Training-Harness (ACO, Multi-Fidelity), Prüfstand-Benchmark, Agenten-README

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

backend/tests/conftest.py

@@ -22,63 +22,10 @@ async def testdb(tmp_path, monkeypatch):
 async def fake_welt(testdb, tmp_path, monkeypatch):
     """E2E ohne LLM: run_agent überall durch die Fake-Welt ersetzt, Tempo-Bremsen raus.
     Alle echten Schichten (_race, Quorum, Panels, Producer, QA-Gate) laufen mit."""
-    import agents
-    import blocks
-    import board_inventory as bi
-    import guide
-    import kanban
-    import pipeline
     import qa
-    import repair
-    from fake_agents import Welt
+    from fake_agents import Welt, aktivieren
 
     welt = Welt()
-
-    async def fake_run_agent(agent_key, prompt, timeout, provider="claude", role="fast",
-                             capabilities="none", lane="batch", scope=None, on_line=None, label=""):
-        return welt.respond(agent_key, prompt, capabilities)
-
-    for mod in (agents, pipeline, blocks, guide, repair):
-        monkeypatch.setattr(mod, "run_agent", fake_run_agent)
-
-    # Tempo: grace/poll/backoff bremsen echte Läufe, nicht den Fake
-    monkeypatch.setattr(blocks, "CONSENSUS_GRACE", 0)
-    monkeypatch.setattr(bi, "_QA_GATE_POLL", 0.05)
-    monkeypatch.setattr(kanban, "RETRY_BACKOFF", 0.05)
-    monkeypatch.setattr(qa, "QA_DIR", tmp_path / "qa")
-    import guide_board
-    monkeypatch.setattr(guide_board, "READABILITY_ACTIVE", False)  # kein Modell-Load im Test
-    import asyncio as _aio
-    monkeypatch.setattr(bi, "_ingest_lock", _aio.Lock())  # Modul-Lock klebt sonst am Vortest-Loop
-
-    class _FakeEmb:  # identischer Text → cos 1.0, sonst 0.0 (deterministisch, ohne Modell)
-        @staticmethod
-        def available():
-            return True
-
-        @staticmethod
-        def embed_sims(texts):
-            import numpy as np
-            uniq = {t: k for k, t in enumerate(dict.fromkeys(texts))}
-            arr = np.zeros((len(texts), max(len(uniq), 1)))
-            for r, t in enumerate(texts):
-                arr[r, uniq[t]] = 1.0
-            return arr @ arr.T
-
-        @staticmethod
-        def embed(texts):
-            import numpy as np
-            uniq = {t: k for k, t in enumerate(dict.fromkeys(texts))}
-            arr = np.zeros((len(texts), max(len(uniq), 1)))
-            for r, t in enumerate(texts):
-                arr[r, uniq[t]] = 1.0
-            return arr
-
-    import board_artefacts as ba
-    for mod in (blocks, ba, qa):
-        monkeypatch.setattr(mod, "embedding", _FakeEmb)
-
-    async def emb_ok(flow):  # Board-1-Vektorpfade aus (wie board_env) — Judge-Wellen reichen
-        return False
-    monkeypatch.setattr(bi, "_emb_ok", emb_ok)
+    aktivieren(welt, setattr_fn=monkeypatch.setattr)
+    monkeypatch.setattr(qa, "QA_DIR", tmp_path / "qa")  # Reports nie in echte Nutzdaten
     return welt