  eval_set)
+
+    for split_name, split in [("train", train_set), ("eval", eval_set)]:
+        c = Counter(lbl for lbl, _ in split)
+        print(f"  {split_name}: {dict(sorted(c.items()))}")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/specialists/scripts/gen_from_projects.py b/specialists/scripts/gen_from_projects.py
new file mode 100644
index 0000000..6fb6b31
--- /dev/null
+++ b/specialists/scripts/gen_from_projects.py
@@ -0,0 +1,332 @@
+#!/usr/bin/env python3
+"""
+gen_from_projects.py
+
+Runs all project benchmark specs through the whetstone_mcp pipeline
+(architect_intake → generate_taskitems) and extracts labeled training
+examples for each specialist from the real pipeline output.
+
+Why this matters:
+  - Training data from sprint plans uses step description text.
+  - Specialists run at inference time on taskitem TITLES + INTENT.
+  - These are similar but not identical formats. Running real specs through
+    the pipeline produces training examples in the exact inference-time format.
+  - The pipeline output is the "correct answer" — workerType from inferWorkerType,
+    prerequisiteOps from inferPrerequisiteOps, verificationType from the contract.
+
+Input files:
+  datasets/project_benchmarks/common_projects_100.jsonl       (100 projects)
+  datasets/project_benchmarks/challenging_*.jsonl             (25 projects)
+
+Output TSVs (appended to existing data or written fresh):
+  specialists/data/generated/prereq_op_projects_train.tsv
+  specialists/data/generated/prereq_op_projects_eval.tsv
+  specialists/data/generated/worker_type_projects_train.tsv
+  specialists/data/generated/worker_type_projects_eval.tsv
+  specialists/data/generated/verification_type_projects_train.tsv
+  specialists/data/generated/verification_type_projects_eval.tsv
+
+Text format for all specialists:
+  "<title> | <intent[:120]>"
+  Using pipe separator so title and intent are both visible but the
+  model sees them as one continuous text sequence.
+
+Usage:
+    python3 specialists/scripts/gen_from_projects.py \\
+        --whetstone-bin editor/build-native/whetstone_mcp \\
+        --workspace .
+"""
+import argparse
+import json
+import random
+import subprocess
+import sys
+from collections import Counter
+from pathlib import Path
+
+# ---------------------------------------------------------------------------
+# Label maps
+# ---------------------------------------------------------------------------
+
+# Text-based prereq classifier — same tokens as gen_prereq_op_data.py so that
+# project-spec task text is labeled consistently with the sprint-plan data.
+# Using text rather than the C++ heuristic's prerequisiteOps because the heuristic
+# never emits "manual-approval" (circular labels = all "standard").
+_PREREQ_REVIEW_TOKENS = {
+    "introduce", "phase", "framework", "architecture", "design", "refactor",
+    "planning", "taxonomy", "grammar", "interface", "api surface", "subsystem",
+    "extension point", "abstraction", "decompos", "scaffold", "polyglot",
+    "cross-language", "cross-project", "multi-language", "orchestrat",
+    "migration plan", "phase plan", "protocol design", "new layer",
+    "foundation", "bootstrap", "initial", "phase 1", "phase 2", "phase 3",
+    "sprint plan", "roadmap",
+}
+_PREREQ_APPROVAL_TOKENS = {
+    "release", "deploy", "publish", "migrate", "security", "compliance",
+    "upgrade", "breaking change", "production", "rollout", "finalize",
+    "sign off", "certification", "approval", "cutover", "go-live",
+    "data migration", "schema migration", "breaking", "deprecat",
+    "audit", "regulatory", "gdpr", "pii", "secret", "credential", "vault",
+    "access control", "rbac", "acl", "permission", "fraud", "encryption",
+}
+
+def prereq_ops_to_class(ops: list[str]) -> str:
+    """Legacy: classify by actual ops list (unused — text_to_prereq_class is preferred)."""
+    s = set(ops)
+    has_review   = "architect-review" in s
+    has_approval = "manual-approval" in s
+    if has_review and has_approval:
+        return "full_gates"
+    if has_review:
+        return "needs_review"
+    if has_approval:
+        return "needs_approval"
+    return "standard"
+
+def text_to_prereq_class(text: str) -> str:
+    """Label prereq class from task title+intent text (same signal the specialist sees)."""
+    t = text.lower()
+    has_review   = any(tok in t for tok in _PREREQ_REVIEW_TOKENS)
+    has_approval = any(tok in t for tok in _PREREQ_APPROVAL_TOKENS)
+    if has_review and has_approval:
+        return "full_gates"
+    if has_review:
+        return "needs_review"
+    if has_approval:
+        return "needs_approval"
+    return "standard"
+
+PREREQ_LABEL_INDEX   = {"standard": 0, "needs_review": 1, "needs_approval": 2, "full_gates": 3}
+WORKER_LABEL_INDEX   = {"implementer": 0, "reviewer": 1, "architect": 2, "qa": 3}
+VERIF_LABEL_INDEX    = {"unit": 0, "integration": 1, "schema": 2, "smoke": 3, "docs": 4}
+
+# ---------------------------------------------------------------------------
+# MCP pipeline helpers
+# ---------------------------------------------------------------------------
+
+def mcp_call(binary: str, workspace: str, tool: str, args: dict) -> dict:
+    payload = json.dumps({
+        "jsonrpc": "2.0", "id": 1,
+        "method": "tools/call",
+        "params": {"name": tool, "arguments": args}
+    })
+    try:
+        r = subprocess.run(
+            [binary, "--workspace", workspace, "--language", "cpp"],
+            input=payload, capture_output=True, text=True, timeout=20
+        )
+        content = (json.loads(r.stdout)
+                   .get("result", {})
+                   .get("content", [{}])[0]
+                   .get("text", ""))
+        return json.loads(content) if content else {}
+    except Exception as e:
+        print(f"  [WARN] mcp_call {tool}: {e}", file=sys.stderr)
+        return {}
+
+
+def run_pipeline(binary: str, workspace: str, spec: str) -> list[dict]:
+    """Run architect_intake → generate_taskitems for one spec. Return task list."""
+    intake = mcp_call(binary, workspace, "whetstone_architect_intake", {"markdown": spec})
+    reqs = intake.get("normalizedRequirements", [])
+    if not reqs:
+        return []
+    result = mcp_call(binary, workspace, "whetstone_generate_taskitems", {
+        "normalizedRequirements": reqs,
+        "conflicts": intake.get("conflicts", [])
+    })
+    return result.get("tasks", [])
+
+
+def task_text(task: dict) -> str:
+    """Combine title + intent for specialist input."""
+    title  = task.get("title", "").strip()
+    intent = task.get("intent", "").strip()[:120]
+    if intent and intent != title:
+        return f"{title} | {intent}"
+    return title
+
+
+# ---------------------------------------------------------------------------
+# Load project specs
+# ---------------------------------------------------------------------------
+
+def load_projects(dataset_dir: Path) -> list[dict]:
+    projects = []
+    # Common 100
+    common = dataset_dir / "common_projects_100.jsonl"
+    if common.exists():
+        with open(common) as f:
+            for line in f:
+                projects.append(json.loads(line))
+    # All challenging sets
+    for path in sorted(dataset_dir.glob("challenging_*.jsonl")):
+        with open(path) as f:
+            for line in f:
+                p = json.loads(line)
+                if p.get("spec"):
+                    projects.append(p)
+    print(f"Loaded {len(projects)} project specs")
+    return projects
+
+
+def spec_markdown(project: dict) -> str:
+    """Wrap a project spec into the sections architect_intake expects."""
+    return (
+        f"## Goals\n{project['spec']}\n\n"
+        f"## Constraints\n- Production quality\n- Well-tested\n\n"
+        f"## Acceptance Criteria\n- All core features implemented\n"
+        f"- Tests passing\n"
+    )
+
+
+# ---------------------------------------------------------------------------
+# Balance + write TSV
+# ---------------------------------------------------------------------------
+
+def balance_and_split(examples: list[tuple[str, str]],
+                      label_index: dict[str, int],
+                      label_order: list[str],
+                      seed: int = 42,
+                      eval_frac: float = 0.15,
+                      min_per_class: int = 20) -> tuple[list, list]:
+    """Return (train_rows, eval_rows) as list of (label_str, text) pairs."""
+    by_class: dict[str, list[str]] = {}
+    for lbl, text in examples:
+        by_class.setdefault(lbl, []).append(text)
+
+    counts = {lbl: len(items) for lbl, items in by_class.items()}
+    print(f"  Raw class distribution: {counts}")
+
+    if not counts:
+        return [], []
+
+    sorted_vals = sorted(counts.values())
+    median = sorted_vals[len(sorted_vals) // 2]
+    target = max(median, min_per_class)
+    print(f"  Balancing to {target} per class")
+
+    rng = random.Random(seed)
+    balanced: list[tuple[str, str]] = []
+    for lbl in label_order:
+        items = by_class.get(lbl, [])
+        if not items:
+            print(f"  WARNING: no examples for '{lbl}'", file=sys.stderr)
+            continue
+        rng.shuffle(items)
+        chosen = items[:target]
+        while len(chosen) < target:
+            chosen += items
+        balanced.extend((lbl, t) for t in chosen[:target])
+
+    rng.shuffle(balanced)
+    n_eval = max(1, int(len(balanced) * eval_frac))
+    return balanced[n_eval:], balanced[:n_eval]
+
+
+def write_tsv(path: Path, rows: list[tuple[str, str]], label_index: dict[str, int]) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    with open(path, "w") as f:
+        for lbl, text in rows:
+            idx = label_index[lbl]
+            f.write(f"{idx}\t0\t{text}\n")
+    print(f"  Wrote {len(rows)} rows → {path}")
+
+
+# ---------------------------------------------------------------------------
+# Main
+# ---------------------------------------------------------------------------
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--whetstone-bin", default="editor/build-native/whetstone_mcp")
+    ap.add_argument("--workspace",     default=".")
+    ap.add_argument("--dataset-dir",   default="datasets/project_benchmarks")
+    ap.add_argument("--out-dir",       default="specialists/data/generated")
+    ap.add_argument("--eval-frac", type=float, default=0.15)
+    ap.add_argument("--seed",      type=int,   default=42)
+    ap.add_argument("--dry-run",   action="store_true",
+                    help="Print extracted examples without writing files")
+    args = ap.parse_args()
+
+    binary    = str(Path(args.workspace) / args.whetstone_bin)
+    workspace = str(Path(args.workspace).resolve())
+    out_dir   = Path(args.out_dir)
+
+    projects = load_projects(Path(args.workspace) / args.dataset_dir)
+
+    prereq_examples:  list[tuple[str, str]] = []
+    worker_examples:  list[tuple[str, str]] = []
+    verif_examples:   list[tuple[str, str]] = []
+
+    for i, project in enumerate(projects):
+        print(f"[{i+1:3}/{len(projects)}] {project.get('project','?')} ({project.get('category','?')})")
+        spec_md = spec_markdown(project)
+        tasks = run_pipeline(binary, workspace, spec_md)
+        if not tasks:
+            print("  -> no tasks generated, skipping")
+            continue
+
+        for task in tasks:
+            text = task_text(task)
+            if not text or len(text) < 8:
+                continue
+
+            # prereq_op_selector — use text-based classifier (same signal as sprint-plan
+            # data) to avoid circular heuristic labels (C++ heuristic never emits
+            # manual-approval, so ops-based labeling produces 100% "standard").
+            prereq_examples.append((text_to_prereq_class(text), text))
+
+            # worker_type
+            wt = task.get("workerType", "")
+            if wt in WORKER_LABEL_INDEX:
+                worker_examples.append((wt, text))
+
+            # verification_type
+            vt = task.get("executionContract", {}).get("verificationType", "")
+            if vt in VERIF_LABEL_INDEX:
+                verif_examples.append((vt, text))
+
+        print(f"  -> {len(tasks)} tasks, prereq={len(prereq_examples)} "
+              f"worker={len(worker_examples)} verif={len(verif_examples)}")
+
+    if args.dry_run:
+        print("\n=== SAMPLE EXAMPLES ===")
+        for name, examples in [("prereq", prereq_examples[:5]),
+                                 ("worker", worker_examples[:5]),
+                                 ("verif",  verif_examples[:5])]:
+            print(f"\n{name}:")
+            for lbl, text in examples:
+                print(f"  [{lbl}] {text[:80]}")
+        return
+
+    PREREQ_ORDER = ["standard", "needs_review", "needs_approval", "full_gates"]
+    WORKER_ORDER = ["implementer", "reviewer", "architect", "qa"]
+    VERIF_ORDER  = ["unit", "integration", "schema", "smoke", "docs"]
+
+    print("\n=== prereq_op from projects ===")
+    train, eval_ = balance_and_split(prereq_examples, PREREQ_LABEL_INDEX, PREREQ_ORDER,
+                                     seed=args.seed, eval_frac=args.eval_frac)
+    write_tsv(out_dir / "prereq_op_projects_train.tsv", train, PREREQ_LABEL_INDEX)
+    write_tsv(out_dir / "prereq_op_projects_eval.tsv",  eval_, PREREQ_LABEL_INDEX)
+
+    print("\n=== worker_type from projects ===")
+    train, eval_ = balance_and_split(worker_examples, WORKER_LABEL_INDEX, WORKER_ORDER,
+                                     seed=args.seed, eval_frac=args.eval_frac)
+    write_tsv(out_dir / "worker_type_projects_train.tsv", train, WORKER_LABEL_INDEX)
+    write_tsv(out_dir / "worker_type_projects_eval.tsv",  eval_, WORKER_LABEL_INDEX)
+
+    print("\n=== verification_type from projects ===")
+    train, eval_ = balance_and_split(verif_examples, VERIF_LABEL_INDEX, VERIF_ORDER,
+                                     seed=args.seed, eval_frac=args.eval_frac)
+    write_tsv(out_dir / "verification_type_projects_train.tsv", train, VERIF_LABEL_INDEX)
+    write_tsv(out_dir / "verification_type_projects_eval.tsv",  eval_, VERIF_LABEL_INDEX)
+
+    print("\nDone. Combine with sprint-plan TSVs using:")
+    print("  cat specialists/data/generated/prereq_op_train.tsv \\")
+    print("      specialists/data/generated/prereq_op_projects_train.tsv \\")
+    print("      > specialists/data/combined/prereq_op_train.tsv")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/specialists/scripts/gen_prereq_op_data.py b/specialists/scripts/gen_prereq_op_data.py
new file mode 100644
index 0000000..e8b312e
--- /dev/null
+++ b/specialists/scripts/gen_prereq_op_data.py
@@ -0,0 +1,181 @@
+#!/usr/bin/env python3
+"""
+gen_prereq_op_data.py
+
+Mines whetstone_DSL sprint plan files to produce labeled training data for
+the prereq_op_selector specialist.
+
+Prerequisite op classes:
+  standard       — [validate-intake, resolve-dependencies]
+                   Default implementation task; no special gates needed.
+  needs_review   — [validate-intake, architect-review]
+                   Introduces new architecture, API surface, subsystem, or
+                   cross-component interface. Requires architect sign-off.
+  needs_approval — [validate-intake, manual-approval]
+                   Touches release, security, compliance, or production.
+                   Requires manual gate regardless of architecture complexity.
+  full_gates     — [validate-intake, architect-review, manual-approval]
+                   Both architectural and production-sensitive risk present.
+
+Output: TSV with columns: label<TAB>hops<TAB>text
+  label: integer index 0–3
+         0=standard  1=needs_review  2=needs_approval  3=full_gates
+  hops:  always 0 (unused by fabricate binary; required by format)
+  text:  the step description text
+
+Usage:
+    python3 specialists/scripts/gen_prereq_op_data.py \\
+        --sprint-dir /path/to/whetstone_DSL \\
+        --train-out specialists/data/generated/prereq_op_train.tsv \\
+        --eval-out  specialists/data/generated/prereq_op_eval.tsv
+"""
+import re
+import sys
+import random
+import argparse
+from pathlib import Path
+from collections import Counter
+
+# ---------------------------------------------------------------------------
+# Classification heuristics
+# ---------------------------------------------------------------------------
+
+# Signals that an architect needs to review this (new API, structure, design)
+REVIEW_TOKENS = {
+    "introduce", "phase", "framework", "architecture", "design", "refactor",
+    "planning", "taxonomy", "grammar", "interface", "api surface", "subsystem",
+    "extension point", "abstraction", "decompos", "scaffold", "polyglot",
+    "cross-language", "cross-project", "multi-language", "orchestrat",
+    "migration plan", "phase plan", "protocol design", "new layer",
+    "foundation", "bootstrap", "initial", "phase 1", "phase 2", "phase 3",
+    "sprint plan", "roadmap",
+}
+
+# Signals that a manual approval gate is required (release, security, production)
+APPROVAL_TOKENS = {
+    "release", "deploy", "publish", "migrate", "security", "compliance",
+    "upgrade", "breaking change", "production", "rollout", "finalize",
+    "sign off", "certification", "approval", "cutover", "go-live",
+    "data migration", "schema migration", "breaking", "deprecat",
+}
+
+LABEL_ORDER = ["standard", "needs_review", "needs_approval", "full_gates"]
+LABEL_INDEX = {lbl: i for i, lbl in enumerate(LABEL_ORDER)}
+
+
+def classify(text: str) -> str:
+    t = text.lower()
+    has_review   = any(tok in t for tok in REVIEW_TOKENS)
+    has_approval = any(tok in t for tok in APPROVAL_TOKENS)
+
+    if has_review and has_approval:
+        return "full_gates"
+    if has_review:
+        return "needs_review"
+    if has_approval:
+        return "needs_approval"
+    return "standard"
+
+
+# ---------------------------------------------------------------------------
+# Step extraction (same regex as gen_verification_type_data.py)
+# ---------------------------------------------------------------------------
+
+STEP_RE = re.compile(
+    r"###\s+Step\s+\d+[a-z]?:\s+(.+?)(?:\s*\(\d+\s+tests?\))?$",
+    re.IGNORECASE,
+)
+
+
+def extract_steps(sprint_path: Path) -> list[tuple[str, str]]:
+    results = []
+    for line in sprint_path.read_text(errors="replace").splitlines():
+        m = STEP_RE.match(line.strip())
+        if not m:
+            continue
+        desc = m.group(1).strip()
+        if not desc or len(desc) < 8:
+            continue
+        label = classify(desc)
+        results.append((label, desc))
+    return results
+
+
+# ---------------------------------------------------------------------------
+# Main
+# ---------------------------------------------------------------------------
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--sprint-dir", default=".",
+                    help="Root of whetstone_DSL (contains sprint*_plan.md files)")
+    ap.add_argument("--train-out", default="specialists/data/generated/prereq_op_train.tsv")
+    ap.add_argument("--eval-out",  default="specialists/data/generated/prereq_op_eval.tsv")
+    ap.add_argument("--eval-frac", type=float, default=0.15)
+    ap.add_argument("--seed", type=int, default=42)
+    args = ap.parse_args()
+
+    sprint_dir = Path(args.sprint_dir)
+    plans = sorted(sprint_dir.glob("sprint*_plan.md"))
+    if not plans:
+        print(f"No sprint plan files found in {sprint_dir}", file=sys.stderr)
+        sys.exit(1)
+    print(f"Found {len(plans)} sprint plan files")
+
+    all_examples: list[tuple[str, str]] = []
+    for p in plans:
+        all_examples.extend(extract_steps(p))
+
+    counts = Counter(lbl for lbl, _ in all_examples)
+    print("Raw class distribution:", dict(counts))
+
+    # Balance: cap at 3x the smallest class, floor at 40
+    min_count = min(counts.values())
+    target = min(min_count * 4, sorted(counts.values())[1] if len(counts) > 1 else min_count)
+    target = max(target, 40)
+    print(f"Balancing to {target} examples per class")
+
+    rng = random.Random(args.seed)
+    by_class: dict[str, list[str]] = {}
+    for lbl, desc in all_examples:
+        by_class.setdefault(lbl, []).append(desc)
+
+    balanced: list[tuple[str, str]] = []
+    for lbl in LABEL_ORDER:
+        descs = by_class.get(lbl, [])
+        if not descs:
+            print(f"WARNING: no examples for class '{lbl}'", file=sys.stderr)
+            continue
+        rng.shuffle(descs)
+        chosen = descs[:target]
+        while len(chosen) < target:
+            chosen += descs
+        balanced.extend((lbl, d) for d in chosen[:target])
+
+    rng.shuffle(balanced)
+    print(f"Balanced total: {len(balanced)} examples")
+
+    n_eval = max(1, int(len(balanced) * args.eval_frac))
+    eval_set  = balanced[:n_eval]
+    train_set = balanced[n_eval:]
+
+    Path(args.train_out).parent.mkdir(parents=True, exist_ok=True)
+    Path(args.eval_out).parent.mkdir(parents=True, exist_ok=True)
+
+    def write_tsv(path, rows):
+        with open(path, "w") as f:
+            for lbl, desc in rows:
+                idx = LABEL_INDEX[lbl]
+                f.write(f"{idx}\t0\t{desc}\n")
+        print(f"Wrote {len(rows)} rows → {path}")
+
+    write_tsv(args.train_out, train_set)
+    write_tsv(args.eval_out,  eval_set)
+
+    for split_name, split in [("train", train_set), ("eval", eval_set)]:
+        c = Counter(lbl for lbl, _ in split)
+        print(f"  {split_name}: {dict(sorted(c.items()))}")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/specialists/scripts/gen_verification_type_data.py b/specialists/scripts/gen_verification_type_data.py
new file mode 100644
index 0000000..860cbc4
--- /dev/null
+++ b/specialists/scripts/gen_verification_type_data.py
@@ -0,0 +1,204 @@
+#!/usr/bin/env python3
+"""
+gen_verification_type_data.py
+
+Mines whetstone_DSL sprint plan files to produce labeled training data for
+the verification_type specialist.
+
+Verification types:
+  unit        — tests a single function/class/module in isolation
+  integration — tests multiple components wired together (MCP tools, APIs, pipelines)
+  schema      — validates a data contract, struct, or schema definition
+  smoke       — sprint integration summary / quick regression / health check
+  docs        — documentation, packet emission, reporting artifacts
+
+Output: TSV with columns: label<TAB>hops<TAB>text
+  label: integer index 0–4 matching --labels "unit,integration,schema,smoke,docs"
+           0=unit  1=integration  2=schema  3=smoke  4=docs
+  hops:  always 0 (unused by fabricate binary; required by format)
+  text:  the step description text
+
+Usage:
+    python3 specialists/scripts/gen_verification_type_data.py \
+        --sprint-dir /path/to/whetstone_DSL \
+        --train-out specialists/data/generated/verification_type_train.tsv \
+        --eval-out  specialists/data/generated/verification_type_eval.tsv \
+        --eval-frac 0.15
+"""
+import re
+import sys
+import random
+import argparse
+from pathlib import Path
+from collections import Counter
+
+# ---------------------------------------------------------------------------
+# Classification heuristics
+# ---------------------------------------------------------------------------
+
+SCHEMA_TOKENS = {
+    "schema", "struct", "contract", "model", "type", "definition",
+    "serialization", "format", "layout", "ir", "ast", "grammar", "spec",
+    "signature", "interface", "protocol", "record", "field", "extension",
+    "packet", "payload", "shape", "descriptor",
+}
+
+INTEGRATION_TOKENS = {
+    "mcp", "tool", "endpoint", "api", "wire", "pipeline", "plugin",
+    "register", "hook", "dispatch", "router", "connection", "rpc",
+    "server", "client", "session", "daemon", "handler", "bridge",
+    "integration", "compose", "mount",
+}
+
+LABEL_ORDER = ["unit", "integration", "schema", "smoke", "docs"]
+LABEL_INDEX = {lbl: i for i, lbl in enumerate(LABEL_ORDER)}
+
+
+SMOKE_TOKENS = {
+    "integration summary", "sprint summary", "regression", "full-program",
+    "milestone", "health", "sanity", "smoke", "end-to-end summary",
+    "transition", "consolidation", "audit", "retrospective",
+}
+
+DOCS_TOKENS = {
+    "documentation", "readme", "comment", "guide", "notes",
+    "changelog", "release notes", "publication", "publish",
+    "announce", "report", "artifact bundle", "evidence bundle",
+    "scorecard", "scoreboard",
+}
+
+# Anything that doesn't match the above defaults to 'unit'.
+
+
+def classify(text: str) -> str:
+    t = text.lower()
+
+    # Smoke: integration summaries are almost always the last step
+    for tok in SMOKE_TOKENS:
+        if tok in t:
+            return "smoke"
+
+    # Docs: documentation/publication artifacts
+    for tok in DOCS_TOKENS:
+        if tok in t:
+            return "docs"
+
+    # Schema: data/contract definition
+    for tok in SCHEMA_TOKENS:
+        if tok in t:
+            return "schema"
+
+    # Integration: wiring multiple components or MCP tools
+    for tok in INTEGRATION_TOKENS:
+        if tok in t:
+            return "integration"
+
+    # Default: unit implementation
+    return "unit"
+
+
+# ---------------------------------------------------------------------------
+# Step extraction
+# ---------------------------------------------------------------------------
+
+STEP_RE = re.compile(
+    r"###\s+Step\s+\d+[a-z]?:\s+(.+?)(?:\s*\(\d+\s+tests?\))?$",
+    re.IGNORECASE,
+)
+
+
+def extract_steps(sprint_path: Path) -> list[tuple[str, str]]:
+    """Return list of (label, description) from a sprint plan file."""
+    results = []
+    for line in sprint_path.read_text(errors="replace").splitlines():
+        m = STEP_RE.match(line.strip())
+        if not m:
+            continue
+        desc = m.group(1).strip()
+        if not desc or len(desc) < 8:
+            continue
+        label = classify(desc)
+        results.append((label, desc))
+    return results
+
+
+# ---------------------------------------------------------------------------
+# Main
+# ---------------------------------------------------------------------------
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--sprint-dir", default=".",
+                    help="Root of whetstone_DSL (contains sprint*_plan.md files)")
+    ap.add_argument("--train-out", default="specialists/data/generated/verification_type_train.tsv")
+    ap.add_argument("--eval-out",  default="specialists/data/generated/verification_type_eval.tsv")
+    ap.add_argument("--eval-frac", type=float, default=0.15)
+    ap.add_argument("--seed", type=int, default=42)
+    args = ap.parse_args()
+
+    sprint_dir = Path(args.sprint_dir)
+    plans = sorted(sprint_dir.glob("sprint*_plan.md"))
+    if not plans:
+        print(f"No sprint plan files found in {sprint_dir}", file=sys.stderr)
+        sys.exit(1)
+    print(f"Found {len(plans)} sprint plan files")
+
+    all_examples = []
+    for p in plans:
+        for label, desc in extract_steps(p):
+            all_examples.append((label, desc))
+
+    counts = Counter(lbl for lbl, _ in all_examples)
+    print("Raw class distribution:", dict(counts))
+
+    # Balance: undersample to the second-smallest class, but cap at 3x min
+    min_count = min(counts.values())
+    target = min(min_count * 4, sorted(counts.values())[1])
+    target = max(target, 40)  # floor
+    print(f"Balancing to {target} examples per class")
+
+    rng = random.Random(args.seed)
+    by_class: dict[str, list[str]] = {}
+    for lbl, desc in all_examples:
+        by_class.setdefault(lbl, []).append(desc)
+
+    balanced = []
+    for lbl, descs in by_class.items():
+        rng.shuffle(descs)
+        chosen = descs[:target]
+        if len(chosen) < target:
+            # oversample with repetition to reach target
+            while len(chosen) < target:
+                chosen += descs
+            chosen = chosen[:target]
+        balanced.extend((lbl, d) for d in chosen)
+
+    rng.shuffle(balanced)
+    print(f"Balanced total: {len(balanced)} examples")
+
+    # Train/eval split
+    n_eval = max(1, int(len(balanced) * args.eval_frac))
+    eval_set  = balanced[:n_eval]
+    train_set = balanced[n_eval:]
+
+    Path(args.train_out).parent.mkdir(parents=True, exist_ok=True)
+    Path(args.eval_out).parent.mkdir(parents=True, exist_ok=True)
+
+    def write_tsv(path, rows):
+        with open(path, "w") as f:
+            for lbl, desc in rows:
+                idx = LABEL_INDEX[lbl]
+                f.write(f"{idx}\t0\t{desc}\n")
+        print(f"Wrote {len(rows)} rows → {path}")
+
+    write_tsv(args.train_out, train_set)
+    write_tsv(args.eval_out,  eval_set)
+
+    # Print final counts
+    for split_name, split in [("train", train_set), ("eval", eval_set)]:
+        c = Counter(lbl for lbl, _ in split)
+        print(f"  {split_name}: {dict(sorted(c.items()))}")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/specialists/scripts/gen_worker_type_data.py b/specialists/scripts/gen_worker_type_data.py
new file mode 100644
index 0000000..082e45b
--- /dev/null
+++ b/specialists/scripts/gen_worker_type_data.py
@@ -0,0 +1,188 @@
+#!/usr/bin/env python3
+"""
+gen_worker_type_data.py
+
+Mines whetstone_DSL sprint plan files to produce labeled training data for
+the worker_type specialist.
+
+Worker types:
+  implementer — Builds, adds, creates, extends a feature or component.
+                Most sprint steps fall here.
+  reviewer    — Validates, audits, checks, or analyzes existing work.
+  architect   — Designs, plans, introduces new structural layers or frameworks.
+  qa          — Writes or runs tests, regression checks, benchmarks.
+
+Output: TSV with columns: label<TAB>hops<TAB>text
+  label: integer index 0–3
+         0=implementer  1=reviewer  2=architect  3=qa
+  hops:  always 0 (unused by fabricate binary; required by format)
+  text:  the step description text
+
+Usage:
+    python3 specialists/scripts/gen_worker_type_data.py \\
+        --sprint-dir /path/to/whetstone_DSL \\
+        --train-out specialists/data/generated/worker_type_train.tsv \\
+        --eval-out  specialists/data/generated/worker_type_eval.tsv
+"""
+import re
+import sys
+import random
+import argparse
+from pathlib import Path
+from collections import Counter
+
+# ---------------------------------------------------------------------------
+# Classification heuristics
+# ---------------------------------------------------------------------------
+
+# QA: writing/running tests, regression, benchmarks — check before architect
+# because "integration summary" etc. are clearly QA not architecture
+QA_TOKENS = {
+    "test", "tests", "regression", "benchmark", "benchmarks", "coverage",
+    "smoke", "integration summary", "sprint summary", "harness", "end-to-end",
+    "e2e", "golden", "fixture", "passing", "failing test",
+}
+
+# Architect: introduces new structure, layers, frameworks, decompositions
+ARCHITECT_TOKENS = {
+    "introduce", "phase", "architecture", "framework", "refactor",
+    "planning", "taxonomy", "grammar", "scaffold", "protocol",
+    "decompos", "organiz", "restructur", "subsystem", "foundation",
+    "bootstrap", "initial design", "phase plan", "design doc",
+    "migration plan", "sprint plan", "roadmap", "new layer",
+}
+
+# Reviewer: validates, audits, checks existing things
+REVIEWER_TOKENS = {
+    "review", "audit", "inspect", "diagnos", "assess",
+    "scan", "lint", "triage", "evaluate", "measure",
+}
+
+# implementer is the default — everything else
+
+LABEL_ORDER = ["implementer", "reviewer", "architect", "qa"]
+LABEL_INDEX = {lbl: i for i, lbl in enumerate(LABEL_ORDER)}
+
+
+def classify(text: str) -> str:
+    t = text.lower()
+
+    # QA first — "integration summary + regression" is clearly qa not arch
+    for tok in QA_TOKENS:
+        if tok in t:
+            return "qa"
+
+    # Architect next — structural introductions
+    for tok in ARCHITECT_TOKENS:
+        if tok in t:
+            return "architect"
+
+    # Reviewer — validation/audit tasks
+    for tok in REVIEWER_TOKENS:
+        if tok in t:
+            return "reviewer"
+
+    return "implementer"
+
+
+# ---------------------------------------------------------------------------
+# Step extraction
+# ---------------------------------------------------------------------------
+
+STEP_RE = re.compile(
+    r"###\s+Step\s+\d+[a-z]?:\s+(.+?)(?:\s*\(\d+\s+tests?\))?$",
+    re.IGNORECASE,
+)
+
+
+def extract_steps(sprint_path: Path) -> list[tuple[str, str]]:
+    results = []
+    for line in sprint_path.read_text(errors="replace").splitlines():
+        m = STEP_RE.match(line.strip())
+        if not m:
+            continue
+        desc = m.group(1).strip()
+        if not desc or len(desc) < 8:
+            continue
+        label = classify(desc)
+        results.append((label, desc))
+    return results
+
+
+# ---------------------------------------------------------------------------
+# Main
+# ---------------------------------------------------------------------------
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--sprint-dir", default=".",
+                    help="Root of whetstone_DSL (contains sprint*_plan.md files)")
+    ap.add_argument("--train-out", default="specialists/data/generated/worker_type_train.tsv")
+    ap.add_argument("--eval-out",  default="specialists/data/generated/worker_type_eval.tsv")
+    ap.add_argument("--eval-frac", type=float, default=0.15)
+    ap.add_argument("--seed", type=int, default=42)
+    args = ap.parse_args()
+
+    sprint_dir = Path(args.sprint_dir)
+    plans = sorted(sprint_dir.glob("sprint*_plan.md"))
+    if not plans:
+        print(f"No sprint plan files found in {sprint_dir}", file=sys.stderr)
+        sys.exit(1)
+    print(f"Found {len(plans)} sprint plan files")
+
+    all_examples: list[tuple[str, str]] = []
+    for p in plans:
+        all_examples.extend(extract_steps(p))
+
+    counts = Counter(lbl for lbl, _ in all_examples)
+    print("Raw class distribution:", dict(counts))
+
+    min_count = min(counts.values())
+    target = min(min_count * 4, sorted(counts.values())[1] if len(counts) > 1 else min_count)
+    target = max(target, 40)
+    print(f"Balancing to {target} examples per class")
+
+    rng = random.Random(args.seed)
+    by_class: dict[str, list[str]] = {}
+    for lbl, desc in all_examples:
+        by_class.setdefault(lbl, []).append(desc)
+
+    balanced: list[tuple[str, str]] = []
+    for lbl in LABEL_ORDER:
+        descs = by_class.get(lbl, [])
+        if not descs:
+            print(f"WARNING: no examples for class '{lbl}'", file=sys.stderr)
+            continue
+        rng.shuffle(descs)
+        chosen = descs[:target]
+        while len(chosen) < target:
+            chosen += descs
+        balanced.extend((lbl, d) for d in chosen[:target])
+
+    rng.shuffle(balanced)
+    print(f"Balanced total: {len(balanced)} examples")
+
+    n_eval = max(1, int(len(balanced) * args.eval_frac))
+    eval_set  = balanced[:n_eval]
+    train_set = balanced[n_eval:]
+
+    Path(args.train_out).parent.mkdir(parents=True, exist_ok=True)
+    Path(args.eval_out).parent.mkdir(parents=True, exist_ok=True)
+
+    def write_tsv(path, rows):
+        with open(path, "w") as f:
+            for lbl, desc in rows:
+                idx = LABEL_INDEX[lbl]
+                f.write(f"{idx}\t0\t{desc}\n")
+        print(f"Wrote {len(rows)} rows → {path}")
+
+    write_tsv(args.train_out, train_set)
+    write_tsv(args.eval_out,  eval_set)
+
+    for split_name, split in [("train", train_set), ("eval", eval_set)]:
+        c = Counter(lbl for lbl, _ in split)
+        print(f"  {split_name}: {dict(sorted(c.items()))}")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/specialists/scripts/infer.py b/specialists/scripts/infer.py
new file mode 100644
index 0000000..579dbc9
--- /dev/null
+++ b/specialists/scripts/infer.py
@@ -0,0 +1,159 @@
+#!/usr/bin/env python3
+"""
+infer.py — Run a trained Whetstone specialist on text input.
+
+Supports both binary specialists (true/false) and N-way classifiers.
+
+Usage:
+    # Single query
+    python3 specialists/scripts/infer.py \
+        --run-dir /mnt/storage/fabricate_runs/whetstone_verification_type \
+        --labels "unit,integration,schema,smoke,docs" \
+        --text "Sprint 200 integration summary + regression"
+
+    # Batch from file (one text per line)
+    python3 specialists/scripts/infer.py \
+        --run-dir /mnt/storage/fabricate_runs/whetstone_verification_type \
+        --labels "unit,integration,schema,smoke,docs" \
+        --input-file my_tasks.txt
+
+    # Interactive REPL
+    python3 specialists/scripts/infer.py \
+        --run-dir /mnt/storage/fabricate_runs/whetstone_verification_type \
+        --labels "unit,integration,schema,smoke,docs"
+"""
+import sys
+import argparse
+import numpy as np
+from pathlib import Path
+
+FABRICATE_TOOLS = Path("/home/bill/Documents/WhetstoneAI_Fabricate/tools")
+
+
+def load_specialist(run_dir: str, labels_str: str | None):
+    """Load checkpoint, vocab, and labels. Return inference-ready state."""
+    sys.path.insert(0, str(FABRICATE_TOOLS))
+    from eval_with_vocab import load_checkpoint, load_vocab, load_layers, build_runtime
+
+    ck  = Path(run_dir) / "checkpoint.bin"
+    db  = Path(run_dir) / "fabricate.db"
+    if not ck.exists():
+        raise FileNotFoundError(f"No checkpoint.bin in {run_dir}")
+    if not db.exists():
+        raise FileNotFoundError(f"No fabricate.db in {run_dir}")
+
+    weights, _ = load_checkpoint(str(ck))
+    _, str_to_id = load_vocab(str(db))
+    layers = load_layers(str(db))
+
+    V = len(str_to_id)
+    _, emb_cnt = layers['token_embedding']
+    D = emb_cnt // V
+    _, ff1_cnt = layers['block0.ff1_weight']
+    F = ff1_cnt // D
+    L = sum(1 for k in layers if k.startswith('block') and k.endswith('.qkv_weight'))
+    cfg = dict(seq_len=96, D=D, F=F, V=V, L=L)
+    runtime = build_runtime(weights, layers, cfg)
+
+    label_list = None
+    if labels_str:
+        p = Path(labels_str)
+        if p.exists():
+            label_list = [l.strip() for l in p.read_text().split(',')]
+        else:
+            label_list = [l.strip() for l in labels_str.split(',')]
+
+    return weights, layers, cfg, runtime, str_to_id, label_list
+
+
+def predict(text: str, weights, layers, cfg, runtime, str_to_id, label_list) -> tuple[str, float]:
+    """Predict label for a single text. Returns (predicted_label, confidence)."""
+    sys.path.insert(0, str(FABRICATE_TOOLS))
+    from eval_with_vocab import tokenize_batch, forward_batch
+
+    token_ids = tokenize_batch([text], str_to_id, cfg['seq_len'])
+    logits = forward_batch(token_ids, weights, layers, cfg, runtime=runtime)  # (1, V)
+
+    if label_list:
+        # N-way: restrict to label vocab IDs, take argmax
+        label_ids = np.array([str_to_id.get(l, 0) for l in label_list], dtype=np.int32)
+        label_logits = logits[0, label_ids]
+        probs = np.exp(label_logits - label_logits.max())
+        probs /= probs.sum()
+        idx = int(np.argmax(probs))
+        return label_list[idx], float(probs[idx])
+    else:
+        # Binary true/false
+        true_id  = str_to_id.get('true',  3)
+        false_id = str_to_id.get('false', 4)
+        true_logit  = logits[0, true_id]
+        false_logit = logits[0, false_id]
+        prob_true = float(np.exp(true_logit) / (np.exp(true_logit) + np.exp(false_logit)))
+        if prob_true >= 0.5:
+            return "true", prob_true
+        else:
+            return "false", 1.0 - prob_true
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--run-dir", required=True,
+                    help="Directory with checkpoint.bin and fabricate.db")
+    ap.add_argument("--labels", default=None,
+                    help="Comma-sep label list, e.g. 'unit,integration,schema,smoke,docs'")
+    ap.add_argument("--text",   default=None, help="Single text to classify")
+    ap.add_argument("--input-file", default=None, help="File with one text per line")
+    ap.add_argument("--eval-tsv", default=None,
+                    help="Evaluate accuracy on a TSV file (label<TAB>hops<TAB>text)")
+    args = ap.parse_args()
+
+    print(f"Loading specialist from {args.run_dir}...", flush=True)
+    state = load_specialist(args.run_dir, args.labels)
+    weights, layers, cfg, runtime, str_to_id, label_list = state
+    print(f"Loaded. V={cfg['V']} D={cfg['D']} L={cfg['L']}  labels={label_list}")
+
+    if args.eval_tsv:
+        correct = total = 0
+        with open(args.eval_tsv) as f:
+            for line in f:
+                parts = line.strip().split('\t', 2)
+                if len(parts) < 3:
+                    continue
+                true_idx = int(parts[0])
+                text = parts[2]
+                true_label = label_list[true_idx] if label_list else str(true_idx)
+                pred, conf = predict(text, *state[:-1], label_list)
+                if pred == true_label:
+                    correct += 1
+                total += 1
+        print(f"Accuracy: {correct}/{total} = {100*correct/max(1,total):.1f}%")
+        return
+
+    if args.input_file:
+        with open(args.input_file) as f:
+            lines = [l.strip() for l in f if l.strip()]
+        for text in lines:
+            pred, conf = predict(text, weights, layers, cfg, runtime, str_to_id, label_list)
+            print(f"{pred:15s}  {conf:.3f}  {text}")
+        return
+
+    if args.text:
+        pred, conf = predict(args.text, weights, layers, cfg, runtime, str_to_id, label_list)
+        print(f"{pred}  ({conf:.3f})")
+        return
+
+    # Interactive REPL
+    print("Enter task descriptions (Ctrl-D to quit):")
+    while True:
+        try:
+            text = input("> ").strip()
+        except (EOFError, KeyboardInterrupt):
+            break
+        if not text:
+            continue
+        pred, conf = predict(text, weights, layers, cfg, runtime, str_to_id, label_list)
+        print(f"  → {pred}  (conf={conf:.3f})")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/specialists/scripts/prereq_op_binary_split.py b/specialists/scripts/prereq_op_binary_split.py
new file mode 100644
index 0000000..aa78a0d
--- /dev/null
+++ b/specialists/scripts/prereq_op_binary_split.py
@@ -0,0 +1,319 @@
+#!/usr/bin/env python3
+"""
+prereq_op_binary_split.py
+
+Tests whether prereq_op is a factorizable gate:
+  - Instead of a 4-way flat classifier, train two binary classifiers:
+      1. requires_architect_review (yes/no)
+      2. requires_manual_approval  (yes/no)
+  - Combine deterministically:
+      no/no  → standard      (0)
+      yes/no → needs_review  (1)
+      no/yes → needs_approval (2)
+      yes/yes → full_gates   (3)
+
+Then compare:
+  - flat 4-way model accuracy
+  - combined binary model accuracy
+
+Also measures: false-gate rate (how often either binary fires when neither should).
+
+Usage:
+    python3 specialists/scripts/prereq_op_binary_split.py \\
+        --data-dir  specialists/data/combined \\
+        --runs-dir  /mnt/storage/fabricate_runs \\
+        --results-out specialists/eval/results/prereq_op_factorize.json
+"""
+
+import argparse
+import json
+import subprocess
+import sys
+from pathlib import Path
+
+ROOT        = Path(__file__).parent.parent.parent
+FABRICATE   = Path("/home/bill/Documents/WhetstoneAI_Fabricate")
+VENV_PYTHON = FABRICATE / ".venv" / "bin" / "python3"
+TRAIN_SCRIPT = ROOT / "specialists" / "scripts" / "train_specialist_pt.py"
+EVAL_SCRIPT  = ROOT / "specialists" / "eval" / "eval_specialist_pt.py"
+
+# prereq_op label mapping
+LABEL_NAMES = ["standard", "needs_review", "needs_approval", "full_gates"]
+# Axis decomposition:
+#   requires_review:   needs_review(1), full_gates(3)  → binary label 1
+#   requires_approval: needs_approval(2), full_gates(3) → binary label 1
+REVIEW_LABELS   = {1, 3}   # 4-way indices that mean "needs review"
+APPROVAL_LABELS = {2, 3}   # 4-way indices that mean "needs approval"
+
+
+def split_to_binary(src_tsv: Path, review_out: Path, approval_out: Path):
+    """Write two binary TSVs from 4-way prereq_op data."""
+    review_rows = []
+    approval_rows = []
+    with open(src_tsv) as f:
+        for line in f:
+            parts = line.strip().split("\t", 2)
+            if len(parts) < 3:
+                continue
+            label_idx = int(parts[0])
+            text = parts[2]
+            review_label   = 1 if label_idx in REVIEW_LABELS   else 0
+            approval_label = 1 if label_idx in APPROVAL_LABELS else 0
+            review_rows.append(f"{review_label}\t0\t{text}")
+            approval_rows.append(f"{approval_label}\t0\t{text}")
+
+    review_out.write_text("\n".join(review_rows) + "\n")
+    approval_out.write_text("\n".join(approval_rows) + "\n")
+    print(f"  review   TSV: {len(review_rows)} rows → {review_out}")
+    print(f"  approval TSV: {len(approval_rows)} rows → {approval_out}")
+    # Report class balance
+    r1 = sum(1 for r in review_rows if r.startswith("1"))
+    a1 = sum(1 for r in approval_rows if r.startswith("1"))
+    print(f"  review   pos={r1} neg={len(review_rows)-r1}")
+    print(f"  approval pos={a1} neg={len(approval_rows)-a1}")
+
+
+def train_binary(run_dir: Path, train_tsv: Path, eval_tsv: Path,
+                 labels: str, history_out: Path) -> bool:
+    run_dir.mkdir(parents=True, exist_ok=True)
+    for f in run_dir.glob("*"):
+        f.unlink()
+    history_out.unlink(missing_ok=True)
+
+    cmd = [
+        str(VENV_PYTHON), str(TRAIN_SCRIPT),
+        "--dataset",      str(train_tsv),
+        "--eval-dataset", str(eval_tsv),
+        "--out-dir",      str(run_dir),
+        "--labels",       labels,
+        "--lr",           "0.001",
+        "--weight-decay", "0.01",
+        "--max-steps",    "10000",
+        "--hidden-dim",   "256",
+        "--layers",       "2",
+        "--heads",        "4",
+        "--batch-size",   "256",
+        "--grok-loss-threshold", "0.05",
+        "--grok-acc-jump",       "10.0",
+        "--stop-after-grokking-blocks", "4",
+        "--reset",
+        "--history-out",  str(history_out),
+    ]
+    r = subprocess.run(cmd)
+    return r.returncode == 0
+
+
+def eval_binary(run_dir: Path, eval_tsv: Path, labels: str) -> dict | None:
+    ck = run_dir / "checkpoint.pt"
+    if not ck.exists():
+        return None
+    r = subprocess.run(
+        [str(VENV_PYTHON), str(EVAL_SCRIPT),
+         "--checkpoint", str(ck),
+         "--dataset", str(eval_tsv), "--labels", labels, "--quiet"],
+        capture_output=True, text=True,
+    )
+    if r.returncode != 0:
+        return None
+    try:
+        return json.loads(r.stdout.strip().splitlines()[-1] if r.stdout.strip() else "{}")
+    except json.JSONDecodeError:
+        return None
+
+
+def evaluate_combined(eval_tsv: Path, review_run: Path, approval_run: Path) -> dict:
+    """Load both PyTorch binary models, combine predictions, compare to 4-way ground truth."""
+    import numpy as np
+    import torch
+    import torch.nn as nn
+
+    # Inline model class (mirrors train_specialist_pt.py)
+    def _tok(text):
+        return text.lower().split()
+
+    class _Model(nn.Module):
+        def __init__(self, cfg):
+            super().__init__()
+            D, L, H, V, seq = cfg["hidden_dim"], cfg["layers"], cfg["heads"], cfg["vocab_size"], cfg["seq_len"]
+            self.tok_emb = nn.Embedding(V, D, padding_idx=0)
+            self.pos_emb = nn.Embedding(seq, D)
+            enc = nn.TransformerEncoderLayer(d_model=D, nhead=H, dim_feedforward=D*4,
+                                             dropout=0.0, batch_first=True)
+            self.encoder = nn.TransformerEncoder(enc, num_layers=L)
+            self.head = nn.Linear(D, cfg["n_labels"])
+            self.seq_len = seq
+
+        def forward(self, x):
+            pad_mask = (x == 0)
+            pos = torch.arange(x.size(1), device=x.device).unsqueeze(0)
+            h = self.tok_emb(x) + self.pos_emb(pos)
+            h = self.encoder(h, src_key_padding_mask=pad_mask)
+            lengths = (~pad_mask).float().sum(1, keepdim=True).clamp(min=1)
+            h = (h * (~pad_mask).unsqueeze(-1).float()).sum(1) / lengths
+            return self.head(h)
+
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+    def load_pt_model(run_dir):
+        ckpt = torch.load(run_dir / "checkpoint.pt", map_location=device)
+        m = _Model(ckpt["config"]).to(device)
+        m.load_state_dict(ckpt["model"])
+        m.eval()
+        vocab = ckpt.get("vocab_obj")
+        if vocab is None:
+            # reconstruct minimal vocab object
+            class _V:
+                UNK = 1
+                def __init__(self, d): self.str_to_id = d; self.seq_len = ckpt["config"]["seq_len"]
+                def encode(self, text, seq_len):
+                    ids = [self.str_to_id.get(t, self.UNK) for t in _tok(text)]
+                    ids = ids[:seq_len]; ids += [0] * (seq_len - len(ids)); return ids
+            vocab = _V(ckpt["vocab"])
+        return m, vocab, ckpt["config"]["seq_len"]
+
+    def predict(model, vocab, seq_len, texts):
+        preds = []
+        with torch.no_grad():
+            for start in range(0, len(texts), 256):
+                batch = texts[start:start+256]
+                ids = [vocab.encode(t, seq_len) for t in batch]
+                x = torch.tensor(ids, dtype=torch.long, device=device)
+                logits = model(x).cpu().numpy()
+                preds.append(np.argmax(logits, axis=1))
+        return np.concatenate(preds)
+
+    # Load data
+    labels_gt, texts = [], []
+    with open(eval_tsv) as f:
+        for line in f:
+            parts = line.strip().split("\t", 2)
+            if len(parts) < 3:
+                continue
+            labels_gt.append(int(parts[0]))
+            texts.append(parts[2])
+    labels_gt = np.array(labels_gt)
+    n = len(labels_gt)
+
+    review_model,   review_vocab,   review_seq   = load_pt_model(review_run)
+    approval_model, approval_vocab, approval_seq = load_pt_model(approval_run)
+
+    # Binary models: label 0 = no, label 1 = yes
+    review_pred   = predict(review_model,   review_vocab,   review_seq,   texts)
+    approval_pred = predict(approval_model, approval_vocab, approval_seq, texts)
+
+    # Combine: review×approval → 4-way index
+    combined_pred = np.where(
+        (review_pred == 0) & (approval_pred == 0), 0,   # standard
+        np.where(
+            (review_pred == 1) & (approval_pred == 0), 1,  # needs_review
+            np.where(
+                (review_pred == 0) & (approval_pred == 1), 2,  # needs_approval
+                3   # full_gates
+            )
+        )
+    )
+
+    correct = (combined_pred == labels_gt)
+    overall = float(correct.mean() * 100)
+
+    per_class = {}
+    for i, name in enumerate(LABEL_NAMES):
+        mask = labels_gt == i
+        if mask.sum() == 0:
+            per_class[name] = {"acc": None, "n": 0}
+        else:
+            per_class[name] = {"acc": float(correct[mask].mean() * 100), "n": int(mask.sum())}
+
+    from collections import Counter
+    errors = Counter()
+    for gt, pred in zip(labels_gt.tolist(), combined_pred.tolist()):
+        if gt != pred:
+            errors[(LABEL_NAMES[gt], LABEL_NAMES[pred])] += 1
+
+    return {
+        "method": "binary_factorized",
+        "overall_accuracy": round(overall, 1),
+        "n_eval": n,
+        "per_class": per_class,
+        "top_errors": [{"gt": k[0], "pred": k[1], "count": v}
+                       for k, v in errors.most_common(6)],
+    }
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--data-dir",    default="specialists/data/combined")
+    ap.add_argument("--runs-dir",    default="/mnt/storage/fabricate_runs")
+    ap.add_argument("--results-out", default="specialists/eval/results/prereq_op_factorize.json")
+    ap.add_argument("--skip-train",  action="store_true",
+                    help="Skip training, just eval existing checkpoints")
+    args = ap.parse_args()
+
+    data_dir  = Path(args.data_dir)
+    runs_dir  = Path(args.runs_dir)
+    results   = Path(args.results_out)
+    results.parent.mkdir(parents=True, exist_ok=True)
+
+    # Intermediate data paths
+    tmp = Path("specialists/data/generated")
+    tmp.mkdir(parents=True, exist_ok=True)
+    review_train   = tmp / "prereq_review_train.tsv"
+    review_eval    = tmp / "prereq_review_eval.tsv"
+    approval_train = tmp / "prereq_approval_train.tsv"
+    approval_eval  = tmp / "prereq_approval_eval.tsv"
+
+    # --- Split data ---
+    print("=== Splitting 4-way data into two binary axes ===")
+    split_to_binary(data_dir / "prereq_op_train.tsv", review_train, approval_train)
+    split_to_binary(data_dir / "prereq_op_eval.tsv",  review_eval,  approval_eval)
+
+    review_run   = runs_dir / "whetstone_prereq_review_binary"
+    approval_run = runs_dir / "whetstone_prereq_approval_binary"
+    review_hist   = ROOT / "specialists" / "runs" / "prereq_review_history.json"
+    approval_hist = ROOT / "specialists" / "runs" / "prereq_approval_history.json"
+
+    if not args.skip_train:
+        # --- Train binary models ---
+        print("\n=== Training: requires_architect_review (binary) ===")
+        train_binary(review_run, review_train, review_eval, "no,yes", review_hist)
+
+        print("\n=== Training: requires_manual_approval (binary) ===")
+        train_binary(approval_run, approval_train, approval_eval, "no,yes", approval_hist)
+
+    # --- Eval individual binary models ---
+    print("\n=== Evaluating binary models independently ===")
+    review_result   = eval_binary(review_run,   review_eval,   "no,yes")
+    approval_result = eval_binary(approval_run, approval_eval, "no,yes")
+
+    if review_result:
+        print(f"  review   binary: {review_result.get('overall_accuracy')}%")
+    if approval_result:
+        print(f"  approval binary: {approval_result.get('overall_accuracy')}%")
+
+    # --- Evaluate combined predictions on 4-way task ---
+    print("\n=== Evaluating combined binary → 4-way accuracy ===")
+    combined = evaluate_combined(
+        data_dir / "prereq_op_eval.tsv", review_run, approval_run
+    )
+    print(f"  Combined 4-way: {combined['overall_accuracy']}%")
+    print("  Per class:")
+    for name, stat in combined["per_class"].items():
+        if stat["acc"] is not None:
+            print(f"    {name:20s}: {stat['acc']:.1f}%  (n={stat['n']})")
+
+    output = {
+        "gate": "prereq_op",
+        "factorized": {
+            "requires_architect_review": review_result,
+            "requires_manual_approval":  approval_result,
+            "combined_4way":             combined,
+        },
+        "note": "Compare combined_4way.overall_accuracy against flat 4-way model accuracy",
+    }
+
+    results.write_text(json.dumps(output, indent=2))
+    print(f"\nResults → {results}")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/specialists/scripts/sweep_all_gates.sh b/specialists/scripts/sweep_all_gates.sh
new file mode 100755
index 0000000..19c4d9d
--- /dev/null
+++ b/specialists/scripts/sweep_all_gates.sh
@@ -0,0 +1,129 @@
+#!/usr/bin/env bash
+# sweep_all_gates.sh
+#
+# Run capacity sweeps for all 4 specialists + prereq_op factorizability test.
+# Runs sweeps sequentially (one GPU job at a time).
+#
+# Usage:
+#   bash specialists/scripts/sweep_all_gates.sh [--pilot-only] [--gates worker_type,prereq_op]
+#
+# Defaults: full sweep for all gates.
+#
+# Output:
+#   specialists/eval/results/<gate>_sweep.json  — per-gate sweep table
+#   specialists/eval/results/prereq_op_factorize.json — factorize comparison
+#   specialists/eval/results/summary.json        — all gates combined
+
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+PYTHON="/home/bill/Documents/WhetstoneAI_Fabricate/.venv/bin/python3"
+RUNS_DIR="/mnt/storage/fabricate_runs"
+RESULTS_DIR="$ROOT/specialists/eval/results"
+mkdir -p "$RESULTS_DIR"
+
+PILOT_ONLY=""
+GATES="verification_type,worker_type,prereq_op,automatability"
+
+for arg in "$@"; do
+  case "$arg" in
+    --pilot-only) PILOT_ONLY="--pilot-only" ;;
+    --gates=*)    GATES="${arg#--gates=}" ;;
+  esac
+done
+
+echo "=== Whetstone Specialist Capacity Sweep ==="
+echo "  Gates:   $GATES"
+echo "  Mode:    ${PILOT_ONLY:-(full sweep)}"
+echo "  Results: $RESULTS_DIR"
+echo ""
+
+run_sweep() {
+  local name="$1"; local train="$2"; local eval="$3"; local labels="$4"
+  if echo "$GATES" | grep -qw "$name"; then
+    echo ""
+    echo ">>> Gate: $name"
+    "$PYTHON" "$ROOT/specialists/scripts/capacity_sweep.py" \
+      --name        "$name" \
+      --train       "$train" \
+      --eval        "$eval" \
+      --labels      "$labels" \
+      --runs-dir    "$RUNS_DIR" \
+      --results-out "$RESULTS_DIR/${name}_sweep.json" \
+      $PILOT_ONLY \
+      2>&1 | tee "$RESULTS_DIR/${name}_sweep.log"
+    echo ""
+  fi
+}
+
+run_sweep "verification_type" \
+  "$ROOT/specialists/data/generated/verification_type_train.tsv" \
+  "$ROOT/specialists/data/generated/verification_type_eval.tsv" \
+  "unit,integration,schema,smoke,docs"
+
+run_sweep "worker_type" \
+  "$ROOT/specialists/data/generated/worker_type_train.tsv" \
+  "$ROOT/specialists/data/generated/worker_type_eval.tsv" \
+  "implementer,reviewer,architect,qa"
+
+run_sweep "prereq_op" \
+  "$ROOT/specialists/data/combined/prereq_op_train.tsv" \
+  "$ROOT/specialists/data/combined/prereq_op_eval.tsv" \
+  "standard,needs_review,needs_approval,full_gates"
+
+run_sweep "automatability" \
+  "$ROOT/specialists/data/generated/automatability_train.tsv" \
+  "$ROOT/specialists/data/generated/automatability_eval.tsv" \
+  "deterministic,template,specialist,slm,llm,human"
+
+# prereq_op factorizability test (runs alongside the sweep)
+if echo "$GATES" | grep -qw "prereq_op"; then
+  echo ""
+  echo ">>> prereq_op: factorizability test (two binary classifiers)"
+  "$PYTHON" "$ROOT/specialists/scripts/prereq_op_binary_split.py" \
+    --data-dir    "$ROOT/specialists/data/combined" \
+    --runs-dir    "$RUNS_DIR" \
+    --results-out "$RESULTS_DIR/prereq_op_factorize.json" \
+    2>&1 | tee "$RESULTS_DIR/prereq_op_factorize.log"
+fi
+
+# --- Combine into summary ---
+"$PYTHON" - <<'PYEOF'
+import json, glob, sys
+from pathlib import Path
+
+results_dir = Path("specialists/eval/results")
+summary = {"gates": {}}
+
+for path in sorted(results_dir.glob("*_sweep.json")):
+    gate = path.stem.replace("_sweep", "")
+    data = json.loads(path.read_text())
+    best_tier = max(
+        (t for t in data["tiers"] if t.get("eval_accuracy")),
+        key=lambda t: t.get("eval_accuracy", 0),
+        default=None,
+    )
+    summary["gates"][gate] = {
+        "best_tier": best_tier["tier"] if best_tier else None,
+        "best_accuracy": best_tier["eval_accuracy"] if best_tier else None,
+        "verdict": best_tier.get("verdict") if best_tier else None,
+        "guardrail_threshold": best_tier.get("guardrail_threshold") if best_tier else None,
+    }
+
+# Add factorize comparison if available
+factorize_path = results_dir / "prereq_op_factorize.json"
+if factorize_path.exists():
+    data = json.loads(factorize_path.read_text())
+    summary["gates"]["prereq_op"]["factorized_accuracy"] = (
+        data.get("factorized", {}).get("combined_4way", {}).get("overall_accuracy")
+    )
+
+out = results_dir / "summary.json"
+out.write_text(json.dumps(summary, indent=2))
+print(f"\nSummary → {out}")
+print(json.dumps(summary, indent=2))
+PYEOF
+
+echo ""
+echo "=== Sweep complete ==="
+echo "  Full results in: $RESULTS_DIR"
diff --git a/specialists/scripts/train_all_large.sh b/specialists/scripts/train_all_large.sh
new file mode 100755
index 0000000..df1bd46
--- /dev/null
+++ b/specialists/scripts/train_all_large.sh
@@ -0,0 +1,91 @@
+#!/usr/bin/env bash
+# Train all 4 specialists with a larger model (hidden_dim=512, layers=2, heads=4).
+# ~10-12x the capacity of the default (hidden_dim=128, layers=1, heads=1).
+# Uses combined sprint-plan + project data for prereq_op.
+#
+# Usage: bash specialists/scripts/train_all_large.sh
+#        (run from whetstone_DSL root)
+
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+FABRICATE_ROOT="/home/bill/Documents/WhetstoneAI_Fabricate"
+RUN_BASE="/mnt/storage/fabricate_runs"
+
+# Model size — ~16x the default 128/1/1
+# Note: gpu_transformer engine requires heads=1
+HIDDEN=512
+LAYERS=4
+HEADS=1
+MAX_STEPS=20000
+
+mkdir -p "$ROOT/specialists/runs"
+
+cd "$FABRICATE_ROOT"
+
+run_specialist() {
+  local name="$1"
+  local train="$2"
+  local eval="$3"
+  local labels="$4"
+  local out="$RUN_BASE/${name}_large"
+  local history="$ROOT/specialists/runs/${name}_large_history.json"
+  local log="$ROOT/specialists/runs/${name}_large.log"
+
+  mkdir -p "$out"
+  echo "=== $name (large) ===" | tee "$log"
+  echo "  train:  $train" | tee -a "$log"
+  echo "  eval:   $eval" | tee -a "$log"
+  echo "  model:  hidden=$HIDDEN layers=$LAYERS heads=$HEADS" | tee -a "$log"
+  echo "" | tee -a "$log"
+
+  .venv/bin/python3 run_grokking_until.py \
+    --dataset              "$train" \
+    --eval-dataset         "$eval" \
+    --out-dir              "$out" \
+    --labels               "$labels" \
+    --lr                   0.0005 \
+    --weight-decay         0.01 \
+    --max-steps            "$MAX_STEPS" \
+    --hidden-dim           "$HIDDEN" \
+    --layers               "$LAYERS" \
+    --heads                "$HEADS" \
+    --batch-size           256 \
+    --grok-loss-threshold  0.05 \
+    --grok-acc-jump        10.0 \
+    --stop-after-grokking-blocks 5 \
+    --morph-interval       99999 \
+    --reset \
+    --history-out          "$history" \
+    2>&1 | tee -a "$log"
+
+  echo "" | tee -a "$log"
+  echo "Done: $out/checkpoint.bin" | tee -a "$log"
+}
+
+# Run sequentially — one model at a time gets full GPU, avoids memory contention.
+# prereq_op — combined sprint-plan + project data (text-labeled)
+run_specialist "whetstone_prereq_op" \
+  "$ROOT/specialists/data/combined/prereq_op_train.tsv" \
+  "$ROOT/specialists/data/combined/prereq_op_eval.tsv" \
+  "standard,needs_review,needs_approval,full_gates"
+
+# worker_type
+run_specialist "whetstone_worker_type" \
+  "$ROOT/specialists/data/generated/worker_type_train.tsv" \
+  "$ROOT/specialists/data/generated/worker_type_eval.tsv" \
+  "implementer,reviewer,architect,qa"
+
+# verification_type
+run_specialist "whetstone_verification_type" \
+  "$ROOT/specialists/data/generated/verification_type_train.tsv" \
+  "$ROOT/specialists/data/generated/verification_type_eval.tsv" \
+  "unit,integration,schema,smoke,docs"
+
+# automatability
+run_specialist "whetstone_automatability" \
+  "$ROOT/specialists/data/generated/automatability_train.tsv" \
+  "$ROOT/specialists/data/generated/automatability_eval.tsv" \
+  "deterministic,template,specialist,slm,llm,human"
+echo ""
+echo "All large-model training runs complete."
diff --git a/specialists/scripts/train_automatability.sh b/specialists/scripts/train_automatability.sh
new file mode 100755
index 0000000..8cc2a0a
--- /dev/null
+++ b/specialists/scripts/train_automatability.sh
@@ -0,0 +1,64 @@
+#!/usr/bin/env bash
+# Train the automatability_strategy 6-way specialist.
+#
+# Classes (matching AutomatabilityAnnotation.strategy in Annotation.h):
+#   0 = deterministic  — pure formula/rule, zero parameters
+#   1 = template       — fixed-pattern code generation
+#   2 = specialist     — bounded probabilistic (~800KB model)
+#   3 = slm            — complex transformation, local model (~1-7GB)
+#   4 = llm            — full generative reasoning (14B+)
+#   5 = human          — requires manual judgment
+#
+# Input format: "name=ClassName"
+#
+# NOTE: 6-way classification with sparse tiers (human=5, specialist=8 raw
+# examples). Grokking may take longer — max-steps set to 10000.
+# If it doesn't grok by 10k, lower LR to 0.0005 and retry.
+#
+# Usage: bash specialists/scripts/train_automatability.sh
+#        (run from whetstone_DSL root)
+
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+FABRICATE_ROOT="/home/bill/Documents/WhetstoneAI_Fabricate"
+
+TRAIN_DATA="$ROOT/specialists/data/generated/automatability_train.tsv"
+EVAL_DATA="$ROOT/specialists/data/generated/automatability_eval.tsv"
+OUT_DIR="/mnt/storage/fabricate_runs/whetstone_automatability"
+HISTORY="$ROOT/specialists/runs/automatability_history.json"
+RUN_LOG="$ROOT/specialists/runs/automatability_training.log"
+
+mkdir -p "$ROOT/specialists/runs" "$OUT_DIR"
+
+if [[ ! -f "$TRAIN_DATA" ]]; then
+  echo "Missing training data. Run:" >&2
+  echo "  python3 specialists/scripts/gen_automatability_data.py" >&2
+  exit 1
+fi
+
+echo "Training automatability_strategy specialist..."
+echo "  train: $TRAIN_DATA"
+echo "  eval:  $EVAL_DATA"
+echo "  out:   $OUT_DIR"
+echo "  log:   $RUN_LOG"
+echo ""
+
+cd "$FABRICATE_ROOT"
+
+.venv/bin/python3 run_grokking_until.py \
+  --dataset        "$TRAIN_DATA" \
+  --eval-dataset   "$EVAL_DATA" \
+  --out-dir        "$OUT_DIR" \
+  --labels         "deterministic,template,specialist,slm,llm,human" \
+  --lr             0.001 \
+  --weight-decay   0.01 \
+  --max-steps      10000 \
+  --grok-loss-threshold   0.10 \
+  --grok-acc-jump         15.0 \
+  --stop-after-grokking-blocks 3 \
+  --history-out    "$HISTORY" \
+  2>&1 | tee "$RUN_LOG"
+
+echo ""
+echo "Done. Checkpoint at: $OUT_DIR/checkpoint.bin"
diff --git a/specialists/scripts/train_confidence_tier.sh b/specialists/scripts/train_confidence_tier.sh
new file mode 100755
index 0000000..26fd124
--- /dev/null
+++ b/specialists/scripts/train_confidence_tier.sh
@@ -0,0 +1,38 @@
+#!/usr/bin/env bash
+# Train the confidence_tier 3-way specialist.
+# Classes: high (conf>=80) / medium (60-79) / low (<60)
+# Input: "conflicts=N ambiguity=M deps=D prereqs=P queueready=yes/no"
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+FABRICATE_ROOT="/home/bill/Documents/WhetstoneAI_Fabricate"
+
+TRAIN_DATA="$ROOT/specialists/data/generated/confidence_tier_train.tsv"
+EVAL_DATA="$ROOT/specialists/data/generated/confidence_tier_eval.tsv"
+OUT_DIR="/mnt/storage/fabricate_runs/whetstone_confidence_tier"
+HISTORY="$ROOT/specialists/runs/confidence_tier_history.json"
+RUN_LOG="$ROOT/specialists/runs/confidence_tier_training.log"
+
+mkdir -p "$ROOT/specialists/runs" "$OUT_DIR"
+
+if [[ ! -f "$TRAIN_DATA" ]]; then
+  echo "Generating training data..." >&2
+  python3 "$ROOT/specialists/scripts/gen_confidence_tier_data.py" \
+    --train-out "$TRAIN_DATA" --eval-out "$EVAL_DATA"
+fi
+
+cd "$FABRICATE_ROOT"
+
+.venv/bin/python3 run_grokking_until.py \
+  --dataset        "$TRAIN_DATA" \
+  --eval-dataset   "$EVAL_DATA" \
+  --out-dir        "$OUT_DIR" \
+  --labels         "high,medium,low" \
+  --lr             0.001 \
+  --weight-decay   0.01 \
+  --max-steps      5000 \
+  --grok-loss-threshold   0.05 \
+  --grok-acc-jump         15.0 \
+  --stop-after-grokking-blocks 3 \
+  --history-out    "$HISTORY" \
+  2>&1 | tee "$RUN_LOG"
diff --git a/specialists/scripts/train_prereq_op.sh b/specialists/scripts/train_prereq_op.sh
new file mode 100755
index 0000000..a8eb813
--- /dev/null
+++ b/specialists/scripts/train_prereq_op.sh
@@ -0,0 +1,56 @@
+#!/usr/bin/env bash
+# Train the prereq_op_selector 4-way specialist.
+#
+# Classes:
+#   0 = standard       [validate-intake, resolve-dependencies]
+#   1 = needs_review   [validate-intake, architect-review]
+#   2 = needs_approval [validate-intake, manual-approval]
+#   3 = full_gates     [validate-intake, architect-review, manual-approval]
+#
+# Usage: bash specialists/scripts/train_prereq_op.sh
+#        (run from whetstone_DSL root)
+
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+FABRICATE_ROOT="/home/bill/Documents/WhetstoneAI_Fabricate"
+
+TRAIN_DATA="$ROOT/specialists/data/generated/prereq_op_train.tsv"
+EVAL_DATA="$ROOT/specialists/data/generated/prereq_op_eval.tsv"
+OUT_DIR="/mnt/storage/fabricate_runs/whetstone_prereq_op"
+HISTORY="$ROOT/specialists/runs/prereq_op_history.json"
+RUN_LOG="$ROOT/specialists/runs/prereq_op_training.log"
+
+mkdir -p "$ROOT/specialists/runs" "$OUT_DIR"
+
+if [[ ! -f "$TRAIN_DATA" ]]; then
+  echo "Missing training data. Run:" >&2
+  echo "  python3 specialists/scripts/gen_prereq_op_data.py" >&2
+  exit 1
+fi
+
+echo "Training prereq_op_selector specialist..."
+echo "  train: $TRAIN_DATA"
+echo "  eval:  $EVAL_DATA"
+echo "  out:   $OUT_DIR"
+echo "  log:   $RUN_LOG"
+echo ""
+
+cd "$FABRICATE_ROOT"
+
+.venv/bin/python3 run_grokking_until.py \
+  --dataset        "$TRAIN_DATA" \
+  --eval-dataset   "$EVAL_DATA" \
+  --out-dir        "$OUT_DIR" \
+  --labels         "standard,needs_review,needs_approval,full_gates" \
+  --lr             0.001 \
+  --weight-decay   0.01 \
+  --max-steps      5000 \
+  --grok-loss-threshold   0.10 \
+  --grok-acc-jump         15.0 \
+  --stop-after-grokking-blocks 3 \
+  --history-out    "$HISTORY" \
+  2>&1 | tee "$RUN_LOG"
+
+echo ""
+echo "Done. Checkpoint at: $OUT_DIR/checkpoint.bin"
diff --git a/specialists/scripts/train_specialist_pt.py b/specialists/scripts/train_specialist_pt.py
new file mode 100644
index 0000000..499107c
--- /dev/null
+++ b/specialists/scripts/train_specialist_pt.py
@@ -0,0 +1,328 @@
+#!/usr/bin/env python3
+"""
+train_specialist_pt.py
+
+PyTorch replacement for run_grokking_until.py + whetstone_fabricate.
+Trains a small transformer text classifier on TSV data.
+No SQLite — saves checkpoint.pt and history.json only.
+
+TSV format: label_idx<TAB>0<TAB>text
+
+Usage:
+    python3 specialists/scripts/train_specialist_pt.py \
+        --dataset    specialists/data/generated/worker_type_train.tsv \
+        --eval-dataset specialists/data/generated/worker_type_eval.tsv \
+        --out-dir    /tmp/runs/worker_type_small \
+        --labels     "implementer,reviewer,architect,qa" \
+        --hidden-dim 256 --layers 2 --heads 4 \
+        --lr 0.001 --weight-decay 0.01 \
+        --max-steps 3000 --batch-size 256 \
+        --history-out /tmp/runs/worker_type_small/history.json \
+        --reset
+"""
+
+import argparse
+import json
+import math
+import random
+import time
+from pathlib import Path
+
+import torch
+import torch.nn as nn
+from torch.utils.data import DataLoader, Dataset
+
+
+# ── Vocabulary ────────────────────────────────────────────────────────────────
+
+class Vocab:
+    PAD, UNK = 0, 1
+
+    def __init__(self):
+        self.str_to_id = {"<pad>": 0, "<unk>": 1}
+        self.id_to_str = ["<pad>", "<unk>"]
+
+    def build(self, texts):
+        for text in texts:
+            for tok in _tokenize(text):
+                if tok not in self.str_to_id:
+                    self.str_to_id[tok] = len(self.id_to_str)
+                    self.id_to_str.append(tok)
+
+    def encode(self, text, seq_len):
+        ids = [self.str_to_id.get(t, self.UNK) for t in _tokenize(text)]
+        ids = ids[:seq_len]
+        ids += [self.PAD] * (seq_len - len(ids))
+        return ids
+
+    def __len__(self):
+        return len(self.id_to_str)
+
+
+def _tokenize(text):
+    return text.lower().split()
+
+
+# ── Dataset ───────────────────────────────────────────────────────────────────
+
+class TSVDataset(Dataset):
+    def __init__(self, path, vocab, seq_len, label_list, build_vocab=False):
+        self.rows = []
+        texts = []
+        raw = []
+        with open(path) as f:
+            for line in f:
+                parts = line.strip().split("\t", 2)
+                if len(parts) < 3:
+                    continue
+                raw.append((int(parts[0]), parts[2]))
+                texts.append(parts[2])
+
+        if build_vocab:
+            vocab.build(texts)
+
+        for label_idx, text in raw:
+            ids = vocab.encode(text, seq_len)
+            self.rows.append((torch.tensor(ids, dtype=torch.long), label_idx))
+
+    def __len__(self):
+        return len(self.rows)
+
+    def __getitem__(self, i):
+        return self.rows[i]
+
+
+# ── Model ─────────────────────────────────────────────────────────────────────
+
+class SpecialistModel(nn.Module):
+    def __init__(self, vocab_size, hidden_dim, layers, heads, n_labels, seq_len):
+        super().__init__()
+        self.seq_len = seq_len
+        self.tok_emb = nn.Embedding(vocab_size, hidden_dim, padding_idx=0)
+        self.pos_emb = nn.Embedding(seq_len, hidden_dim)
+        encoder_layer = nn.TransformerEncoderLayer(
+            d_model=hidden_dim,
+            nhead=heads,
+            dim_feedforward=hidden_dim * 4,
+            dropout=0.0,
+            batch_first=True,
+        )
+        self.encoder = nn.TransformerEncoder(encoder_layer, num_layers=layers)
+        self.head = nn.Linear(hidden_dim, n_labels)
+        self._init_weights()
+
+    def _init_weights(self):
+        for p in self.parameters():
+            if p.dim() > 1:
+                nn.init.xavier_uniform_(p)
+
+    def forward(self, x):
+        # x: (B, seq_len)
+        pad_mask = (x == 0)  # True = ignore
+        pos = torch.arange(x.size(1), device=x.device).unsqueeze(0)
+        h = self.tok_emb(x) + self.pos_emb(pos)
+        h = self.encoder(h, src_key_padding_mask=pad_mask)
+        # Mean-pool over non-padding positions
+        lengths = (~pad_mask).float().sum(dim=1, keepdim=True).clamp(min=1)
+        h = (h * (~pad_mask).unsqueeze(-1).float()).sum(dim=1) / lengths
+        return self.head(h)
+
+
+# ── Training ──────────────────────────────────────────────────────────────────
+
+BLOCK = 1000  # steps per eval block (matches Fabricate convention)
+
+
+def evaluate(model, loader, device):
+    model.eval()
+    correct = total = 0
+    loss_sum = 0.0
+    criterion = nn.CrossEntropyLoss()
+    with torch.no_grad():
+        for x, y in loader:
+            x, y = x.to(device), y.to(device)
+            logits = model(x)
+            loss_sum += criterion(logits, y).item() * len(y)
+            correct += (logits.argmax(1) == y).sum().item()
+            total += len(y)
+    return correct / total * 100.0, loss_sum / max(total, 1)
+
+
+def train(args):
+    out_dir = Path(args.out_dir)
+    out_dir.mkdir(parents=True, exist_ok=True)
+
+    ckpt_path    = out_dir / "checkpoint.pt"
+    history_path = Path(args.history_out) if args.history_out else out_dir / "history.json"
+
+    if args.reset:
+        for f in [ckpt_path, history_path]:
+            f.unlink(missing_ok=True)
+
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    print(f"Device: {device}", flush=True)
+
+    label_list = [l.strip() for l in args.labels.split(",")]
+    n_labels = len(label_list)
+    seq_len = args.seq_len
+
+    # ── Vocab + data ──
+    vocab = Vocab()
+    train_ds = TSVDataset(args.dataset, vocab, seq_len, label_list, build_vocab=True)
+    eval_ds  = TSVDataset(args.eval_dataset, vocab, seq_len, label_list, build_vocab=False)
+
+    train_loader = DataLoader(train_ds, batch_size=args.batch_size, shuffle=True,
+                              pin_memory=(device.type == "cuda"), num_workers=0)
+    eval_loader  = DataLoader(eval_ds,  batch_size=args.batch_size, shuffle=False,
+                              pin_memory=(device.type == "cuda"), num_workers=0)
+
+    print(f"Train: {len(train_ds)} examples   Eval: {len(eval_ds)} examples", flush=True)
+    print(f"Vocab: {len(vocab)}   Labels: {label_list}", flush=True)
+
+    # ── Resume or fresh ──
+    start_step = 0
+    history_entries = []
+    best_acc = 0.0
+
+    model = SpecialistModel(
+        vocab_size=len(vocab),
+        hidden_dim=args.hidden_dim,
+        layers=args.layers,
+        heads=args.heads,
+        n_labels=n_labels,
+        seq_len=seq_len,
+    ).to(device)
+
+    optimizer = torch.optim.AdamW(
+        model.parameters(), lr=args.lr, weight_decay=args.weight_decay
+    )
+    criterion = nn.CrossEntropyLoss()
+
+    if ckpt_path.exists() and not args.reset:
+        ckpt = torch.load(ckpt_path, map_location=device)
+        model.load_state_dict(ckpt["model"])
+        optimizer.load_state_dict(ckpt["optimizer"])
+        vocab = ckpt["vocab_obj"]
+        start_step = ckpt.get("step", 0)
+        if history_path.exists():
+            history_entries = json.loads(history_path.read_text()).get("history", [])
+            best_acc = max((e.get("overall", 0.0) for e in history_entries), default=0.0)
+        print(f"Resuming from step {start_step}, best_acc={best_acc:.1f}%", flush=True)
+
+    n_params = sum(p.numel() for p in model.parameters())
+    print(f"Model: hidden={args.hidden_dim} layers={args.layers} heads={args.heads} "
+          f"params={n_params:,}", flush=True)
+
+    # ── Load all training data onto GPU once (tiny datasets — avoids DataLoader overhead) ──
+    all_x = torch.stack([row[0] for row in train_ds]).to(device)
+    all_y = torch.tensor([row[1] for row in train_ds], dtype=torch.long, device=device)
+    n_train = len(all_x)
+
+    # ── Grokking detection state ──
+    grokked = False
+    blocks_since_grok = 0
+    prev_block_loss = None
+    prev_block_acc  = 0.0
+
+    # ── Training loop ──
+    step = start_step
+    model.train()
+    block_start = time.perf_counter()
+
+    while step < args.max_steps:
+        idx = torch.randint(0, n_train, (args.batch_size,), device=device)
+        x, y = all_x[idx], all_y[idx]
+        optimizer.zero_grad()
+        loss = criterion(model(x), y)
+        loss.backward()
+        nn.utils.clip_grad_norm_(model.parameters(), 1.0)
+        optimizer.step()
+        step += 1
+
+        # ── Block boundary ──
+        if step % BLOCK == 0 or step == args.max_steps:
+            eval_acc, eval_loss = evaluate(model, eval_loader, device)
+            elapsed = time.perf_counter() - block_start
+
+            entry = {
+                "step": step,
+                "overall": round(eval_acc, 2),
+                "loss": round(eval_loss, 4),
+                "elapsed_s": round(elapsed, 1),
+            }
+            history_entries.append(entry)
+            history_path.write_text(json.dumps({"history": history_entries}, indent=2))
+
+            # Save checkpoint
+            torch.save({
+                "model": model.state_dict(),
+                "optimizer": optimizer.state_dict(),
+                "vocab_obj": vocab,
+                "vocab": vocab.str_to_id,
+                "config": {
+                    "hidden_dim": args.hidden_dim,
+                    "layers": args.layers,
+                    "heads": args.heads,
+                    "n_labels": n_labels,
+                    "seq_len": seq_len,
+                    "vocab_size": len(vocab),
+                },
+                "labels": label_list,
+                "step": step,
+            }, ckpt_path)
+
+            if eval_acc > best_acc:
+                best_acc = eval_acc
+
+            print(f"  step={step:6d}  acc={eval_acc:5.1f}%  loss={eval_loss:.4f}  "
+                  f"best={best_acc:.1f}%  {elapsed:.0f}s", flush=True)
+
+            # Grokking detection
+            if not grokked:
+                loss_drop = (prev_block_loss is not None and
+                             eval_loss < args.grok_loss_threshold)
+                acc_jump = (eval_acc - prev_block_acc) >= args.grok_acc_jump
+                if loss_drop or acc_jump:
+                    grokked = True
+                    print(f"  *** GROKKING detected at step={step} "
+                          f"({'loss_drop' if loss_drop else 'acc_jump'}) ***", flush=True)
+
+            if grokked:
+                blocks_since_grok += 1
+                if blocks_since_grok >= args.stop_after_grokking_blocks:
+                    print(f"  Stopping after {blocks_since_grok} post-grokking blocks.", flush=True)
+                    break
+
+            prev_block_loss = eval_loss
+            prev_block_acc  = eval_acc
+            block_start = time.perf_counter()
+            model.train()
+
+    print(f"\nDone. best_acc={best_acc:.1f}%  steps={step}  checkpoint={ckpt_path}", flush=True)
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--dataset",       required=True)
+    ap.add_argument("--eval-dataset",  required=True)
+    ap.add_argument("--out-dir",       required=True)
+    ap.add_argument("--labels",        required=True)
+    ap.add_argument("--lr",            type=float, default=0.001)
+    ap.add_argument("--weight-decay",  type=float, default=0.01)
+    ap.add_argument("--max-steps",     type=int,   default=10000)
+    ap.add_argument("--hidden-dim",    type=int,   default=256)
+    ap.add_argument("--layers",        type=int,   default=2)
+    ap.add_argument("--heads",         type=int,   default=4)
+    ap.add_argument("--batch-size",    type=int,   default=256)
+    ap.add_argument("--seq-len",       type=int,   default=96)
+    ap.add_argument("--history-out",   default=None)
+    ap.add_argument("--reset",         action="store_true")
+    ap.add_argument("--grok-loss-threshold",        type=float, default=0.05)
+    ap.add_argument("--grok-acc-jump",              type=float, default=10.0)
+    ap.add_argument("--stop-after-grokking-blocks", type=int,   default=4)
+    args = ap.parse_args()
+    train(args)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/specialists/scripts/train_verification_type.sh b/specialists/scripts/train_verification_type.sh
new file mode 100755
index 0000000..c528b22
--- /dev/null
+++ b/specialists/scripts/train_verification_type.sh
@@ -0,0 +1,54 @@
+#!/usr/bin/env bash
+# Train the verification_type 5-way specialist.
+#
+# Uses the WhetstoneAI_Fabricate binary + run_grokking_until.py harness.
+# The 5 classes map directly to the router architecture:
+#   label index 0 = unit, 1 = integration, 2 = schema, 3 = smoke, 4 = docs
+#
+# Usage: bash specialists/scripts/train_verification_type.sh
+#        (run from whetstone_DSL root)
+
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+FABRICATE_ROOT="/home/bill/Documents/WhetstoneAI_Fabricate"
+
+TRAIN_DATA="$ROOT/specialists/data/generated/verification_type_train.tsv"
+EVAL_DATA="$ROOT/specialists/data/generated/verification_type_eval.tsv"
+OUT_DIR="/mnt/storage/fabricate_runs/whetstone_verification_type"
+HISTORY="$ROOT/specialists/runs/verification_type_history.json"
+RUN_LOG="$ROOT/specialists/runs/verification_type_training.log"
+
+mkdir -p "$ROOT/specialists/runs" "$OUT_DIR"
+
+if [[ ! -f "$TRAIN_DATA" ]]; then
+  echo "Missing training data. Run:" >&2
+  echo "  python3 specialists/scripts/gen_verification_type_data.py" >&2
+  exit 1
+fi
+
+echo "Training verification_type specialist..."
+echo "  train: $TRAIN_DATA"
+echo "  eval:  $EVAL_DATA"
+echo "  out:   $OUT_DIR"
+echo "  log:   $RUN_LOG"
+echo ""
+
+cd "$FABRICATE_ROOT"
+
+.venv/bin/python3 run_grokking_until.py \
+  --dataset        "$TRAIN_DATA" \
+  --eval-dataset   "$EVAL_DATA" \
+  --out-dir        "$OUT_DIR" \
+  --labels         "unit,integration,schema,smoke,docs" \
+  --lr             0.001 \
+  --weight-decay   0.01 \
+  --max-steps      5000 \
+  --grok-loss-threshold   0.10 \
+  --grok-acc-jump         15.0 \
+  --stop-after-grokking-blocks 3 \
+  --history-out    "$HISTORY" \
+  2>&1 | tee "$RUN_LOG"
+
+echo ""
+echo "Done. Checkpoint at: $OUT_DIR/checkpoint.bin"
diff --git a/specialists/scripts/train_worker_type.sh b/specialists/scripts/train_worker_type.sh
new file mode 100755
index 0000000..19e57e1
--- /dev/null
+++ b/specialists/scripts/train_worker_type.sh
@@ -0,0 +1,56 @@
+#!/usr/bin/env bash
+# Train the worker_type 4-way specialist.
+#
+# Classes:
+#   0 = implementer  — builds/creates/extends features
+#   1 = reviewer     — validates/audits/checks existing work
+#   2 = architect    — designs/introduces new structure or frameworks
+#   3 = qa           — writes/runs tests, regression, benchmarks
+#
+# Usage: bash specialists/scripts/train_worker_type.sh
+#        (run from whetstone_DSL root)
+
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+FABRICATE_ROOT="/home/bill/Documents/WhetstoneAI_Fabricate"
+
+TRAIN_DATA="$ROOT/specialists/data/generated/worker_type_train.tsv"
+EVAL_DATA="$ROOT/specialists/data/generated/worker_type_eval.tsv"
+OUT_DIR="/mnt/storage/fabricate_runs/whetstone_worker_type"
+HISTORY="$ROOT/specialists/runs/worker_type_history.json"
+RUN_LOG="$ROOT/specialists/runs/worker_type_training.log"
+
+mkdir -p "$ROOT/specialists/runs" "$OUT_DIR"
+
+if [[ ! -f "$TRAIN_DATA" ]]; then
+  echo "Missing training data. Run:" >&2
+  echo "  python3 specialists/scripts/gen_worker_type_data.py" >&2
+  exit 1
+fi
+
+echo "Training worker_type specialist..."
+echo "  train: $TRAIN_DATA"
+echo "  eval:  $EVAL_DATA"
+echo "  out:   $OUT_DIR"
+echo "  log:   $RUN_LOG"
+echo ""
+
+cd "$FABRICATE_ROOT"
+
+.venv/bin/python3 run_grokking_until.py \
+  --dataset        "$TRAIN_DATA" \
+  --eval-dataset   "$EVAL_DATA" \
+  --out-dir        "$OUT_DIR" \
+  --labels         "implementer,reviewer,architect,qa" \
+  --lr             0.001 \
+  --weight-decay   0.01 \
+  --max-steps      5000 \
+  --grok-loss-threshold   0.10 \
+  --grok-acc-jump         15.0 \
+  --stop-after-grokking-blocks 3 \
+  --history-out    "$HISTORY" \
+  2>&1 | tee "$RUN_LOG"
+
+echo ""
+echo "Done. Checkpoint at: $OUT_DIR/checkpoint.bin"