whetstone_DSL/sprint187_plan.md

Sprint 187 Plan: Execution-Specific Scoring Calibration II

Context

Calibration must generalize beyond single-sprint corpora.

Goals

1. Add cross-project calibration datasets
2. Tune score penalties for weak grounding
3. Validate threshold stability across corpora

Steps

Step 1974: Cross-project calibration ingestion (10 tests)

Load calibration data from sprint plans and run-spec families.

Step 1975: Penalty tuning for generic taskitems (10 tests)

Tune penalties for weak file/tool/test grounding.

Step 1976: Stability checker across datasets (8 tests)

Compare threshold drift across cohorts.

Step 1977: Deterministic threshold profile selection (8 tests)

Select strict profile with deterministic tie-breaking.

Step 1978: Sprint 187 Integration Summary (8 tests)

Add editor/src/Sprint187IntegrationSummary.h.

Architecture Gate

• Selected thresholds must remain stable across at least three dataset cohorts.