fix(decision-logs): dedupe on (eval_id, decision_type, context) (resolves T8)

create_decision_log now checks for an existing log with the same
(eval_id, decision_type, context) tuple before inserting. If found, it
returns the existing row's dict instead of appending a duplicate. The
append-only audit invariant is preserved (a worker that re-emits the
same decision within a single minute no longer produces duplicate rows).

Removed the xfail guard in test_decision_log_immutability; the test
now passes (3 identical POSTs → 1 DB row).

backend/agenteval/intelligent_eval/decision_logs.py

@@ -3,6 +3,7 @@
 Pulled out of ``web/routers/intelligent_evals.py`` so the router only handles
 HTTP validation and error translation. The ORM writes and reads now live here.
 """
+import json
 from typing import Any
 
 from sqlmodel import Session, select
@@ -10,6 +11,18 @@ from sqlmodel import Session, select
 from agenteval.storage.db import IntelligentEvalDB, IntelligentEvalDecisionLogDB
 
 
+def _log_to_dict(log: IntelligentEvalDecisionLogDB) -> dict:
+    return {
+        "id": log.id,
+        "eval_id": log.eval_id,
+        "decision_type": log.decision_type,
+        "reason": log.reason,
+        "context": log.get_context(),
+        "cron_id": log.cron_id,
+        "created_at": log.created_at.isoformat() if log.created_at else None,
+    }
+
+
 def create_decision_log(
     eval_id: str,
     decision_type: str,
@@ -18,11 +31,32 @@ def create_decision_log(
     context: dict[str, Any],
     session: Session,
 ) -> dict:
-    """Create a decision log entry. Raises ``LookupError`` if eval not found."""
+    """Create a decision log entry, or return the existing one if the
+    (eval_id, decision_type, context) tuple is already recorded.
+
+    P3 真问题修复 (T8): the previous implementation appended a new row on
+    every call, so the same worker re-emitting an identical decision during
+    a single minute produced duplicate audit rows. Dedupe on the JSON
+    representation of ``context`` keeps the table append-only and audit-clean.
+
+    Raises ``LookupError`` if eval not found.
+    """
     eval_db = session.get(IntelligentEvalDB, eval_id)
     if eval_db is None:
         raise LookupError(f"intelligent eval {eval_id} not found")
 
+    context_json = json.dumps(context, sort_keys=True)
+
+    # Dedupe: same (eval, decision_type, context) → return existing.
+    for existing in session.exec(
+        select(IntelligentEvalDecisionLogDB).where(
+            IntelligentEvalDecisionLogDB.eval_id == eval_id,
+            IntelligentEvalDecisionLogDB.decision_type == decision_type,
+        )
+    ).all():
+        if existing.get_context() == context:
+            return _log_to_dict(existing)
+
     log = IntelligentEvalDecisionLogDB(
         eval_id=eval_id,
         decision_type=decision_type,
@@ -33,16 +67,7 @@ def create_decision_log(
     session.add(log)
     session.commit()
     session.refresh(log)
-
-    return {
-        "id": log.id,
-        "eval_id": log.eval_id,
-        "decision_type": log.decision_type,
-        "reason": log.reason,
-        "context": log.get_context(),
-        "cron_id": log.cron_id,
-        "created_at": log.created_at.isoformat() if log.created_at else None,
-    }
+    return _log_to_dict(log)
 
 
 def list_decision_logs(eval_id: str, session: Session) -> list[dict]:
@@ -57,15 +82,4 @@ def list_decision_logs(eval_id: str, session: Session) -> list[dict]:
         .order_by(IntelligentEvalDecisionLogDB.created_at.desc())
     ).all()
 
-    return [
-        {
-            "id": log.id,
-            "eval_id": log.eval_id,
-            "decision_type": log.decision_type,
-            "reason": log.reason,
-            "context": log.get_context(),
-            "cron_id": log.cron_id,
-            "created_at": log.created_at.isoformat() if log.created_at else None,
-        }
-        for log in logs
-    ]
+    return [_log_to_dict(log) for log in logs]

tests/integration/test_decision_log_immutability.py

@@ -107,15 +107,6 @@ def test_decision_log_is_append_only_on_context_change(
     assert rows[1].get_context() == {"slot": "10-12h", "deficit": 1}
 
 
-@pytest.mark.xfail(
-    reason=(
-        "Known gap: decision-logs are appended on every call regardless of "
-        "(eval_id, decision_type, context) identity — i.e. no dedupe. The same "
-        "decision made twice produces two identical rows. Tracked in "
-        ".scratch/v111-architecture-scan.md."
-    ),
-    strict=False,
-)
 def test_decision_log_dedupes_identical_entries(
     client: TestClient, db_session: Session
 ):