Loop 77: Phase F F7 — calibration link from negative-result entries to intake reports

docs/evidence/METRICS_CURRENT.md

@@ -5,7 +5,9 @@ Machine-readable snapshot: `outputs/metrics_snapshot.json` regenerated with `mak
 > **Purpose:** One authoritative table of current pipeline metrics. If any doc disagrees
 > with this file, this file wins. Updated whenever benchmark/benchmark config changes.
 >
-> **Last updated:** 2026-07-09 (Phase F negative-result informativeness — v0.5.77)
+> **Last updated:** 2026-07-09 (Phase F F7 — calibration link from negative-result entries to intake reports — v0.5.78)
+> **New in v0.5.78:** Calibration link from negative-result entries to intake reports (F7) — closes learning loop by tracing each negative-result entry back to its prediction-vs-actual data in the calibration intake report. Schema addition: `intake_report_id` optional field on `negative_result_entry.schema.json`. CLI (`scripts/link_negative_result_to_intake.py`) reads negative-result archive JSON + calibration intake report, links by candidate_id, reports matched/unmatched entries, orphan intake candidates, lab summary for matched entries, and validates intake_report_id references. Produces structured JSON + Markdown link report. 25 tests. 2822 total.
+>
 > **New in v0.5.77:** Negative-result informativeness guide (F6) — comprehensive documentation defining 7-dimension informativeness framework for negative-result entries, with 14 examples across all 6 reason categories, before/after transformation pairs, and quick-reference checklist. CLI classifier (`scripts/classify_negative_result_informativeness.py`) that scores entries on 7 dimensions and produces informative/neutral/non-informative classification. 37 tests. 2796 total.
 >
 > **New in v0.5.76:** Safe-publication filter (F5) — CLI that reads candidate panel data with safety metadata and checks each candidate against publication safety constraints before external release. FAIL-CLOSED: defaults to reject unless all checks pass. Checks: dry_lab_only (must be true), proof_ladder_level (≤4), toxicity_screened, hemolysis_screened, dual_use_reviewed (all must be true). Produces structured JSON filter result and human-readable Markdown summary. 33 tests. 2758 total.
@@ -64,7 +66,7 @@ Machine-readable snapshot: `outputs/metrics_snapshot.json` regenerated with `mak
 > **New in v0.5.31:** Added dipeptide-order features for sequence-order awareness. `dipeptide_order_score` achieves AUROC 0.7861 on AMP-vs-scrambled discrimination — the strongest order-dependent feature in the pipeline. Only 7/31 features survive scrambling (amphipathicity/helix-wheel + dipeptide). All composition features are purely position-independent (exactly 0.5000 AUROC on scrambled test).
 > **New in v0.5.30:** Easy baseline benchmark added — charge density alone (AUROC 0.8166) outperforms the full pipeline ensemble (0.7792) on AMP-vs-Swiss-Prot-decoy discrimination. Honest finding documented: expected because pipeline optimizes for safety, not raw discrimination.
 > **New in v0.5.29:** Expanded benchmark to 500 AMPs + 500 composition-matched decoys (n=1000). AUROC 0.7792 (CI₉₅: 0.7505–0.8065) confirms signal generalizes. Cluster-aware CI: 0.746–0.8102. Representative AUROC: 0.778. Standard benchmark (n=191) retained for backward comparison.
-> **Pipeline version:** v0.5.77
+> **Pipeline version:** v0.5.78
 > **Branch:** main
 
 ---

docs/research/ROADMAP.md

@@ -1,5 +1,44 @@
 # Roadmap
 
+## v0.5.78 — Loop 77: Phase F F7 — Calibration Link from Negative-Result Entries to Intake Reports ✓ (2026-07-09)
+
+Closes the learning loop by tracing each negative-result entry back to its
+prediction-vs-actual data in the calibration intake report. Schema, CLI, and tests.
+
+Changes:
+- `schemas/negative_result_entry.schema.json` — Added `intake_report_id` optional
+  field (string) to link negative-result entries back to their calibration intake
+  report source.
+- `scripts/link_negative_result_to_intake.py` (F7) — CLI that reads a negative-result
+  archive JSON and a calibration intake report JSON, links entries by candidate_id,
+  and produces a structured link report with: matched/unmatched counts, linkage rate,
+  orphan intake detection, per-candidate predictions + lab summary for matched entries,
+  intake_report_id validation, control failures and cohort metrics from the intake report.
+  Supports `--out-json` and `--out-md` output.
+- `tests/evidence/test_link_negative_result_to_intake.py` — 25 tests covering: linkage
+  counts and rate, matched entries with predictions, unmatched entries without predictions,
+  orphan intake candidate detection, intake_report_id validation (valid + invalid),
+  by-reason-category aggregation, metadata and caveat generation, lab summary in linked
+  entries, markdown section content, empty entries handling, all-unmatched case,
+  file loading error paths, CLI output (JSON + Markdown + both), missing input and
+  empty entries error paths, control failures and cohort metrics forwarding, list-format
+  input support, lab summary absent when no lab result.
+- `docs/evidence/METRICS_CURRENT.md` — v0.5.78 changelog. Pipeline version bumped.
+  Test count: 2822.
+- `tests/test_test_count_regression.py` — baseline updated to 2822.
+
+Honest boundaries:
+- Linkage is by candidate_id only. Entries without a matching intake row are reported
+  as unmatched but are not invalidated.
+- Orphan intake candidates (in intake but not in negative-result archive) may be active,
+  pending, or withdrawn — absence of a negative entry does not imply success.
+- The intake_report_id field is advisory and does not affect linkage logic; validation
+  reports mismatches but does not block report generation.
+- The link report is informational only and requires qualified review before any claim
+  about candidate quality or pipeline performance.
+- Cohort metrics and control failures are forwarded from the intake report; this script
+  does not independently validate them.
+
 ## v0.5.77 — Loop 76: Phase F F6 — Negative-Result Informativeness Guide ✓ (2026-07-09)
 
 Comprehensive guide defining informative vs non-informative negative results with

schemas/negative_result_entry.schema.json

@@ -27,6 +27,7 @@
     "score_novelty": {"type": "number", "minimum": 0, "maximum": 1},
     "score_ensemble": {"type": "number", "minimum": 0, "maximum": 1},
     "recalibration_used": {"type": "string", "enum": ["yes", "no"]},
+    "intake_report_id": {"type": "string", "description": "Optional reference to the calibration intake report that contains this candidate's prediction-vs-actual data"},
     "superseded_by": {"type": "integer"},
     "reviewer_notes": {"type": "string"}
   }