Files
t2a_v2/docs/gold_debug/NUKE3_GOLD_TOP_ERRORS.md
dom cad0dd22b1 tests: alias DLBCL + garde-fou Trackare + e2e PDFs réels + gold CRH + benchmark enrichi
- 11 tests unitaires : TestAliasAndConclusionBonus (7) + TestTrackareSymptomGuard (4)
- Tests e2e sur PDFs réels (skip si absent) : méningite A87.0 + DLBCL C83.3 top1
- Gold CRH enrichi : 5 cas (2 réels ajoutés : 115_23066188, 132_23080179)
- Benchmark synthese : récupération conclusion depuis source_excerpt des DAS/traitements
- .gitignore : protection anti-PHI (real_crh_pdfs/, data/crh_samples/*.pdf)
- docs/PHI_POLICY.md : 7 règles de sécurité PHI
- Rapports debug : case 132 REVIEW (garde-fou actif), top errors, DIM pack

1043 tests passent, 0 régression.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-24 14:35:57 +01:00

705 B

NUKE-3 — Top erreurs gold CRH

Date : 2026-02-24 14:34
Cas : 5

# Case ID Choisi Attendu Strict Accept. Verdict Conf. Delta Reason
1 132_23080179 R59.0 C83.3 FAIL FAIL REVIEW medium 0 other
2 74_23141536 D50 I25.1 FAIL FAIL REVIEW medium 0.0 low_delta
3 115_23066188 A87.0 A87.0 OK OK CONFIRMED high 0 other
4 106_23056475 I26.9 I26.9 OK OK REVIEW medium 1.0 low_delta
5 73_23139637 R06.0 R06.0 OK OK REVIEW medium 1.0 mono_fragile

Généré le 2026-02-24 14:34