Files
whetstone_DSL/sprint29_plan.md

2.3 KiB

Sprint 29 Plan: Workflow Intelligence + Cost Discipline

Context

Sprint 29 tightens orchestration intelligence around constrained editing: better routing, better context bundles, and explicit cost-control policies.

Primary goals:

  1. Context width minimization without quality regression
  2. Routing confidence calibration for constrained mode
  3. Cost policy enforcement and optimization feedback

Phase 29a: Routing and Context Optimization (Steps 544-548)

Step 544: Constrained Routing Ruleset Extension (12 tests)

  • Add constrained-mode routing policy variants
  • Promote deterministic/template paths when legal-choice confidence is high

Step 545: Context Bundle Minimizer (12 tests)

  • Build minimal sufficient context bundles per taskitem contract
  • Remove irrelevant file/project context from narrow operations

Step 546: Confidence Calibration for Constrained Paths (12 tests)

  • Calibrate routing confidence with historical constraint outcomes
  • Penalize paths with repeated post-apply failures

Step 547: Cost Policy Guard (12 tests)

  • Enforce per-step and per-workflow token ceilings
  • Require rationale/escalation when exceeding policy

Step 548: Phase 29a Integration (8 tests)

  • Lower context budgets with maintained pass rates
  • Cost guard blocks runaway orchestration paths

Phase 29b: Optimization Feedback Loops (Steps 549-553)

Step 549: Suggestion Engine for Cost Reductions (12 tests)

  • Suggest template substitution, batching, and context narrowing opportunities

Step 550: Worker Efficiency Dashboard Data Model (12 tests)

  • Aggregate completion/cost/rejection/latency by worker and task class

Step 551: Review-Gate Policy Refinement (12 tests)

  • Reduce unnecessary human escalations while preserving safety/quality

Step 552: Cost vs Quality Regression Suite (12 tests)

  • Assert quality does not regress as optimization policies tighten

Step 553: Sprint 29 Integration + Summary (8 tests)

  • Routing intelligence and cost controls operate as a coherent policy layer

Step & Test Summary

Phase Steps Tests Theme
29a 544-548 56 Routing/context/cost policy optimization
29b 549-553 56 Feedback loops and quality-preserving optimization
Total 544-553 ~112 10 steps