specialists: switch training to PyTorch, add capacity sweep infrastructure

- train_specialist_pt.py: pure PyTorch trainer replacing Fabricate binary.
  No SQLite/WAL — saves checkpoint.pt + history.json only. Fabricate was
  causing D-state IO blocking on the HDD due to continuous WAL writes.
- eval_specialist_pt.py: PyTorch eval with confusion matrix, threshold
  analysis, guardrail assessment, and deployment verdict.
- capacity_sweep.py: updated to call PyTorch scripts; TIERS now include
  heads field (4/6/8) since PyTorch supports multi-head unlike Fabricate.
- prereq_op_binary_split.py: evaluate_combined rewritten in PyTorch.
- Performance fix in train_specialist_pt.py: load training data onto GPU
  once and sample via torch.randint instead of DataLoader (eliminates
  Python/CPU overhead on tiny datasets). Fix applied, not yet timed.

Next session: time the fix, then run sweep_all_gates.sh --pilot-only.
See HANDOFF-2026-03-31.md for full context.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

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"