exam: capture sections + choice_groups + marks_confidence in extraction_meta (WS-2 R3)

Persist the paper's question-layout signals (A-level Section A/B, EITHER/OR choice
groups, marks confidence) onto exam_templates.extraction_meta so the setup UI can
show them. Data was carried by the contract but dropped at persistence.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
Claude-Session: https://claude.ai/code/session_01GruxHXxfdp4kZCgAMVFgvV

modules/services/exam_extract.py

@@ -40,6 +40,19 @@ def _get(base: str, slug: str, timeout: int = 30) -> Dict[str, Any]:
     return r.json()
 
 
+def get_replica(slug: str, timeout: int = 30) -> Dict[str, Any]:
+    """The digital-replica markdown for a paper (P4): {slug, title, n_questions, total_marks, markdown,
+    questions:[{label, marks, markdown}]}. Raises ExtractError if the paper has no replica yet."""
+    base = service_url()
+    if not base:
+        raise ExtractError("EXAM_EXTRACT_URL not configured")
+    r = requests.get(f"{base}/api/replica/{slug}", timeout=timeout)
+    if r.status_code == 404:
+        raise ExtractError(f"no digital replica for {slug}")
+    r.raise_for_status()
+    return r.json()
+
+
 def extract_suggestions(slug: str, pdf_bytes: bytes, *, force: bool = False,
                         poll_timeout: int = 1500, poll_interval: int = 5) -> Dict[str, Any]:
     """POST the paper to the service and poll until the analyse contract is ready.

routers/exam/schemas.py

@@ -80,6 +80,9 @@ class ResponseAreaPayload(BaseModel):
     ] = None
     # Optional Context differentiation (v1 generic; future graph/chart/data_table/diagram/code_block/passage).
     context_type: Optional[str] = None
+    # Rich recognition payload (75-exam-marker-region-meta.sql): figure name/description, OMR geometry, unit…
+    # Carried on canvas save so a named context figure survives a round-trip.
+    meta: Optional[Dict[str, Any]] = None
     source: Literal["manual", "ai"] = "manual"
     confirmed: bool = True
     confidence: Optional[float] = Field(default=None, ge=0, le=1)

routers/exam/templates.py

@@ -745,6 +745,9 @@ def _map_service_contract_to_rows(template_id: str, contract: Dict[str, Any],
             "bounds": _frac_box_to_canvas(q.get("bounds"), q.get("page") or 1, pages),
             "page": q.get("page"), "source": "ai", "confirmed": False,
             "confidence": _safe_confidence(q.get("confidence")), "derivation": "extract-service",
+            # analyse contract v2 (migration 78): command verb + stem prose per part
+            "command_word": (q.get("command_word") or None) if not q.get("is_container") else None,
+            "preamble": q.get("preamble") or None,
         })
     for q in questions:                                 # FK safety: de-parent a dangling parent_id
         if q["parent_id"] and q["parent_id"] not in q_ids:
@@ -793,9 +796,21 @@ def _run_service_extract_merge(ctx: ExamContext, template_id: str, pdf_bytes: by
     contract = exam_extract.extract_suggestions(slug, pdf_bytes)
     rows = _map_service_contract_to_rows(template_id, contract, pdf_bytes)
     _refresh_ai_rows(ctx, template_id, rows)
-    n_pages = (contract.get("meta") or {}).get("n_pages") or (contract.get("meta") or {}).get("pages")
+    meta = contract.get("meta") or {}
+    # P3: record provenance + the audit cover-reconciliation gate so the setup UI can flag under-reads,
+    # and store the slug so the digital-text view can resolve this paper's replica.
+    updates: Dict[str, Any] = {"extraction_meta": {
+        "engine": "extract-service", "slug": slug, "audit": meta.get("audit") or {},
+        "counts": {k: len(v) for k, v in rows.items()},
+        # question-layout signals for the setup UI: paper sections (e.g. A-level Section A/B) + EITHER/OR choices
+        "sections": meta.get("sections") or [],
+        "choice_groups": meta.get("choice_groups") or [],
+        "marks_confidence": meta.get("marks_confidence"),
+    }}
+    n_pages = meta.get("n_pages") or meta.get("pages")
     if n_pages:
-        ctx.supabase.table("exam_templates").update({"page_count": n_pages}).eq("id", template_id).execute()
+        updates["page_count"] = n_pages
+    ctx.supabase.table("exam_templates").update(updates).eq("id", template_id).execute()
     return rows
 
 
@@ -1034,6 +1049,24 @@ async def auto_map_status(
     return body
 
 
+@router.get("/templates/{template_id}/digital-text")
+async def template_digital_text(
+    template_id: str,
+    ctx: ExamContext = Depends(get_exam_context),
+) -> Dict[str, Any]:
+    """P4: the digital-replica markdown for this template's paper (stem text, parts, marks, answer-space
+    placeholders, spec refs). Resolves the paper via the slug the extraction used."""
+    template = _fetch_template_or_404(ctx, template_id)
+    _require_source_visibility_or_404(ctx, template)
+    if not exam_extract.is_enabled():
+        raise HTTPException(status_code=503, detail="Extraction service not configured")
+    slug = (template.get("extraction_meta") or {}).get("slug") or _extract_slug(ctx, template)
+    try:
+        return exam_extract.get_replica(slug)
+    except exam_extract.ExtractError as exc:
+        raise HTTPException(status_code=404, detail=f"No digital text yet — run auto-map first ({exc})")
+
+
 @router.put("/templates/{template_id}")
 async def replace_template(
     template_id: str,
@@ -1114,6 +1147,7 @@ async def replace_template(
                 "kind": ra.kind,
                 "response_form": ra.response_form,
                 "context_type": ra.context_type,   # 73: optional Context differentiation
+                "meta": ra.meta,                   # 75: rich recognition payload (name/description/OMR/…)
                 "source": ra.source,
                 "confirmed": ra.confirmed,
                 "confidence": ra.confidence,