30 lines
976 B
Markdown
30 lines
976 B
Markdown
|
# Sprint 186 Plan: Execution-Specific Scoring Calibration I
|
||
|
|
|
||
|
|
## Context
|
||
|
|
`executionSpecificityScore` exists but calibration is shallow.
|
||
|
|
|
||
|
|
## Goals
|
||
|
|
1. Build calibration corpus from recent taskitem runs
|
||
|
|
2. Measure false-ready and false-blocked rates
|
||
|
|
3. Define scoring threshold candidates by gap class
|
||
|
|
|
||
|
|
## Steps
|
||
|
|
|
||
|
|
### Step 1969: Calibration corpus loader (10 tests)
|
||
|
|
Aggregate historical taskitem artifacts into labeled calibration records.
|
||
|
|
|
||
|
|
### Step 1970: Score-vs-outcome analyzer (10 tests)
|
||
|
|
Compute score distributions across green vs blocked outcomes.
|
||
|
|
|
||
|
|
### Step 1971: Threshold recommendation engine (8 tests)
|
||
|
|
Emit deterministic threshold candidates for strict mode.
|
||
|
|
|
||
|
|
### Step 1972: Misclassification report export (8 tests)
|
||
|
|
Write machine-readable false-positive/false-negative reports.
|
||
|
|
|
||
|
|
### Step 1973: Sprint 186 Integration Summary (8 tests)
|
||
|
|
Add `editor/src/Sprint186IntegrationSummary.h`.
|
||
|
|
|
||
|
|
## Architecture Gate
|
||
|
|
- Calibration outputs must be reproducible and artifact-backed.
|