[DTP-Worker] Sprint 8 · Générateur Matrice d'acceptation / traçabilité MVP (15 promesses roadmap · 21 modules gated tracés bijectif · partition sprint exacte · 12 hors-périmètre sourcés) (QA · roadmap L75/L80-87)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

05_deliverables_mvp/qa/acceptance/.gitignore

@@ -0,0 +1,2 @@
+__pycache__/
+*.pyc

05_deliverables_mvp/qa/acceptance/README.md

@@ -0,0 +1,75 @@
+# Matrice d'acceptation / traçabilité MVP — `qa/acceptance`
+
+**Sprint 8 · QA (buffer L75 · corrections finales / recette).** Document de
+**RECETTE** de méta-niveau : il mappe chaque **promesse** de la roadmap
+(`04_roadmap/ROADMAP_8_WEEKS_OR_LESS.md`) vers sa **preuve** livrée, et prouve
+qu'aucune promesse n'est silencieusement oubliée.
+
+- **8 livrables de sprint** (L33..L76) → modules gated qui les réalisent en-repo.
+- **7 métriques succès MVP** (L81..L87) → preuve gated ou hors-périmètre sourcé.
+
+## Question de recette à laquelle il répond
+
+> « Le MVP promis est-il honoré, et ce qui reste est-il **explicitement** hors
+> périmètre worker (VPS #8 · builds stores · services runtime) ? »
+
+## Pourquoi un module distinct (non redondant · #5)
+
+| Harnais méta | Axe prouvé |
+|---|---|
+| `qa/audit_4big` | qualité **statique** de chaque module (95+/100) |
+| `qa/regression` | chaque suite **s'exécute** au vert |
+| `devops/deploy_runbook` | **ordre** de déploiement VPS des modules |
+| **`qa/acceptance`** (ici) | **couverture des promesses** roadmap (la recette) |
+
+## Périmètre PROUVÉ, pas déclaré (anti-invention · #6)
+
+- L'ensemble des **modules-preuve** est **dérivé du CI** (`q4lib.registry.parse_ci`,
+  réutilisé · zéro duplication) et confronté de façon **BIJECTIVE** à la réalité :
+  un module gated **non tracé** OU une preuve **non gated** → génération **refusée**.
+- La **fenêtre de sprint** de chaque module est **lue** dans le registre de
+  l'auditeur 4Big (`qa/audit_4big/quality_spec.json`) — jamais re-déclarée ici
+  (anti-dérive). L'auditeur lui-même (absent de son propre registre · SoD) reçoit
+  sa fenêtre via `extra_module_sprint`, **avec source**.
+- **Partition exacte** par sprint : chaque livrable de sprint SX cite exactement
+  les modules gated de fenêtre SX (ni trou ni chevauchement).
+- Les **nombres** des énoncés (`<1h`, `95/100`, `7 dashboards`, `2 stores`) sont
+  des **citations verbatim** de la roadmap — la matrice ne **prétend** aucune
+  mesure de performance.
+- Tout **hors-périmètre** porte une **source** (`CLAUDE.md #8` VPS, builds stores,
+  services runtime) : aucune omission silencieuse.
+
+## Usage
+
+```bash
+python3 acceptance_gen.py build      # écrit out/acceptance_matrix.json + out/MANIFEST.json
+python3 acceptance_gen.py validate   # (re)génère en mémoire + valide (schéma + 10 invariants)
+python3 -m unittest discover -s tests -p 'test_*.py' -v
+```
+
+## Sorties (`out/`)
+
+- `acceptance_matrix.json` — une ligne par promesse : preuves modules résolues
+  (avec job CI), artefacts, hors-périmètre sourcé, drapeaux + **verdict** global.
+- `MANIFEST.json` — comptes, preuve de **couverture bijective** vs CI, **partition
+  par sprint**.
+
+Sortie **déterministe** (ordre du spec, listes triées, aucun horodatage).
+
+## Invariants (10 familles)
+
+1. ids uniques ; livrables S1..S8 + métriques M1..M7 ordonnés.
+2. toute preuve module est **gated** (drapeau recalculé depuis le CI).
+3. **couverture bijective** cités ⇔ gated (hors self) — cœur anti-invention.
+4. fenêtre de sprint connue pour tout module gated ; extra sans conflit registre.
+5. **partition exacte** par sprint.
+6. tout hors-périmètre porte une source non vide.
+7. cohérence statut/preuve ; tout artefact cité **existe** sur disque.
+8. `mvp_metric` ⇒ sprint nul ; `sprint_deliverable` ⇒ sprint renseigné.
+9. **SoD** : la matrice ne se cite jamais elle-même comme preuve.
+10. `roadmap_line` entier positif.
+
+## Hors périmètre worker (#8)
+
+Ce module **ne déploie rien** : il produit un document de recette en-repo.
+L'application réelle (démo publique, VPS) reste à l'agent DevOps / la direction.

05_deliverables_mvp/qa/acceptance/acceptance.schema.json

@@ -0,0 +1,188 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "$id": "https://oto-enterprise-os-dtp/qa/acceptance.schema.json",
+  "title": "Matrice d'acceptation / traçabilité MVP (roadmap → livrable → preuve)",
+  "description": "Contrat de sortie de acceptance_gen.py. Validé par le validateur maison Publiciste (zéro pip) dans le job CI qa-acceptance-tests. Mappe chaque promesse roadmap (livrables de sprint + métriques succès MVP) vers sa preuve gated (module CI) ou un hors-périmètre sourcé. Aucun chiffre métier fabriqué : les nombres des énoncés sont des citations verbatim de la roadmap.",
+  "type": "object",
+  "additionalProperties": false,
+  "required": ["matrix", "manifest", "verdict"],
+  "properties": {
+    "matrix": {
+      "type": "array",
+      "minItems": 1,
+      "items": { "$ref": "#/definitions/row" }
+    },
+    "verdict": { "type": "boolean" },
+    "manifest": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": [
+        "generated_from",
+        "reference_cadre",
+        "roadmap_ref",
+        "cible_portage",
+        "self_module",
+        "counts",
+        "coverage",
+        "roadmap"
+      ],
+      "properties": {
+        "generated_from": { "type": "string", "minLength": 1 },
+        "reference_cadre": { "type": "string", "minLength": 1 },
+        "roadmap_ref": { "type": "string", "minLength": 1 },
+        "cible_portage": { "type": "string", "minLength": 1 },
+        "self_module": { "type": "string", "const": "qa/acceptance" },
+        "counts": {
+          "type": "object",
+          "additionalProperties": false,
+          "required": [
+            "sprint_deliverables",
+            "mvp_metrics",
+            "deliverables_total",
+            "in_repo",
+            "out_of_scope_full",
+            "gated_modules",
+            "cited_modules",
+            "out_of_scope_items"
+          ],
+          "properties": {
+            "sprint_deliverables": { "type": "integer", "minimum": 1 },
+            "mvp_metrics": { "type": "integer", "minimum": 1 },
+            "deliverables_total": { "type": "integer", "minimum": 1 },
+            "in_repo": { "type": "integer", "minimum": 0 },
+            "out_of_scope_full": { "type": "integer", "minimum": 0 },
+            "gated_modules": { "type": "integer", "minimum": 1 },
+            "cited_modules": { "type": "integer", "minimum": 0 },
+            "out_of_scope_items": { "type": "integer", "minimum": 0 }
+          }
+        },
+        "coverage": {
+          "type": "object",
+          "additionalProperties": false,
+          "required": [
+            "ci_modules_count",
+            "cited_modules_count",
+            "bijective",
+            "missing_in_matrix",
+            "phantom_evidence",
+            "partition_ok",
+            "sprint_partition",
+            "unknown_sprint",
+            "missing_artifacts",
+            "self_module_excluded"
+          ],
+          "properties": {
+            "ci_modules_count": { "type": "integer", "minimum": 0 },
+            "cited_modules_count": { "type": "integer", "minimum": 0 },
+            "bijective": { "type": "boolean" },
+            "missing_in_matrix": { "type": "array", "items": { "type": "string" } },
+            "phantom_evidence": { "type": "array", "items": { "type": "string" } },
+            "partition_ok": { "type": "boolean" },
+            "sprint_partition": {
+              "type": "array",
+              "items": { "$ref": "#/definitions/partition" }
+            },
+            "unknown_sprint": { "type": "array", "items": { "type": "string" } },
+            "missing_artifacts": { "type": "array", "items": { "type": "string" } },
+            "self_module_excluded": { "type": "string", "const": "qa/acceptance" }
+          }
+        },
+        "roadmap": {
+          "type": "object",
+          "additionalProperties": false,
+          "required": ["sprint_ids", "metric_ids"],
+          "properties": {
+            "sprint_ids": { "type": "array", "items": { "type": "string" } },
+            "metric_ids": { "type": "array", "items": { "type": "string" } }
+          }
+        }
+      }
+    }
+  },
+  "definitions": {
+    "row": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": [
+        "id",
+        "kind",
+        "sprint",
+        "roadmap_line",
+        "statement",
+        "status",
+        "modules",
+        "artifacts",
+        "out_of_scope",
+        "n_modules",
+        "all_modules_gated",
+        "all_artifacts_exist",
+        "has_evidence"
+      ],
+      "properties": {
+        "id": { "type": "string", "minLength": 1 },
+        "kind": { "type": "string", "enum": ["sprint_deliverable", "mvp_metric"] },
+        "sprint": { "type": ["string", "null"] },
+        "roadmap_line": { "type": "integer", "minimum": 1 },
+        "statement": { "type": "string", "minLength": 1 },
+        "status": { "type": "string", "enum": ["in_repo", "out_of_scope"] },
+        "modules": {
+          "type": "array",
+          "items": { "$ref": "#/definitions/evmodule" }
+        },
+        "artifacts": {
+          "type": "array",
+          "items": { "$ref": "#/definitions/evartifact" }
+        },
+        "out_of_scope": {
+          "type": "array",
+          "items": { "$ref": "#/definitions/oos" }
+        },
+        "n_modules": { "type": "integer", "minimum": 0 },
+        "all_modules_gated": { "type": "boolean" },
+        "all_artifacts_exist": { "type": "boolean" },
+        "has_evidence": { "type": "boolean" }
+      }
+    },
+    "evmodule": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["module", "ci_job", "gated"],
+      "properties": {
+        "module": { "type": "string", "minLength": 1 },
+        "ci_job": { "type": ["string", "null"] },
+        "gated": { "type": "boolean" }
+      }
+    },
+    "evartifact": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["path", "exists"],
+      "properties": {
+        "path": { "type": "string", "minLength": 1 },
+        "exists": { "type": "boolean" }
+      }
+    },
+    "oos": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["item", "source"],
+      "properties": {
+        "item": { "type": "string", "minLength": 1 },
+        "source": { "type": "string", "minLength": 1 }
+      }
+    },
+    "partition": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["sprint", "expected", "cited", "missing", "extra", "exact"],
+      "properties": {
+        "sprint": { "type": "string", "minLength": 1 },
+        "expected": { "type": "array", "items": { "type": "string" } },
+        "cited": { "type": "array", "items": { "type": "string" } },
+        "missing": { "type": "array", "items": { "type": "string" } },
+        "extra": { "type": "array", "items": { "type": "string" } },
+        "exact": { "type": "boolean" }
+      }
+    }
+  }
+}

05_deliverables_mvp/qa/acceptance/acceptance_gen.py

@@ -0,0 +1,282 @@
+#!/usr/bin/env python3
+"""Matrice d'acceptation / traçabilité MVP (Sprint 8 · QA · buffer L75).
+
+Document de RECETTE de MÉTA-NIVEAU : il mappe chaque PROMESSE de la roadmap
+(8 livrables de sprint L33..L76 + 7 métriques succès MVP L81..L87) vers sa PREUVE
+livrée (un module gated du CI) OU un hors-périmètre worker SOURCÉ. Il répond à la
+seule question de recette : « le MVP promis est-il honoré, et ce qui reste
+est-il explicitement hors périmètre worker ? »
+
+Axe distinct des autres harnais méta (non redondant · #5) :
+  - qa/audit_4big         note la QUALITÉ statique de chaque module ;
+  - qa/regression         prouve que chaque suite S'EXÉCUTE au vert ;
+  - devops/deploy_runbook ORDONNE le déploiement VPS des modules ;
+  - qa/acceptance (ici)   prouve la COUVERTURE des promesses roadmap.
+
+Périmètre PROUVÉ, pas déclaré : l'ensemble des modules-preuve est confronté à
+`.gitea/workflows/ci.yml` (réutilise `q4lib.registry.parse_ci` — zéro duplication
+· #5) de façon BIJECTIVE, et la fenêtre de sprint de chaque module est LUE dans
+le registre de l'auditeur 4Big (anti-dérive). Tout module gated non tracé OU tout
+module tracé non gated → génération refusée.
+
+Anti-invention (#6) : les nombres des énoncés (<1h, 95/100, 7 dashboards, 2
+stores) sont des CITATIONS verbatim de la roadmap ; la matrice ne PRÉTEND aucune
+mesure de performance ; tout hors-périmètre porte une source.
+
+Produit :
+  - acceptance_matrix.json → la matrice (une ligne par promesse : preuves
+                             résolues, hors-périmètre sourcé, drapeaux) + verdict.
+  - MANIFEST.json          → manifeste agrégé (comptes, preuve de couverture
+                             bijective vs CI, partition par sprint).
+
+Sous-commandes :
+  build    [-o OUT]  → écrit acceptance_matrix.json / MANIFEST.json (après validation)
+  validate [-o OUT]  → (re)génère en mémoire, valide vs acceptance.schema.json +
+                       invariants — sort en erreur si un invariant casse.
+
+Sortie déterministe (ordre du spec, listes triées, aucun horodatage).
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import sys
+
+_HERE = os.path.dirname(os.path.abspath(__file__))
+sys.path.insert(0, _HERE)
+
+from acclib import builder, deps  # noqa: E402
+
+_SPEC_PATH = os.path.join(_HERE, "acceptance_spec.json")
+_SCHEMA_PATH = os.path.join(_HERE, "acceptance.schema.json")
+_DEFAULT_OUT = os.path.join(_HERE, "out")
+
+_STATUS = {"in_repo", "out_of_scope"}
+_KINDS = {"sprint_deliverable", "mvp_metric"}
+
+
+def _eprint(*args) -> None:
+    print(*args, file=sys.stderr)
+
+
+def _load(path: str) -> dict:
+    with open(path, encoding="utf-8") as fh:
+        return json.load(fh)
+
+
+def _write_json(path: str, data) -> None:
+    with open(path, "w", encoding="utf-8") as fh:
+        json.dump(data, fh, ensure_ascii=False, indent=2)
+        fh.write("\n")
+
+
+def _build_bundle() -> dict:
+    """Charge le spec, lit la réalité du CI + le registre 4Big, assemble."""
+    spec = _load(_SPEC_PATH)
+    ci = deps.parse_ci()
+    audit = deps.audit_sprints()
+    return builder.build_matrix(spec, ci, audit, deps)
+
+
+def validate_bundle(bundle: dict, spec: dict, ci: dict, audit: dict) -> list[str]:
+    """Valide la matrice contre acceptance.schema.json + invariants de recette.
+
+    Fonction pure sur (bundle, spec, ci, audit) → les tests injectent des
+    dictionnaires synthétiques pour prouver le basculement en erreur.
+    """
+    schema = _load(_SCHEMA_PATH)
+    errors = list(deps.validate(bundle, schema))
+
+    matrix = bundle["matrix"]
+    manifest = bundle["manifest"]
+    cov = manifest["coverage"]
+    self_module = spec["self_module"]
+
+    deliverables = builder.all_deliverables(spec)
+
+    # INV1 · ids uniques ; livrables de sprint = S1..S8 ordonnés ; métriques =
+    # M1..M7 ordonnées (aucune promesse roadmap omise ni dupliquée).
+    ids = [r["id"] for r in matrix]
+    if len(set(ids)) != len(ids):
+        errors.append(f"lignes: ids en double {ids}.")
+    sprint_ids = [d["id"] for d in spec.get("sprint_deliverables", [])]
+    metric_ids = [d["id"] for d in spec.get("mvp_metrics", [])]
+    if sprint_ids != sorted(sprint_ids):
+        errors.append(f"livrables de sprint non ordonnés {sprint_ids}.")
+    if metric_ids != sorted(metric_ids):
+        errors.append(f"métriques MVP non ordonnées {metric_ids}.")
+
+    # INV2 · toute preuve module citée est gated (aucune preuve fantôme). On
+    # RECALCULE l'ensemble gated depuis le CI (on ne fait pas confiance au
+    # manifeste).
+    gated = set(builder.gated_modules(ci, self_module))
+    for r in matrix:
+        for m in r["modules"]:
+            if m["module"] not in gated:
+                errors.append(f"ligne {r['id']}: preuve non gated {m['module']!r}.")
+            if m["gated"] != (m["module"] in gated):
+                errors.append(f"ligne {r['id']}: drapeau `gated` faux pour {m['module']!r}.")
+
+    # INV3 · COUVERTURE BIJECTIVE (cœur anti-invention · #6) : les modules cités
+    # (lignes in_repo) == l'ensemble gated (hors self). Aucun oubli, aucune
+    # invention.
+    cited: set[str] = set()
+    for d in deliverables:
+        if d.get("status") == "in_repo":
+            cited.update(d.get("evidence_modules", []))
+    missing = sorted(gated - cited)
+    phantom = sorted(cited - gated)
+    if missing:
+        errors.append(f"couverture: modules gated non tracés {missing}.")
+    if phantom:
+        errors.append(f"couverture: preuves non gated {phantom}.")
+    if not cov["bijective"] and not (missing or phantom):
+        errors.append("couverture: `bijective=false` mais aucun écart recalculé.")
+    if cov["bijective"] and (missing or phantom):
+        errors.append("couverture: `bijective=true` mais écart recalculé présent.")
+
+    # INV4 · la fenêtre de sprint de chaque module gated est connue (registre
+    # 4Big + extra du spec) — aucun module gated sans sprint.
+    ms_map = builder.module_sprint_map(spec, audit)
+    unknown = sorted(m for m in gated if m not in ms_map)
+    if unknown:
+        errors.append(f"sprint: modules gated sans fenêtre {unknown}.")
+    for path, meta in spec.get("extra_module_sprint", {}).items():
+        if path in audit and audit[path] != meta["sprint"]:
+            errors.append(
+                f"sprint: extra {path!r} en conflit avec le registre 4Big "
+                f"({meta['sprint']} vs {audit[path]})."
+            )
+
+    # INV5 · PARTITION EXACTE par sprint : chaque livrable de sprint SX cite
+    # EXACTEMENT les modules gated de fenêtre SX (ni trou ni chevauchement).
+    gated_by_sprint: dict[str, set] = {}
+    for m in gated:
+        if m in ms_map:
+            gated_by_sprint.setdefault(ms_map[m], set()).add(m)
+    for d in spec.get("sprint_deliverables", []):
+        sx = d["sprint"]
+        want = gated_by_sprint.get(sx, set())
+        have = set(d.get("evidence_modules", []))
+        if want != have:
+            errors.append(
+                f"partition {sx}: attendu {sorted(want)} ≠ cité {sorted(have)}."
+            )
+    if cov["partition_ok"] != all(p["exact"] for p in cov["sprint_partition"]) or \
+            (cov["unknown_sprint"] and cov["partition_ok"]):
+        errors.append("partition: drapeau `partition_ok` incohérent.")
+
+    # INV6 · tout hors-périmètre porte une source non vide (aucune omission
+    # silencieuse d'une promesse roadmap).
+    for d in deliverables:
+        for o in d.get("out_of_scope", []):
+            if not o.get("source", "").strip():
+                errors.append(f"ligne {d['id']}: hors-périmètre sans source: {o.get('item')!r}.")
+
+    # INV7 · cohérence statut/preuve : in_repo ⇒ ≥1 module OU ≥1 artefact ;
+    # out_of_scope ⇒ 0 module + ≥1 hors-périmètre sourcé. Tout artefact cité
+    # existe sur disque (fait · #6).
+    for d in deliverables:
+        st = d.get("status")
+        if st not in _STATUS:
+            errors.append(f"ligne {d['id']}: statut invalide {st!r}.")
+        mods = d.get("evidence_modules", [])
+        arts = d.get("evidence_artifacts", [])
+        oos = d.get("out_of_scope", [])
+        if st == "in_repo" and not mods and not arts:
+            errors.append(f"ligne {d['id']}: in_repo sans aucune preuve.")
+        if st == "out_of_scope":
+            if mods:
+                errors.append(f"ligne {d['id']}: out_of_scope avec preuve module {mods}.")
+            if not oos:
+                errors.append(f"ligne {d['id']}: out_of_scope sans hors-périmètre sourcé.")
+        for a in arts:
+            if not deps.artifact_exists(a):
+                errors.append(f"ligne {d['id']}: artefact-preuve inexistant {a!r}.")
+
+    # INV8 · kind cohérent : sprint_deliverable ⇒ sprint renseigné ; mvp_metric
+    # ⇒ sprint null (les métriques transcendent les sprints).
+    for d in deliverables:
+        if d["kind"] not in _KINDS:
+            errors.append(f"ligne {d['id']}: kind invalide {d['kind']!r}.")
+        if d["kind"] == "sprint_deliverable" and not d.get("sprint"):
+            errors.append(f"ligne {d['id']}: sprint_deliverable sans sprint.")
+        if d["kind"] == "mvp_metric" and d.get("sprint") is not None:
+            errors.append(f"ligne {d['id']}: mvp_metric avec sprint non nul {d.get('sprint')!r}.")
+
+    # INV9 · le manifeste s'exclut lui-même (SoD · ISA 315) — la matrice ne se
+    # cite jamais comme preuve.
+    if self_module in cited:
+        errors.append("SoD: la matrice se cite elle-même comme preuve.")
+    if cov["self_module_excluded"] != self_module:
+        errors.append("SoD: self_module_excluded divergent.")
+
+    # INV10 · roadmap_line = entier positif ; pas de doublon de ligne.
+    lines = [d["roadmap_line"] for d in deliverables]
+    if any(not isinstance(x, int) or x < 1 for x in lines):
+        errors.append(f"roadmap_line: valeur non positive {lines}.")
+
+    return errors
+
+
+def _validate_or_die(bundle: dict) -> None:
+    spec = _load(_SPEC_PATH)
+    ci = deps.parse_ci()
+    audit = deps.audit_sprints()
+    errs = validate_bundle(bundle, spec, ci, audit)
+    if errs:
+        _eprint("❌ Validation échouée :")
+        for e in errs:
+            _eprint(f"   · {e}")
+        sys.exit(1)
+
+
+def cmd_build(args) -> None:
+    bundle = _build_bundle()
+    _validate_or_die(bundle)
+    out_dir = args.out or _DEFAULT_OUT
+    os.makedirs(out_dir, exist_ok=True)
+    _write_json(os.path.join(out_dir, "acceptance_matrix.json"),
+                {"matrix": bundle["matrix"], "verdict": bundle["verdict"]})
+    _write_json(os.path.join(out_dir, "MANIFEST.json"), bundle["manifest"])
+    cov = bundle["manifest"]["coverage"]
+    c = bundle["manifest"]["counts"]
+    verdict = "✅ RECETTE PASS" if bundle["verdict"] else "❌ RECETTE FAIL"
+    print(f"{verdict} — {c['sprint_deliverables']} livrables de sprint + "
+          f"{c['mvp_metrics']} métriques MVP · {c['gated_modules']} modules gated "
+          f"tracés (bijectif={cov['bijective']}) · {c['out_of_scope_items']} "
+          f"hors-périmètre sourcés.")
+    print(f"   → {os.path.join(out_dir, 'acceptance_matrix.json')}")
+    print(f"   → {os.path.join(out_dir, 'MANIFEST.json')}")
+
+
+def cmd_validate(args) -> None:
+    bundle = _build_bundle()
+    _validate_or_die(bundle)
+    cov = bundle["manifest"]["coverage"]
+    print(f"✅ Matrice valide — bijective={cov['bijective']} · "
+          f"partition_ok={cov['partition_ok']} · verdict={bundle['verdict']}.")
+
+
+def main(argv=None) -> None:
+    p = argparse.ArgumentParser(
+        prog="acceptance_gen",
+        description="Matrice d'acceptation / traçabilité MVP (roadmap → preuve).")
+    sub = p.add_subparsers(dest="cmd", required=True)
+
+    b = sub.add_parser("build", help="écrit acceptance_matrix.json / MANIFEST.json")
+    b.add_argument("-o", "--out", help="répertoire de sortie (défaut: ./out)")
+    b.set_defaults(func=cmd_build)
+
+    v = sub.add_parser("validate", help="valide sans écrire")
+    v.add_argument("-o", "--out", help="(ignoré · symétrie CLI)")
+    v.set_defaults(func=cmd_validate)
+
+    args = p.parse_args(argv)
+    args.func(args)
+
+
+if __name__ == "__main__":
+    main()

05_deliverables_mvp/qa/acceptance/acceptance_spec.json

@@ -0,0 +1,262 @@
+{
+  "matrix": "MVP Acceptance / Traçabilité roadmap → livrable → preuve",
+  "version": "1.0",
+  "reference_cadre": "Roadmap 04_roadmap/ROADMAP_8_WEEKS_OR_LESS.md · livrables de sprint (L33..L76) + métriques succès MVP (L81..L87) · CLAUDE.md #8 (VPS pour TOUS projets · le worker n'écrit jamais sur le VPS) · Sprint 8 « Bufer pour corrections finales » (L75)",
+  "roadmap_ref": "04_roadmap/ROADMAP_8_WEEKS_OR_LESS.md",
+  "self_module": "qa/acceptance",
+  "cible_portage": "Michel / comité de recette — document de recette (hand-off démo publique)",
+  "extra_module_sprint": {
+    "qa/audit_4big": {
+      "sprint": "S7",
+      "source": "roadmap Sprint 7 · QA « Audit 4Big niveau 95+/100 sur 100% deliverables » (L69) — l'auditeur ne figure pas dans son propre registre (SoD · ISA 315), sa fenêtre de sprint est déclarée ici avec source"
+    }
+  },
+  "sprint_deliverables": [
+    {
+      "id": "S1",
+      "kind": "sprint_deliverable",
+      "sprint": "S1",
+      "roadmap_line": 33,
+      "statement": "Repo Gitea live · gap analysis complète · baseline tests OK",
+      "status": "in_repo",
+      "evidence_modules": [],
+      "evidence_artifacts": ["GAP_ANALYSIS_SPRINT1.md"],
+      "out_of_scope": [
+        {
+          "item": "baseline Playwright /crm · /waf-home · /qa (dépend du site LIVE) et push initial VPS",
+          "source": "CLAUDE.md #8 (VPS pour TOUS projets · worker n'écrit jamais sur le VPS)"
+        }
+      ]
+    },
+    {
+      "id": "S2",
+      "kind": "sprint_deliverable",
+      "sprint": "S2",
+      "roadmap_line": 39,
+      "statement": "7 dashboards uniformes · CRM avec pipeline live · RBAC 50 rôles",
+      "status": "in_repo",
+      "evidence_modules": [
+        "publiciste",
+        "rbac",
+        "rbac/fixtures_gen",
+        "rbac/userperm_gen",
+        "rbac/roleprofile_gen",
+        "rbac/apply_plan",
+        "faisabilite/generator"
+      ],
+      "evidence_artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "import réel des fixtures RBAC + rendu HTML des dashboards sur le desk ERPNext LIVE",
+          "source": "CLAUDE.md #8 (VPS) · #1 (ERPNext natif = priorité — import via bench par agent Backend)"
+        }
+      ]
+    },
+    {
+      "id": "S3",
+      "kind": "sprint_deliverable",
+      "sprint": "S3",
+      "roadmap_line": 46,
+      "statement": "P07 Aqua Terra faisabilité complète auto-générée en <1h",
+      "status": "in_repo",
+      "evidence_modules": ["faisabilite/bancable"],
+      "evidence_artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "pipeline Blender→IFC→Speckle→Flux (aec.py) exécuté sur runtime BIM VPS + viewer 3D console LIVE",
+          "source": "CLAUDE.md #8 (VPS) — le rendu 3D/BIM tourne hors repo (bim-cloud.otov7.com)"
+        }
+      ]
+    },
+    {
+      "id": "S4",
+      "kind": "sprint_deliverable",
+      "sprint": "S4",
+      "roadmap_line": 52,
+      "statement": "5 portails RBAC live · workflow vente end-to-end",
+      "status": "in_repo",
+      "evidence_modules": [
+        "crm/workflow_vente",
+        "crm/dossier_vente",
+        "crm/commissions",
+        "fiscal/ecf_dgii",
+        "frontend/portails"
+      ],
+      "evidence_artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "création module + import DocTypes/Workflow + transmission e-CF réelle DGII (Compupar) sur ERPNext LIVE",
+          "source": "CLAUDE.md #8 (VPS) · #10 (Cardnet/DGII endpoints prod)"
+        }
+      ]
+    },
+    {
+      "id": "S5",
+      "kind": "sprint_deliverable",
+      "sprint": "S5",
+      "roadmap_line": 58,
+      "statement": "Dépôts P05/P07 ONAPI · Apps mobile live 2 stores",
+      "status": "in_repo",
+      "evidence_modules": ["legal/confotur", "qa/audit_5d"],
+      "evidence_artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "dépôt ONAPI/CONFOTUR réel + builds Expo 54 et submit App Store / Play Store",
+          "source": "roadmap Sprint 5 · Mobile (L57) — builds & submissions stores hors repo · CLAUDE.md #8"
+        }
+      ]
+    },
+    {
+      "id": "S6",
+      "kind": "sprint_deliverable",
+      "sprint": "S6",
+      "roadmap_line": 64,
+      "statement": "Indexation Google trilingue · OTOIA voice orchestre AEC",
+      "status": "in_repo",
+      "evidence_modules": ["seo", "frontend/chat_otoia"],
+      "evidence_artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "indexation Google réelle (crawl runtime) + voix Amélie QC (multilingual_v2) exécutée sur runtime OTOIA VPS",
+          "source": "CLAUDE.md #8 (VPS) — indexation et TTS sont des services runtime hors repo"
+        }
+      ]
+    },
+    {
+      "id": "S7",
+      "kind": "sprint_deliverable",
+      "sprint": "S7",
+      "roadmap_line": 70,
+      "statement": "Plateforme prête pour démo publique (polish · scénarios démo · audit 4Big 95+)",
+      "status": "in_repo",
+      "evidence_modules": ["demo/scenarios", "qa/audit_4big"],
+      "evidence_artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "polish final du site otov7.com (Hero · nav) sur le site LIVE",
+          "source": "roadmap Sprint 7 · Frontend « Site otov7.com polish final » (L67) — dépend du site LIVE · CLAUDE.md #8"
+        }
+      ]
+    },
+    {
+      "id": "S8",
+      "kind": "sprint_deliverable",
+      "sprint": "S8",
+      "roadmap_line": 76,
+      "statement": "Plateforme démo live · 5 portails · CRM · BIM auto · OTOIA voice",
+      "status": "in_repo",
+      "evidence_modules": ["qa/regression", "devops/deploy_runbook"],
+      "evidence_artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "déploiement production complet + monitoring exécutés sur le VPS (run-book appliqué)",
+          "source": "CLAUDE.md #8 (VPS) — l'exécution du run-book de déploiement revient à l'agent DevOps / ERPNext Backend"
+        }
+      ]
+    }
+  ],
+  "mvp_metrics": [
+    {
+      "id": "M1",
+      "kind": "mvp_metric",
+      "sprint": null,
+      "roadmap_line": 81,
+      "statement": "Faisabilité auto générée en <1h (vs 3-5 jours actuels)",
+      "status": "in_repo",
+      "evidence_modules": ["publiciste", "faisabilite/generator", "faisabilite/bancable"],
+      "evidence_artifacts": [],
+      "out_of_scope": []
+    },
+    {
+      "id": "M2",
+      "kind": "mvp_metric",
+      "sprint": null,
+      "roadmap_line": 82,
+      "statement": "Score 4Big >=95/100 sur 100% deliverables",
+      "status": "in_repo",
+      "evidence_modules": ["qa/audit_4big"],
+      "evidence_artifacts": [],
+      "out_of_scope": []
+    },
+    {
+      "id": "M3",
+      "kind": "mvp_metric",
+      "sprint": null,
+      "roadmap_line": 83,
+      "statement": "7 dashboards entités uniformes luxury",
+      "status": "in_repo",
+      "evidence_modules": ["frontend/portails", "rbac/roleprofile_gen"],
+      "evidence_artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "rendu luxury dark+doré sur le desk ERPNext LIVE (Custom HTML)",
+          "source": "CLAUDE.md #8 (VPS) · #4 (design luxury rendu côté desk)"
+        }
+      ]
+    },
+    {
+      "id": "M4",
+      "kind": "mvp_metric",
+      "sprint": null,
+      "roadmap_line": 84,
+      "statement": "CRM ERPNext natif fonctionnel end-to-end",
+      "status": "in_repo",
+      "evidence_modules": ["crm/workflow_vente", "crm/dossier_vente", "crm/commissions"],
+      "evidence_artifacts": [],
+      "out_of_scope": []
+    },
+    {
+      "id": "M5",
+      "kind": "mvp_metric",
+      "sprint": null,
+      "roadmap_line": 85,
+      "statement": "Mobile app publié 2 stores",
+      "status": "out_of_scope",
+      "evidence_modules": [],
+      "evidence_artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "builds Expo 54 + submit App Store #32 / Play Store — aucune fabrication en repo",
+          "source": "roadmap Sprint 5 · Mobile (L56-57) — builds & submissions hors repo · CLAUDE.md #8"
+        }
+      ]
+    },
+    {
+      "id": "M6",
+      "kind": "mvp_metric",
+      "sprint": null,
+      "roadmap_line": 86,
+      "statement": "SEO trilingue indexé",
+      "status": "in_repo",
+      "evidence_modules": ["seo"],
+      "evidence_artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "indexation Google réelle (soumission sitemaps + crawl)",
+          "source": "CLAUDE.md #8 (VPS) — worker systemd SEO tourne hors repo"
+        }
+      ]
+    },
+    {
+      "id": "M7",
+      "kind": "mvp_metric",
+      "sprint": null,
+      "roadmap_line": 87,
+      "statement": "OTOIA voice orchestre bout-en-bout",
+      "status": "in_repo",
+      "evidence_modules": ["frontend/chat_otoia"],
+      "evidence_artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "voix Amélie QC (multilingual_v2) + orchestration AEC exécutées sur runtime OTOIA VPS",
+          "source": "CLAUDE.md #8 (VPS) — TTS et pilotage aec.py sont des services runtime hors repo"
+        }
+      ]
+    }
+  ],
+  "notes": [
+    "Anti-invention (#6) : aucun chiffre metier n'est fabrique. Les nombres presents dans les enonces (<1h, 95/100, 7 dashboards, 2 stores) sont des CITATIONS verbatim de la roadmap, pas des mesures inventees ; la matrice ne PRETEND aucune mesure de performance (pas de « genere en 42 min »).",
+    "Perimetre PROUVE : l'ensemble des modules-preuve est confronte de facon BIJECTIVE a la realite du CI (parse_ci) — un module gated non cite OU un module cite non gated => generation refusee.",
+    "Cross-coherence : la fenetre de sprint de chaque module est DERIVEE du registre de l'auditeur 4Big (qa/audit_4big/quality_spec.json · zero duplication · #5), jamais re-declaree ici (sauf l'auditeur lui-meme, avec source)."
+  ]
+}

05_deliverables_mvp/qa/acceptance/acclib/__init__.py

@@ -0,0 +1,23 @@
+"""acclib · briques de la Matrice d'acceptation / traçabilité MVP (Sprint 8 · QA).
+
+Document de MÉTA-NIVEAU orienté RECETTE : il mappe chaque PROMESSE de la roadmap
+(8 livrables de sprint L33..L76 + 7 métriques succès MVP L81..L87) vers sa PREUVE
+livrée (un module gated du CI) OU un hors-périmètre SOURCÉ (VPS #8 · builds
+stores · services runtime). Axe distinct des autres harnais méta :
+
+  - qa/audit_4big        → note la QUALITÉ statique de chaque module ;
+  - qa/regression        → prouve que chaque suite S'EXÉCUTE au vert ;
+  - devops/deploy_runbook→ ORDONNE le déploiement VPS des modules ;
+  - qa/acceptance (ici)  → prouve que chaque PROMESSE roadmap est HONORÉE
+                           (couverte par un livrable) ou explicitement hors
+                           périmètre worker — la RECETTE du mandat.
+
+  deps     → réutilise q4lib.registry.parse_ci + le validateur maison Publiciste
+             + le registre de sprints de l'auditeur 4Big (zéro duplication · #5)
+  builder  → assemble la matrice déterministe + le manifeste (preuve de couverture
+             bijective vs CI + partition par sprint + hors-périmètre sourcé)
+
+Anti-invention (#6) : les nombres des énoncés sont des citations verbatim de la
+roadmap ; aucune mesure de performance n'est fabriquée ; tout module-preuve est
+confronté au CI ; tout hors-périmètre porte une source.
+"""

05_deliverables_mvp/qa/acceptance/acclib/builder.py

@@ -0,0 +1,189 @@
+"""Assemblage de la matrice d'acceptation à partir du spec + de la réalité CI.
+
+La matrice mappe chaque promesse roadmap → preuve gated OU hors-périmètre sourcé.
+La liste des modules-preuve n'est PAS crue sur parole : elle est confrontée à
+`.gitea/workflows/ci.yml` (via q4lib.registry.parse_ci, réutilisé) de façon
+BIJECTIVE, et la fenêtre de sprint de chaque module est LUE dans le registre de
+l'auditeur 4Big (zéro re-déclaration · anti-dérive).
+
+Fonctions PURES (aucune I/O, aucun horodatage) → sortie déterministe, diffable,
+re-générable bit-à-bit. Le CLI injecte le spec chargé, `parse_ci()` et
+`audit_sprints()` ; les tests injectent des dictionnaires synthétiques (facile à
+mettre en défaut → tests négatifs).
+"""
+
+from __future__ import annotations
+
+from typing import Any
+
+
+def gated_modules(ci: dict, self_module: str) -> list[str]:
+    """Chemins de modules gated par le CI, hors la matrice elle-même (SoD)."""
+    paths = set(ci.get("job_to_path", {}).values())
+    paths.discard(self_module)
+    return sorted(paths)
+
+
+def _path_to_job(ci: dict) -> dict[str, str]:
+    """Inverse job→path en path→job (job minimal en cas de collision)."""
+    out: dict[str, str] = {}
+    for job, path in ci.get("job_to_path", {}).items():
+        if path not in out or job < out[path]:
+            out[path] = job
+    return out
+
+
+def module_sprint_map(spec: dict, audit_sprints: dict) -> dict[str, str]:
+    """{module -> sprint}, dérivé du registre 4Big + complété par le spec.
+
+    Le registre de l'auditeur est la source ; l'auditeur lui-même n'y figure pas
+    (SoD) → son sprint vient de `extra_module_sprint` (avec source). Une entrée
+    `extra` qui écraserait un module déjà daté par le registre est un conflit :
+    on la garde ici mais l'incohérence est signalée par l'invariant du CLI."""
+    out = dict(audit_sprints)
+    for path, meta in spec.get("extra_module_sprint", {}).items():
+        out.setdefault(path, meta["sprint"])
+    return out
+
+
+def all_deliverables(spec: dict) -> list[dict]:
+    """Livrables de sprint + métriques MVP, dans l'ordre du spec."""
+    return list(spec.get("sprint_deliverables", [])) + list(spec.get("mvp_metrics", []))
+
+
+def _resolve_row(d: dict, gated_set: set, p2j: dict, deps_mod) -> dict[str, Any]:
+    """Une ligne de matrice : preuves résolues + drapeaux de cohérence."""
+    ev_modules = list(d.get("evidence_modules", []))
+    modules = [
+        {"module": m, "ci_job": p2j.get(m), "gated": m in gated_set}
+        for m in ev_modules
+    ]
+    artifacts = [
+        {"path": a, "exists": deps_mod.artifact_exists(a)}
+        for a in d.get("evidence_artifacts", [])
+    ]
+    oos = [
+        {"item": o["item"], "source": o.get("source")}
+        for o in d.get("out_of_scope", [])
+    ]
+    has_evidence = bool(ev_modules) or bool(d.get("evidence_artifacts"))
+    return {
+        "id": d["id"],
+        "kind": d["kind"],
+        "sprint": d.get("sprint"),
+        "roadmap_line": d["roadmap_line"],
+        "statement": d["statement"],
+        "status": d["status"],
+        "modules": modules,
+        "artifacts": artifacts,
+        "out_of_scope": oos,
+        "n_modules": len(ev_modules),
+        "all_modules_gated": all(m in gated_set for m in ev_modules),
+        "all_artifacts_exist": all(a["exists"] for a in artifacts),
+        "has_evidence": has_evidence,
+    }
+
+
+def build_matrix(spec: dict, ci: dict, audit_sprints: dict, deps_mod) -> dict[str, Any]:
+    """Consolide spec + CI + registre 4Big en matrice + manifeste agrégé.
+
+    Ne lève JAMAIS : les incohérences (module manquant / en trop / partition de
+    sprint fausse / artefact absent) sont REMONTÉES dans `manifest.coverage`
+    pour que le CLI/les tests les signalent explicitement. Une recette ne doit
+    pas oublier silencieusement une promesse."""
+    self_module = spec["self_module"]
+    gated = gated_modules(ci, self_module)
+    gated_set = set(gated)
+    p2j = _path_to_job(ci)
+    ms_map = module_sprint_map(spec, audit_sprints)
+
+    deliverables = all_deliverables(spec)
+    rows = [_resolve_row(d, gated_set, p2j, deps_mod) for d in deliverables]
+
+    # --- Couverture BIJECTIVE (cœur anti-invention · #6) ------------------
+    # cited = tout module-preuve des lignes IN_REPO ; doit égaler l'ensemble
+    # gated (hors self). Un module gated non cité = promesse orpheline ; un
+    # module cité non gated = preuve fantôme.
+    cited: set[str] = set()
+    for d in deliverables:
+        if d.get("status") == "in_repo":
+            cited.update(d.get("evidence_modules", []))
+    missing = sorted(gated_set - cited)   # gated mais non tracé
+    phantom = sorted(cited - gated_set)   # tracé mais non gated
+    bijective = not missing and not phantom
+
+    # --- Partition par sprint --------------------------------------------
+    # Chaque livrable de sprint SX doit citer EXACTEMENT les modules gated dont
+    # la fenêtre est SX (ni plus, ni moins). Prouve que la matrice répartit les
+    # livraisons par sprint sans trou ni chevauchement.
+    gated_by_sprint: dict[str, set] = {}
+    unknown_sprint = sorted(m for m in gated if m not in ms_map)
+    for m in gated:
+        if m in ms_map:
+            gated_by_sprint.setdefault(ms_map[m], set()).add(m)
+
+    sprint_partition: list[dict] = []
+    for d in spec.get("sprint_deliverables", []):
+        sx = d["sprint"]
+        want = gated_by_sprint.get(sx, set())
+        have = set(d.get("evidence_modules", []))
+        sprint_partition.append({
+            "sprint": sx,
+            "expected": sorted(want),
+            "cited": sorted(have),
+            "missing": sorted(want - have),
+            "extra": sorted(have - want),
+            "exact": want == have,
+        })
+    partition_ok = all(p["exact"] for p in sprint_partition) and not unknown_sprint
+
+    # --- Artefacts hors-CI (docs) ----------------------------------------
+    missing_artifacts = sorted(
+        a["path"]
+        for r in rows for a in r["artifacts"] if not a["exists"]
+    )
+
+    # --- Complétude roadmap ----------------------------------------------
+    sprint_ids = [d["id"] for d in spec.get("sprint_deliverables", [])]
+    metric_ids = [d["id"] for d in spec.get("mvp_metrics", [])]
+    oos_count = sum(len(r["out_of_scope"]) for r in rows)
+    in_repo = sum(1 for r in rows if r["status"] == "in_repo")
+    out_of_scope_full = sum(1 for r in rows if r["status"] == "out_of_scope")
+
+    manifest = {
+        "generated_from": f"acceptance_spec.json v{spec.get('version')}",
+        "reference_cadre": spec.get("reference_cadre"),
+        "roadmap_ref": spec.get("roadmap_ref"),
+        "cible_portage": spec.get("cible_portage"),
+        "self_module": self_module,
+        "counts": {
+            "sprint_deliverables": len(sprint_ids),
+            "mvp_metrics": len(metric_ids),
+            "deliverables_total": len(deliverables),
+            "in_repo": in_repo,
+            "out_of_scope_full": out_of_scope_full,
+            "gated_modules": len(gated),
+            "cited_modules": len(cited),
+            "out_of_scope_items": oos_count,
+        },
+        "coverage": {
+            "ci_modules_count": len(gated),
+            "cited_modules_count": len(cited),
+            "bijective": bijective,
+            "missing_in_matrix": missing,
+            "phantom_evidence": phantom,
+            "partition_ok": partition_ok,
+            "sprint_partition": sprint_partition,
+            "unknown_sprint": unknown_sprint,
+            "missing_artifacts": missing_artifacts,
+            "self_module_excluded": self_module,
+        },
+        "roadmap": {
+            "sprint_ids": sprint_ids,
+            "metric_ids": metric_ids,
+        },
+    }
+
+    verdict = bijective and partition_ok and not missing_artifacts
+
+    return {"matrix": rows, "manifest": manifest, "verdict": verdict}

05_deliverables_mvp/qa/acceptance/acclib/deps.py

@@ -0,0 +1,72 @@
+"""Réutilisation des briques déjà livrées (workflow #5 · zéro duplication).
+
+La matrice d'acceptation ne redéfinit RIEN qui existe ailleurs :
+
+  - `parse_ci`        : parseur de `.gitea/workflows/ci.yml` de l'auditeur 4Big
+                        (`qa/audit_4big/q4lib/registry.py`) — la SEULE autorité
+                        sur les jobs de test réellement gated. On le réutilise
+                        pour DÉRIVER l'ensemble des modules-preuve (jamais une
+                        liste à la main) et prouver la couverture bijective.
+  - `validate`        : validateur JSON-Schema maison du Publiciste (draft-07,
+                        sous-ensemble) — la matrice est validée SANS pip, car le
+                        gate CI Gitea Actions tourne sans réseau (CLAUDE.md #2).
+  - `audit_sprints`   : la fenêtre de sprint de chaque module est LUE dans le
+                        registre de l'auditeur 4Big (`quality_spec.json`) — une
+                        seule source de vérité pour tout le mandat (anti-dérive).
+
+Import par manipulation de `sys.path` (idiome des modules voisins :
+`q4lib/deps.py`, `reglib/deps.py`) plutôt que duplication.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import sys
+
+_HERE = os.path.dirname(os.path.abspath(__file__))
+# qa/acceptance/acclib → 05_deliverables_mvp
+_DELIVERABLES = os.path.normpath(os.path.join(_HERE, "..", "..", ".."))
+_PUB = os.path.join(_DELIVERABLES, "publiciste")
+_AUDIT = os.path.join(_DELIVERABLES, "qa", "audit_4big")
+
+for _p in (_PUB, _AUDIT):
+    if _p not in sys.path:
+        sys.path.insert(0, _p)
+
+from lib import validator          # type: ignore  # noqa: E402
+from q4lib import registry         # type: ignore  # noqa: E402
+
+validate = validator.validate
+parse_ci = registry.parse_ci
+
+# Racine des livrables, exposée aux autres modules acclib.
+DELIVERABLES_ROOT = _DELIVERABLES
+
+_AUDIT_SPEC = os.path.join(_AUDIT, "quality_spec.json")
+
+
+def audit_sprints() -> dict:
+    """{chemin module -> sprint} tel que déclaré par le registre de l'auditeur
+    4Big. Une seule source de vérité (anti-dérive · #5) ; l'auditeur ne se
+    référence pas lui-même — son sprint est complété par le spec (extra)."""
+    with open(_AUDIT_SPEC, encoding="utf-8") as fh:
+        spec = json.load(fh)
+    return {m["path"]: m["sprint"] for m in spec["modules"]}
+
+
+def artifact_exists(rel_path: str) -> bool:
+    """Vrai ssi l'artefact cité (chemin relatif à 05_deliverables_mvp) existe.
+
+    Fait de disque pur — un artefact-preuve inexistant est un mensonge de
+    couverture (#6)."""
+    return os.path.exists(os.path.join(DELIVERABLES_ROOT, rel_path))
+
+
+__all__ = [
+    "validate",
+    "parse_ci",
+    "audit_sprints",
+    "artifact_exists",
+    "DELIVERABLES_ROOT",
+]

05_deliverables_mvp/qa/acceptance/out/MANIFEST.json

@@ -0,0 +1,171 @@
+{
+  "generated_from": "acceptance_spec.json v1.0",
+  "reference_cadre": "Roadmap 04_roadmap/ROADMAP_8_WEEKS_OR_LESS.md · livrables de sprint (L33..L76) + métriques succès MVP (L81..L87) · CLAUDE.md #8 (VPS pour TOUS projets · le worker n'écrit jamais sur le VPS) · Sprint 8 « Bufer pour corrections finales » (L75)",
+  "roadmap_ref": "04_roadmap/ROADMAP_8_WEEKS_OR_LESS.md",
+  "cible_portage": "Michel / comité de recette — document de recette (hand-off démo publique)",
+  "self_module": "qa/acceptance",
+  "counts": {
+    "sprint_deliverables": 8,
+    "mvp_metrics": 7,
+    "deliverables_total": 15,
+    "in_repo": 14,
+    "out_of_scope_full": 1,
+    "gated_modules": 21,
+    "cited_modules": 21,
+    "out_of_scope_items": 12
+  },
+  "coverage": {
+    "ci_modules_count": 21,
+    "cited_modules_count": 21,
+    "bijective": true,
+    "missing_in_matrix": [],
+    "phantom_evidence": [],
+    "partition_ok": true,
+    "sprint_partition": [
+      {
+        "sprint": "S1",
+        "expected": [],
+        "cited": [],
+        "missing": [],
+        "extra": [],
+        "exact": true
+      },
+      {
+        "sprint": "S2",
+        "expected": [
+          "faisabilite/generator",
+          "publiciste",
+          "rbac",
+          "rbac/apply_plan",
+          "rbac/fixtures_gen",
+          "rbac/roleprofile_gen",
+          "rbac/userperm_gen"
+        ],
+        "cited": [
+          "faisabilite/generator",
+          "publiciste",
+          "rbac",
+          "rbac/apply_plan",
+          "rbac/fixtures_gen",
+          "rbac/roleprofile_gen",
+          "rbac/userperm_gen"
+        ],
+        "missing": [],
+        "extra": [],
+        "exact": true
+      },
+      {
+        "sprint": "S3",
+        "expected": [
+          "faisabilite/bancable"
+        ],
+        "cited": [
+          "faisabilite/bancable"
+        ],
+        "missing": [],
+        "extra": [],
+        "exact": true
+      },
+      {
+        "sprint": "S4",
+        "expected": [
+          "crm/commissions",
+          "crm/dossier_vente",
+          "crm/workflow_vente",
+          "fiscal/ecf_dgii",
+          "frontend/portails"
+        ],
+        "cited": [
+          "crm/commissions",
+          "crm/dossier_vente",
+          "crm/workflow_vente",
+          "fiscal/ecf_dgii",
+          "frontend/portails"
+        ],
+        "missing": [],
+        "extra": [],
+        "exact": true
+      },
+      {
+        "sprint": "S5",
+        "expected": [
+          "legal/confotur",
+          "qa/audit_5d"
+        ],
+        "cited": [
+          "legal/confotur",
+          "qa/audit_5d"
+        ],
+        "missing": [],
+        "extra": [],
+        "exact": true
+      },
+      {
+        "sprint": "S6",
+        "expected": [
+          "frontend/chat_otoia",
+          "seo"
+        ],
+        "cited": [
+          "frontend/chat_otoia",
+          "seo"
+        ],
+        "missing": [],
+        "extra": [],
+        "exact": true
+      },
+      {
+        "sprint": "S7",
+        "expected": [
+          "demo/scenarios",
+          "qa/audit_4big"
+        ],
+        "cited": [
+          "demo/scenarios",
+          "qa/audit_4big"
+        ],
+        "missing": [],
+        "extra": [],
+        "exact": true
+      },
+      {
+        "sprint": "S8",
+        "expected": [
+          "devops/deploy_runbook",
+          "qa/regression"
+        ],
+        "cited": [
+          "devops/deploy_runbook",
+          "qa/regression"
+        ],
+        "missing": [],
+        "extra": [],
+        "exact": true
+      }
+    ],
+    "unknown_sprint": [],
+    "missing_artifacts": [],
+    "self_module_excluded": "qa/acceptance"
+  },
+  "roadmap": {
+    "sprint_ids": [
+      "S1",
+      "S2",
+      "S3",
+      "S4",
+      "S5",
+      "S6",
+      "S7",
+      "S8"
+    ],
+    "metric_ids": [
+      "M1",
+      "M2",
+      "M3",
+      "M4",
+      "M5",
+      "M6",
+      "M7"
+    ]
+  }
+}

05_deliverables_mvp/qa/acceptance/out/acceptance_matrix.json

@@ -0,0 +1,468 @@
+{
+  "matrix": [
+    {
+      "id": "S1",
+      "kind": "sprint_deliverable",
+      "sprint": "S1",
+      "roadmap_line": 33,
+      "statement": "Repo Gitea live · gap analysis complète · baseline tests OK",
+      "status": "in_repo",
+      "modules": [],
+      "artifacts": [
+        {
+          "path": "GAP_ANALYSIS_SPRINT1.md",
+          "exists": true
+        }
+      ],
+      "out_of_scope": [
+        {
+          "item": "baseline Playwright /crm · /waf-home · /qa (dépend du site LIVE) et push initial VPS",
+          "source": "CLAUDE.md #8 (VPS pour TOUS projets · worker n'écrit jamais sur le VPS)"
+        }
+      ],
+      "n_modules": 0,
+      "all_modules_gated": true,
+      "all_artifacts_exist": true,
+      "has_evidence": true
+    },
+    {
+      "id": "S2",
+      "kind": "sprint_deliverable",
+      "sprint": "S2",
+      "roadmap_line": 39,
+      "statement": "7 dashboards uniformes · CRM avec pipeline live · RBAC 50 rôles",
+      "status": "in_repo",
+      "modules": [
+        {
+          "module": "publiciste",
+          "ci_job": "publiciste-tests",
+          "gated": true
+        },
+        {
+          "module": "rbac",
+          "ci_job": "rbac-tests",
+          "gated": true
+        },
+        {
+          "module": "rbac/fixtures_gen",
+          "ci_job": "rbac-fixtures-tests",
+          "gated": true
+        },
+        {
+          "module": "rbac/userperm_gen",
+          "ci_job": "rbac-userperm-tests",
+          "gated": true
+        },
+        {
+          "module": "rbac/roleprofile_gen",
+          "ci_job": "rbac-roleprofile-tests",
+          "gated": true
+        },
+        {
+          "module": "rbac/apply_plan",
+          "ci_job": "rbac-applyplan-tests",
+          "gated": true
+        },
+        {
+          "module": "faisabilite/generator",
+          "ci_job": "faisabilite-gen-tests",
+          "gated": true
+        }
+      ],
+      "artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "import réel des fixtures RBAC + rendu HTML des dashboards sur le desk ERPNext LIVE",
+          "source": "CLAUDE.md #8 (VPS) · #1 (ERPNext natif = priorité — import via bench par agent Backend)"
+        }
+      ],
+      "n_modules": 7,
+      "all_modules_gated": true,
+      "all_artifacts_exist": true,
+      "has_evidence": true
+    },
+    {
+      "id": "S3",
+      "kind": "sprint_deliverable",
+      "sprint": "S3",
+      "roadmap_line": 46,
+      "statement": "P07 Aqua Terra faisabilité complète auto-générée en <1h",
+      "status": "in_repo",
+      "modules": [
+        {
+          "module": "faisabilite/bancable",
+          "ci_job": "bancable-tests",
+          "gated": true
+        }
+      ],
+      "artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "pipeline Blender→IFC→Speckle→Flux (aec.py) exécuté sur runtime BIM VPS + viewer 3D console LIVE",
+          "source": "CLAUDE.md #8 (VPS) — le rendu 3D/BIM tourne hors repo (bim-cloud.otov7.com)"
+        }
+      ],
+      "n_modules": 1,
+      "all_modules_gated": true,
+      "all_artifacts_exist": true,
+      "has_evidence": true
+    },
+    {
+      "id": "S4",
+      "kind": "sprint_deliverable",
+      "sprint": "S4",
+      "roadmap_line": 52,
+      "statement": "5 portails RBAC live · workflow vente end-to-end",
+      "status": "in_repo",
+      "modules": [
+        {
+          "module": "crm/workflow_vente",
+          "ci_job": "crm-workflow-vente-tests",
+          "gated": true
+        },
+        {
+          "module": "crm/dossier_vente",
+          "ci_job": "crm-dossier-vente-tests",
+          "gated": true
+        },
+        {
+          "module": "crm/commissions",
+          "ci_job": "crm-commissions-tests",
+          "gated": true
+        },
+        {
+          "module": "fiscal/ecf_dgii",
+          "ci_job": "fiscal-ecf-tests",
+          "gated": true
+        },
+        {
+          "module": "frontend/portails",
+          "ci_job": "frontend-portails-tests",
+          "gated": true
+        }
+      ],
+      "artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "création module + import DocTypes/Workflow + transmission e-CF réelle DGII (Compupar) sur ERPNext LIVE",
+          "source": "CLAUDE.md #8 (VPS) · #10 (Cardnet/DGII endpoints prod)"
+        }
+      ],
+      "n_modules": 5,
+      "all_modules_gated": true,
+      "all_artifacts_exist": true,
+      "has_evidence": true
+    },
+    {
+      "id": "S5",
+      "kind": "sprint_deliverable",
+      "sprint": "S5",
+      "roadmap_line": 58,
+      "statement": "Dépôts P05/P07 ONAPI · Apps mobile live 2 stores",
+      "status": "in_repo",
+      "modules": [
+        {
+          "module": "legal/confotur",
+          "ci_job": "legal-confotur-tests",
+          "gated": true
+        },
+        {
+          "module": "qa/audit_5d",
+          "ci_job": "qa-audit-5d-tests",
+          "gated": true
+        }
+      ],
+      "artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "dépôt ONAPI/CONFOTUR réel + builds Expo 54 et submit App Store / Play Store",
+          "source": "roadmap Sprint 5 · Mobile (L57) — builds & submissions stores hors repo · CLAUDE.md #8"
+        }
+      ],
+      "n_modules": 2,
+      "all_modules_gated": true,
+      "all_artifacts_exist": true,
+      "has_evidence": true
+    },
+    {
+      "id": "S6",
+      "kind": "sprint_deliverable",
+      "sprint": "S6",
+      "roadmap_line": 64,
+      "statement": "Indexation Google trilingue · OTOIA voice orchestre AEC",
+      "status": "in_repo",
+      "modules": [
+        {
+          "module": "seo",
+          "ci_job": "seo-tests",
+          "gated": true
+        },
+        {
+          "module": "frontend/chat_otoia",
+          "ci_job": "chat-otoia-tests",
+          "gated": true
+        }
+      ],
+      "artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "indexation Google réelle (crawl runtime) + voix Amélie QC (multilingual_v2) exécutée sur runtime OTOIA VPS",
+          "source": "CLAUDE.md #8 (VPS) — indexation et TTS sont des services runtime hors repo"
+        }
+      ],
+      "n_modules": 2,
+      "all_modules_gated": true,
+      "all_artifacts_exist": true,
+      "has_evidence": true
+    },
+    {
+      "id": "S7",
+      "kind": "sprint_deliverable",
+      "sprint": "S7",
+      "roadmap_line": 70,
+      "statement": "Plateforme prête pour démo publique (polish · scénarios démo · audit 4Big 95+)",
+      "status": "in_repo",
+      "modules": [
+        {
+          "module": "demo/scenarios",
+          "ci_job": "demo-scenario-tests",
+          "gated": true
+        },
+        {
+          "module": "qa/audit_4big",
+          "ci_job": "qa-audit-4big-tests",
+          "gated": true
+        }
+      ],
+      "artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "polish final du site otov7.com (Hero · nav) sur le site LIVE",
+          "source": "roadmap Sprint 7 · Frontend « Site otov7.com polish final » (L67) — dépend du site LIVE · CLAUDE.md #8"
+        }
+      ],
+      "n_modules": 2,
+      "all_modules_gated": true,
+      "all_artifacts_exist": true,
+      "has_evidence": true
+    },
+    {
+      "id": "S8",
+      "kind": "sprint_deliverable",
+      "sprint": "S8",
+      "roadmap_line": 76,
+      "statement": "Plateforme démo live · 5 portails · CRM · BIM auto · OTOIA voice",
+      "status": "in_repo",
+      "modules": [
+        {
+          "module": "qa/regression",
+          "ci_job": "qa-regression-tests",
+          "gated": true
+        },
+        {
+          "module": "devops/deploy_runbook",
+          "ci_job": "devops-deploy-runbook-tests",
+          "gated": true
+        }
+      ],
+      "artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "déploiement production complet + monitoring exécutés sur le VPS (run-book appliqué)",
+          "source": "CLAUDE.md #8 (VPS) — l'exécution du run-book de déploiement revient à l'agent DevOps / ERPNext Backend"
+        }
+      ],
+      "n_modules": 2,
+      "all_modules_gated": true,
+      "all_artifacts_exist": true,
+      "has_evidence": true
+    },
+    {
+      "id": "M1",
+      "kind": "mvp_metric",
+      "sprint": null,
+      "roadmap_line": 81,
+      "statement": "Faisabilité auto générée en <1h (vs 3-5 jours actuels)",
+      "status": "in_repo",
+      "modules": [
+        {
+          "module": "publiciste",
+          "ci_job": "publiciste-tests",
+          "gated": true
+        },
+        {
+          "module": "faisabilite/generator",
+          "ci_job": "faisabilite-gen-tests",
+          "gated": true
+        },
+        {
+          "module": "faisabilite/bancable",
+          "ci_job": "bancable-tests",
+          "gated": true
+        }
+      ],
+      "artifacts": [],
+      "out_of_scope": [],
+      "n_modules": 3,
+      "all_modules_gated": true,
+      "all_artifacts_exist": true,
+      "has_evidence": true
+    },
+    {
+      "id": "M2",
+      "kind": "mvp_metric",
+      "sprint": null,
+      "roadmap_line": 82,
+      "statement": "Score 4Big >=95/100 sur 100% deliverables",
+      "status": "in_repo",
+      "modules": [
+        {
+          "module": "qa/audit_4big",
+          "ci_job": "qa-audit-4big-tests",
+          "gated": true
+        }
+      ],
+      "artifacts": [],
+      "out_of_scope": [],
+      "n_modules": 1,
+      "all_modules_gated": true,
+      "all_artifacts_exist": true,
+      "has_evidence": true
+    },
+    {
+      "id": "M3",
+      "kind": "mvp_metric",
+      "sprint": null,
+      "roadmap_line": 83,
+      "statement": "7 dashboards entités uniformes luxury",
+      "status": "in_repo",
+      "modules": [
+        {
+          "module": "frontend/portails",
+          "ci_job": "frontend-portails-tests",
+          "gated": true
+        },
+        {
+          "module": "rbac/roleprofile_gen",
+          "ci_job": "rbac-roleprofile-tests",
+          "gated": true
+        }
+      ],
+      "artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "rendu luxury dark+doré sur le desk ERPNext LIVE (Custom HTML)",
+          "source": "CLAUDE.md #8 (VPS) · #4 (design luxury rendu côté desk)"
+        }
+      ],
+      "n_modules": 2,
+      "all_modules_gated": true,
+      "all_artifacts_exist": true,
+      "has_evidence": true
+    },
+    {
+      "id": "M4",
+      "kind": "mvp_metric",
+      "sprint": null,
+      "roadmap_line": 84,
+      "statement": "CRM ERPNext natif fonctionnel end-to-end",
+      "status": "in_repo",
+      "modules": [
+        {
+          "module": "crm/workflow_vente",
+          "ci_job": "crm-workflow-vente-tests",
+          "gated": true
+        },
+        {
+          "module": "crm/dossier_vente",
+          "ci_job": "crm-dossier-vente-tests",
+          "gated": true
+        },
+        {
+          "module": "crm/commissions",
+          "ci_job": "crm-commissions-tests",
+          "gated": true
+        }
+      ],
+      "artifacts": [],
+      "out_of_scope": [],
+      "n_modules": 3,
+      "all_modules_gated": true,
+      "all_artifacts_exist": true,
+      "has_evidence": true
+    },
+    {
+      "id": "M5",
+      "kind": "mvp_metric",
+      "sprint": null,
+      "roadmap_line": 85,
+      "statement": "Mobile app publié 2 stores",
+      "status": "out_of_scope",
+      "modules": [],
+      "artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "builds Expo 54 + submit App Store #32 / Play Store — aucune fabrication en repo",
+          "source": "roadmap Sprint 5 · Mobile (L56-57) — builds & submissions hors repo · CLAUDE.md #8"
+        }
+      ],
+      "n_modules": 0,
+      "all_modules_gated": true,
+      "all_artifacts_exist": true,
+      "has_evidence": false
+    },
+    {
+      "id": "M6",
+      "kind": "mvp_metric",
+      "sprint": null,
+      "roadmap_line": 86,
+      "statement": "SEO trilingue indexé",
+      "status": "in_repo",
+      "modules": [
+        {
+          "module": "seo",
+          "ci_job": "seo-tests",
+          "gated": true
+        }
+      ],
+      "artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "indexation Google réelle (soumission sitemaps + crawl)",
+          "source": "CLAUDE.md #8 (VPS) — worker systemd SEO tourne hors repo"
+        }
+      ],
+      "n_modules": 1,
+      "all_modules_gated": true,
+      "all_artifacts_exist": true,
+      "has_evidence": true
+    },
+    {
+      "id": "M7",
+      "kind": "mvp_metric",
+      "sprint": null,
+      "roadmap_line": 87,
+      "statement": "OTOIA voice orchestre bout-en-bout",
+      "status": "in_repo",
+      "modules": [
+        {
+          "module": "frontend/chat_otoia",
+          "ci_job": "chat-otoia-tests",
+          "gated": true
+        }
+      ],
+      "artifacts": [],
+      "out_of_scope": [
+        {
+          "item": "voix Amélie QC (multilingual_v2) + orchestration AEC exécutées sur runtime OTOIA VPS",
+          "source": "CLAUDE.md #8 (VPS) — TTS et pilotage aec.py sont des services runtime hors repo"
+        }
+      ],
+      "n_modules": 1,
+      "all_modules_gated": true,
+      "all_artifacts_exist": true,
+      "has_evidence": true
+    }
+  ],
+  "verdict": true
+}

05_deliverables_mvp/qa/acceptance/tests/test_acceptance.py

@@ -0,0 +1,299 @@
+"""Tests de la Matrice d'acceptation / traçabilité MVP (stdlib pur · zéro pip).
+
+Couvre : le build réel (couverture bijective vs CI Gitea + registre 4Big), la
+conformité au schéma, les 10 familles d'invariants, le déterminisme, et une
+batterie d'injections négatives (chaque invariant DOIT rougir quand on le viole).
+
+Les négatifs s'appuient sur une fixture SYNTHÉTIQUE minimale (CI + registre +
+spec factices) pour contrôler exactement l'écart injecté, sans dépendre de la
+forme du dépôt réel.
+"""
+
+from __future__ import annotations
+
+import copy
+import os
+import sys
+import unittest
+
+_HERE = os.path.dirname(os.path.abspath(__file__))
+_MOD = os.path.normpath(os.path.join(_HERE, ".."))
+sys.path.insert(0, _MOD)
+
+import acceptance_gen as gen  # noqa: E402
+from acclib import builder, deps  # noqa: E402
+
+
+def _spec() -> dict:
+    return gen._load(gen._SPEC_PATH)
+
+
+# --------------------------------------------------------------------------- #
+# Fixture synthétique : deux modules gated (a/one=S2, b/two=S3) + la matrice.  #
+# --------------------------------------------------------------------------- #
+_SELF = "qa/acceptance"
+
+
+def _mini_ci(paths, self_module=_SELF) -> dict:
+    j2p = {f"{p.replace('/', '-')}-tests": p for p in paths}
+    j2p["qa-acceptance-tests"] = self_module
+    return {"job_to_path": j2p, "gate_needs": list(j2p)}
+
+
+def _mini_audit() -> dict:
+    return {"a/one": "S2", "b/two": "S3"}
+
+
+def _mini_spec() -> dict:
+    return {
+        "version": "test",
+        "reference_cadre": "fixture",
+        "roadmap_ref": "ROADMAP.md",
+        "self_module": _SELF,
+        "cible_portage": "recette",
+        "extra_module_sprint": {},
+        "sprint_deliverables": [
+            {"id": "S2", "kind": "sprint_deliverable", "sprint": "S2",
+             "roadmap_line": 39, "statement": "livrable S2", "status": "in_repo",
+             "evidence_modules": ["a/one"], "evidence_artifacts": [],
+             "out_of_scope": []},
+            {"id": "S3", "kind": "sprint_deliverable", "sprint": "S3",
+             "roadmap_line": 46, "statement": "livrable S3", "status": "in_repo",
+             "evidence_modules": ["b/two"], "evidence_artifacts": [],
+             "out_of_scope": []},
+        ],
+        "mvp_metrics": [
+            {"id": "M1", "kind": "mvp_metric", "sprint": None,
+             "roadmap_line": 81, "statement": "métrique M1", "status": "in_repo",
+             "evidence_modules": ["a/one"], "evidence_artifacts": [],
+             "out_of_scope": []},
+        ],
+    }
+
+
+def _mini_env():
+    """(spec, ci, audit) — la fixture minimale valide."""
+    return _mini_spec(), _mini_ci(["a/one", "b/two"]), _mini_audit()
+
+
+def _mini_errors(spec, ci=None, audit=None) -> list:
+    ci = ci or _mini_ci(["a/one", "b/two"])
+    audit = audit or _mini_audit()
+    bundle = builder.build_matrix(spec, ci, audit, deps)
+    return gen.validate_bundle(bundle, spec, ci, audit)
+
+
+# --------------------------------------------------------------------------- #
+class TestMiniClean(unittest.TestCase):
+    """La fixture synthétique doit valider proprement (base des négatifs)."""
+
+    def test_mini_validates_clean(self):
+        self.assertEqual(_mini_errors(*(_mini_env()[:1]),), [],
+                         "la fixture de base doit être verte")
+
+    def test_mini_bijective(self):
+        spec, ci, audit = _mini_env()
+        b = builder.build_matrix(spec, ci, audit, deps)
+        cov = b["manifest"]["coverage"]
+        self.assertTrue(cov["bijective"])
+        self.assertTrue(cov["partition_ok"])
+        self.assertTrue(b["verdict"])
+
+
+# --------------------------------------------------------------------------- #
+class TestRealBuild(unittest.TestCase):
+    """Build réel contre le CI, le registre 4Big et le spec du dépôt."""
+
+    def setUp(self):
+        self.spec = _spec()
+        self.ci = deps.parse_ci()
+        self.audit = deps.audit_sprints()
+        self.bundle = builder.build_matrix(self.spec, self.ci, self.audit, deps)
+
+    def test_build_validates_clean(self):
+        errs = gen.validate_bundle(self.bundle, self.spec, self.ci, self.audit)
+        self.assertEqual(errs, [], f"invariants rouges: {errs}")
+
+    def test_schema_conformance(self):
+        schema = gen._load(gen._SCHEMA_PATH)
+        self.assertEqual(list(deps.validate(self.bundle, schema)), [])
+
+    def test_coverage_bijective(self):
+        cov = self.bundle["manifest"]["coverage"]
+        self.assertTrue(cov["bijective"])
+        self.assertEqual(cov["missing_in_matrix"], [])
+        self.assertEqual(cov["phantom_evidence"], [])
+
+    def test_partition_exact_all_sprints(self):
+        cov = self.bundle["manifest"]["coverage"]
+        self.assertTrue(cov["partition_ok"])
+        for p in cov["sprint_partition"]:
+            self.assertTrue(p["exact"], f"partition {p['sprint']} non exacte")
+        self.assertEqual(cov["unknown_sprint"], [])
+
+    def test_verdict_pass(self):
+        self.assertTrue(self.bundle["verdict"])
+
+    def test_counts_match_roadmap(self):
+        c = self.bundle["manifest"]["counts"]
+        self.assertEqual(c["sprint_deliverables"], 8)   # S1..S8
+        self.assertEqual(c["mvp_metrics"], 7)           # M1..M7
+        self.assertEqual(c["gated_modules"], c["cited_modules"])
+
+    def test_every_gated_module_cited_once(self):
+        gated = set(builder.gated_modules(self.ci, self.spec["self_module"]))
+        cited = set()
+        for d in builder.all_deliverables(self.spec):
+            if d["status"] == "in_repo":
+                cited.update(d["evidence_modules"])
+        self.assertEqual(cited, gated)
+
+    def test_self_module_not_cited(self):
+        for d in builder.all_deliverables(self.spec):
+            self.assertNotIn(self.spec["self_module"], d["evidence_modules"])
+
+    def test_out_of_scope_all_sourced(self):
+        for d in builder.all_deliverables(self.spec):
+            for o in d.get("out_of_scope", []):
+                self.assertTrue(o.get("source", "").strip(),
+                                f"{d['id']}: hors-périmètre sans source")
+
+    def test_evidence_artifacts_exist(self):
+        for d in builder.all_deliverables(self.spec):
+            for a in d.get("evidence_artifacts", []):
+                self.assertTrue(deps.artifact_exists(a), f"{a} absent")
+
+    def test_determinism(self):
+        b2 = builder.build_matrix(self.spec, self.ci, self.audit, deps)
+        self.assertEqual(self.bundle, b2)
+
+    def test_sprint_windows_from_audit_registry(self):
+        # Toute fenêtre de sprint vient du registre 4Big (sauf l'auditeur, via
+        # extra) — jamais re-déclarée dans le spec d'acceptation.
+        ms = builder.module_sprint_map(self.spec, self.audit)
+        gated = builder.gated_modules(self.ci, self.spec["self_module"])
+        for m in gated:
+            self.assertIn(m, ms, f"{m} sans fenêtre de sprint")
+
+    def test_audit_4big_declared_via_extra(self):
+        self.assertIn("qa/audit_4big", self.spec["extra_module_sprint"])
+        self.assertTrue(
+            self.spec["extra_module_sprint"]["qa/audit_4big"]["source"].strip())
+
+
+# --------------------------------------------------------------------------- #
+class TestNegativeInjections(unittest.TestCase):
+    """Chaque invariant DOIT rougir quand on injecte l'écart correspondant."""
+
+    def test_inv2_phantom_module_flagged(self):
+        spec = _mini_spec()
+        spec["sprint_deliverables"][0]["evidence_modules"].append("z/ghost")
+        self.assertTrue(any("non gated" in e or "fantôme" in e
+                            for e in _mini_errors(spec)))
+
+    def test_inv3_missing_gated_module(self):
+        # b/two gated mais retiré de son livrable → couverture non bijective.
+        spec = _mini_spec()
+        spec["sprint_deliverables"][1]["evidence_modules"] = []
+        # laisser un artefact pour ne pas déclencher AUSSI INV7
+        spec["sprint_deliverables"][1]["evidence_artifacts"] = ["GAP_ANALYSIS_SPRINT1.md"]
+        errs = _mini_errors(spec)
+        self.assertTrue(any("non tracés" in e for e in errs), errs)
+
+    def test_inv5_partition_wrong_sprint(self):
+        # Placer b/two (fenêtre S3) dans le livrable S2 → partition fausse.
+        spec = _mini_spec()
+        spec["sprint_deliverables"][0]["evidence_modules"] = ["a/one", "b/two"]
+        spec["sprint_deliverables"][1]["evidence_modules"] = []
+        spec["sprint_deliverables"][1]["evidence_artifacts"] = ["GAP_ANALYSIS_SPRINT1.md"]
+        self.assertTrue(any("partition" in e for e in _mini_errors(spec)))
+
+    def test_inv4_module_without_sprint(self):
+        # Registre 4Big ne connaît pas b/two → module gated sans fenêtre.
+        spec = _mini_spec()
+        bad_audit = {"a/one": "S2"}  # b/two manquant
+        errs = _mini_errors(spec, audit=bad_audit)
+        self.assertTrue(any("sans fenêtre" in e for e in errs), errs)
+
+    def test_inv4_extra_conflicts_with_registry(self):
+        spec = _mini_spec()
+        spec["extra_module_sprint"] = {"a/one": {"sprint": "S9", "source": "x"}}
+        self.assertTrue(any("conflit" in e for e in _mini_errors(spec)))
+
+    def test_inv6_out_of_scope_without_source(self):
+        spec = _mini_spec()
+        spec["sprint_deliverables"][0]["out_of_scope"] = [{"item": "x", "source": "  "}]
+        self.assertTrue(any("sans source" in e for e in _mini_errors(spec)))
+
+    def test_inv7_in_repo_without_evidence(self):
+        spec = _mini_spec()
+        spec["mvp_metrics"][0]["evidence_modules"] = []
+        self.assertTrue(any("sans aucune preuve" in e for e in _mini_errors(spec)))
+
+    def test_inv7_out_of_scope_with_module(self):
+        spec = _mini_spec()
+        spec["mvp_metrics"].append({
+            "id": "M2", "kind": "mvp_metric", "sprint": None, "roadmap_line": 82,
+            "statement": "m2", "status": "out_of_scope",
+            "evidence_modules": ["a/one"], "evidence_artifacts": [],
+            "out_of_scope": [{"item": "x", "source": "s"}]})
+        self.assertTrue(any("out_of_scope avec preuve" in e for e in _mini_errors(spec)))
+
+    def test_inv7_missing_artifact(self):
+        spec = _mini_spec()
+        spec["sprint_deliverables"][0]["evidence_artifacts"] = ["does/not/exist.md"]
+        self.assertTrue(any("inexistant" in e for e in _mini_errors(spec)))
+
+    def test_inv8_metric_with_sprint(self):
+        spec = _mini_spec()
+        spec["mvp_metrics"][0]["sprint"] = "S4"
+        self.assertTrue(any("sprint non nul" in e for e in _mini_errors(spec)))
+
+    def test_inv8_sprint_deliverable_without_sprint(self):
+        spec = _mini_spec()
+        spec["sprint_deliverables"][0]["sprint"] = None
+        self.assertTrue(any("sans sprint" in e or "sprint" in e
+                            for e in _mini_errors(spec)))
+
+    def test_inv9_self_cited(self):
+        spec = _mini_spec()
+        # citer la matrice elle-même comme preuve (et la rendre gated)
+        ci = _mini_ci(["a/one", "b/two"])
+        ci["job_to_path"]["extra-job"] = _SELF  # reste self → exclu du gated
+        spec["sprint_deliverables"][0]["evidence_modules"] = ["a/one", _SELF]
+        self.assertTrue(any("elle-même" in e or "non gated" in e
+                            for e in _mini_errors(spec, ci=ci)))
+
+    def test_inv1_duplicate_id(self):
+        spec = _mini_spec()
+        spec["mvp_metrics"][0]["id"] = "S2"  # collision avec un sprint
+        self.assertTrue(any("double" in e for e in _mini_errors(spec)))
+
+    def test_inv3_flags_bijective_false_consistency(self):
+        # Retirer un module ET son artefact → bijective doit passer à False.
+        spec = _mini_spec()
+        spec["sprint_deliverables"][1]["evidence_modules"] = []
+        spec["sprint_deliverables"][1]["evidence_artifacts"] = ["GAP_ANALYSIS_SPRINT1.md"]
+        b = builder.build_matrix(spec, _mini_ci(["a/one", "b/two"]), _mini_audit(), deps)
+        self.assertFalse(b["manifest"]["coverage"]["bijective"])
+        self.assertFalse(b["verdict"])
+
+
+# --------------------------------------------------------------------------- #
+class TestCLI(unittest.TestCase):
+    """Le CLI build/validate produit les artefacts et sort proprement."""
+
+    def test_build_writes_outputs(self):
+        import tempfile
+        with tempfile.TemporaryDirectory() as tmp:
+            gen.main(["build", "-o", tmp])
+            self.assertTrue(os.path.exists(os.path.join(tmp, "acceptance_matrix.json")))
+            self.assertTrue(os.path.exists(os.path.join(tmp, "MANIFEST.json")))
+
+    def test_validate_exits_zero(self):
+        # cmd_validate ne lève pas sur le dépôt réel (sinon SystemExit).
+        gen.main(["validate"])
+
+
+if __name__ == "__main__":
+    unittest.main()