fix: Corrections qualité Phase 1 — 261 fuites en moins, 0 régression

Audit sur 30 fichiers aléatoires (OGC 12-690) révélant un overfitting
sur les 59 premiers OGC. Corrections appliquées avec test de non-régression
à chaque étape :

- NDA pieds de page Trackare : regex Episode N. (227→0 fuites)
- ONDANSETRON : word boundary \b sur RE_NUMERO_DOSSIER (32→0)
- RPPS isolés : détection 11 chiffres dans docs Trackare (3→0)
- Stop words : retrait noms réels (ute, dogue, cambo, bains), ajout
  termes médicaux (AINS, ponction, hanche, burkitt, ORL, GDS, OAP...)
- Pattern DR. Prénom NOM : capture prénoms médecins (Ute ×19, Tam...)
- force_names : contextes structurés (DR., Signé, Note d'évolution)
  bypassent les stop words pour masquer les vrais noms de soignants
- Phase 2b : PiiHit trackare (EPISODE, RPPS) appliqués au texte .txt
- Framework de non-régression (regression_tests/) + batch audit 30 fichiers

Résultat : 322→61 fuites détectées, 113→109 faux positifs, 0 régression.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
2026-03-06 17:32:28 +01:00
parent f9532d5543
commit bc2fe667a0
63 changed files with 30187 additions and 24 deletions

View File

@@ -0,0 +1,22 @@
{"page": 0, "kind": "ETAB", "original": "Pôle Spécialités Médicales", "placeholder": "[MASK]", "bbox_hint": null}
{"page": 0, "kind": "ETAB", "original": "Service Hématologie", "placeholder": "[MASK]", "bbox_hint": null}
{"page": 0, "kind": "NOM", "original": "Anne BANOS", "placeholder": "[NOM]", "bbox_hint": null}
{"page": 0, "kind": "ETAB", "original": "Clinique des Hôpitaux de Bordeaux Docteur Pierre COUDANNE", "placeholder": "[ETABLISSEMENT]", "bbox_hint": null}
{"page": 0, "kind": "CODE_POSTAL", "original": "64240 URT", "placeholder": "[CODE_POSTAL]", "bbox_hint": null}
{"page": 0, "kind": "NOM", "original": "Frédéric BAUDUER", "placeholder": "[NOM]", "bbox_hint": null}
{"page": 0, "kind": "NOM", "original": "Sophie BERNARD", "placeholder": "[NOM]", "bbox_hint": null}
{"page": 0, "kind": "DATE_NAISSANCE", "original": "né le 20/02/1981", "placeholder": "[DATE_NAISSANCE]", "bbox_hint": null}
{"page": 0, "kind": "NOM", "original": "COUDANNE Pierre", "placeholder": "[NOM]", "bbox_hint": null}
{"page": 0, "kind": "NOM", "original": "Louise BOUARD", "placeholder": "[NOM]", "bbox_hint": null}
{"page": 0, "kind": "NOM", "original": "Claudie CAPDUPUY", "placeholder": "[NOM]", "bbox_hint": null}
{"page": 0, "kind": "NOM", "original": "Julie GAY Lumbago", "placeholder": "[NOM]", "bbox_hint": null}
{"page": 0, "kind": "NOM", "original": "Marielle LE-MECHEC", "placeholder": "[NOM]", "bbox_hint": null}
{"page": 0, "kind": "NOM", "original": "Corinne MONNIER", "placeholder": "[NOM]", "bbox_hint": null}
{"page": 0, "kind": "NOM", "original": "Laurent DAVID", "placeholder": "[NOM]", "bbox_hint": null}
{"page": 0, "kind": "EMAIL", "original": "secr.hema@ch-cotebasque.fr", "placeholder": "[EMAIL]", "bbox_hint": null}
{"page": 0, "kind": "EMAIL", "original": "hospi.hemato@ch-cotebasque.fr", "placeholder": "[EMAIL]", "bbox_hint": null}
{"page": 1, "kind": "DATE_NAISSANCE", "original": "né le 20/02/1981", "placeholder": "[DATE_NAISSANCE]", "bbox_hint": null}
{"page": 1, "kind": "NOM", "original": "COUDANNE Pierre", "placeholder": "[NOM]", "bbox_hint": null}
{"page": 2, "kind": "DATE_NAISSANCE", "original": "né le 20/02/1981", "placeholder": "[DATE_NAISSANCE]", "bbox_hint": null}
{"page": 2, "kind": "NOM", "original": "COUDANNE Pierre", "placeholder": "[NOM]", "bbox_hint": null}
{"page": 2, "kind": "NOM", "original": "Sophie BERNARD", "placeholder": "[NOM]", "bbox_hint": null}