update

backend/guide_board.py

@@ -35,6 +35,8 @@ STAGE_LABELS = {"lernziele": "Lernziele", "zuweisung": "Zuweisung", "writer": "W
                 "fakten_gate": "Fakten-Gate", "coverage": "Coverage",
                 "lesbarkeit": "Lesbarkeit", "done": "Fertig"}
 MAX_WRITER_ROUNDS = 2   # coverage → writer feedback loop cap (gains die after round 1–2)
+GATE_FIX_MIN = 3        # fact-gate claims below this: log only — a full fix rewrite for
+                        # 1–2 residual claims fired on 19/20 blocks (40 agent-minutes)
 # Simultaneous cards = the per-topic agent cap: every card busies exactly ONE agent at a
 # time (its stages run serially), so a lower number just idles slots (was hardcoded 10
 # from the old 10-slot era while the .env already allowed 24).
@@ -395,6 +397,11 @@ async def _stage_fakten_gate(env: _Env, card: dict) -> bool:
     if status == FAILED:
         claims = []  # gate failure must not block the card — logged, text stands
         _log(env.topic, f"Fakten-Gate {card['block']}: kein Ergebnis — Text bleibt ungeprüft")
+    if claims and len(claims) < GATE_FIX_MIN:
+        # 1–2 Rest-Claims rechtfertigen keinen Voll-Rewrite: der Fix-Pass lief für 19/20
+        # Blöcke und kostete 40 Agent-Minuten; die Guide-QA-Fachlichkeits-Stichprobe misst nach
+        _log(env.topic, f"Fakten-Gate {card['block']}: {len(claims)} Claim(s) unter Schwelle — kein Fix")
+        claims = []
     if claims:
         _log(env.topic, f"Fakten-Gate {card['block']}: {len(claims)} unbelegte Claims → Fix")
         fixp = env.slot(f"gatefix-{_safe(norm)}-r{card['writer_rounds']}.md")