Skip to content

docs(sprint1): METRICS_CURRENT.md + CI stale-metric detector + breaking news cleanup - #106

Merged
cschanhniem merged 2 commits into
mainfrom
feat/sprint1-trust-cleanup
Jun 29, 2026
Merged

cschanhniem merged 2 commits into
mainfrom
feat/sprint1-trust-cleanup

Conversation

@cschanhniem

Copy link
Copy Markdown
Collaborator

Sprint 1 — Trust cleanup

3 deliverables:

1. docs/METRICS_CURRENT.md

Single source of truth for current pipeline metrics:

  • AUROC 0.7832 (CI₉₅: 0.72–0.84, n=191), Phase3 0.7448, AUPRC 0.8164
  • Historical baselines table (demo set = 0.8420)
  • Candidate panel summary, external predictor status, test suite stats
  • Key limitations

2. CI stale-metric detector (tests/test_docs_consistent.py)

5 tests that auto-fail on stale docs:

  • Stale AUROC in current context → FAIL
  • METRICS_CURRENT.md missing → FAIL
  • 'breaking news' terminology → FAIL
  • 92-97% overconfident claims → FAIL
  • Missing honest correction in probability doc → FAIL

3. Residual stale refs

All remaining 'breaking news' → 'high-impact scenario' across 5 docs.
All stale 92-97% table entries updated with honest corrections.

Tests: 1343 total (+5 doc consistency tests)

cschanhniem and others added 2 commits June 29, 2026 09:33
…reaking news cleanup

Sprint 1 — Trust cleanup:

1. docs/METRICS_CURRENT.md: Single source of truth for all current pipeline metrics
   (AUROC 0.7832, n=191, phase3 AUROC 0.7448, etc.). Includes historical baselines,
   candidate panel summary, external predictor status, test suite stats, and limitations.

2. tests/test_docs_consistent.py: CI stale-metric detector — 5 tests that fail if:
   - Any doc uses stale AUROC in current context (not historical)
   - METRICS_CURRENT.md is missing or inconsistent
   - "breaking news" terminology remains (must be "high-impact scenario")
   - Overconfident 92-97% claims appear without honesty correction
   - WET_LAB_PROBABILITY.md lacks honest lower-bound estimate

3. Residual stale refs fixed across 5 docs (ROADMAP.md, DISCOVERY_PREDICTION.md,
   WET_LAB_PROBABILITY.md, WAVE2_PLAN.md, DISCOVERY_PREDICTION.md)
   - All "breaking news" → "high-impact scenario" (including headers and sidebars)
   - All stale 92-97% table entries with honest corrections
   - ROADMAP.md v0.4 entry qualified as "original demo set"

Total: 1343 tests (5 new doc consistency tests).

Co-authored-by: CommandCodeBot <noreply@commandcode.ai>
…g note

Three HIGH issues fixed:
1. Hyphenated "breaking-news" in DISCOVERY_PREDICTION.md (2 occurrences)
2. Test regex updated to catch both space-separated and hyphenated forms
3. Tracking note self-reference fixed

Co-authored-by: CommandCodeBot <noreply@commandcode.ai>
@cschanhniem
cschanhniem merged commit 2a69331 into main Jun 29, 2026
1 of 2 checks passed
@cschanhniem
cschanhniem deleted the feat/sprint1-trust-cleanup branch June 29, 2026 02:41
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant