fix(lineage): wire the adjudication client into corpus-wide rebuild

backend/app/analysis_run_worker.py

@@ -79,6 +79,8 @@ async def consume_analysis_run_stream_once(
                         exc.status_code,
                         exc.detail,
                     )
+                except Exception:
+                    logger.exception("analysis-run %s delivery failed", analysis_run_id)
             last_id = str(entry_id)
     return last_id
 

lineageweave/adjudication_client.py

@@ -15,7 +15,7 @@
 import re
 from typing import Protocol
 
-from .http_client import chat_completion_content, post_json
+from .http_client import HttpClientError, chat_completion_content, post_json
 
 
 class AdjudicationClient(Protocol):
@@ -53,9 +53,11 @@ def judge_prompt(candidate_label: str, record_label: str) -> str:
 
 
 def parse_confidence(content: str) -> float:
-    """Clamp the judge's numeric reply into [0, 1]; no number reads as 0."""
+    """Clamp a numeric reply into ``[0, 1]`` or fail without inventing zero."""
     parsed = parse_confidence_or_none(content)
-    return 0.0 if parsed is None else parsed
+    if parsed is None:
+        raise HttpClientError("adjudication response had no confidence score")
+    return parsed
 
 
 def parse_confidence_or_none(content: str) -> float | None:
@@ -102,4 +104,8 @@ def judge(self, candidate_label: str, record_label: str) -> float:
             headers={"authorization": f"Bearer {self._api_key}"},
             timeout=self._timeout,
         )
-        return parse_confidence(chat_completion_content(body))
+        try:
+            content = chat_completion_content(body)
+        except (TypeError, ValueError) as exc:
+            raise HttpClientError("adjudication response did not contain text") from exc
+        return parse_confidence(content)

tests/test_adjudication_client.py

@@ -1,6 +1,9 @@
 from __future__ import annotations
 
+import pytest
+
 from lineageweave.adjudication_client import ContextualOrchestratorAdjudicationClient
+from lineageweave.http_client import HttpClientError
 
 
 def test_adjudication_uses_supported_auto_mode_and_long_local_timeout(monkeypatch) -> None:
@@ -21,3 +24,24 @@ def fake_post_json(url, payload, *, headers, timeout):
     assert captured["payload"]["mode"] == "auto"
     assert captured["payload"]["reasoning_effort"] == "auto"
     assert captured["timeout"] == 180.0
+
+
+@pytest.mark.parametrize(
+    "body",
+    [
+        {"choices": [{"message": {"content": "not a score"}}]},
+        {"choices": []},
+        {"choices": [{"message": {"content": None}}]},
+    ],
+)
+def test_adjudication_fails_closed_for_unscoreable_responses(monkeypatch, body) -> None:
+    """A provider failure must not become a genuine unrelated score of zero."""
+    monkeypatch.setattr(
+        "lineageweave.adjudication_client.post_json", lambda *args, **kwargs: body
+    )
+    client = ContextualOrchestratorAdjudicationClient(
+        base_url="http://orchestrator:8000", api_key="synthetic-token"
+    )
+
+    with pytest.raises(HttpClientError):
+        client.judge("workshop", "follow-up bid")

tests/test_analysis_run_worker.py

@@ -130,3 +130,28 @@ async def fake_deliver(conn, **kwargs):
 
     assert last_id == "2-1"
     assert delivered == ["00000000-0000-0000-0000-000000000002"]
+
+
+@pytest.mark.anyio
+async def test_one_unexpected_delivery_failure_does_not_end_the_worker(monkeypatch):
+    """A malformed provider reply must not stop later durable deliveries."""
+    delivered = []
+
+    async def fake_deliver(conn, **kwargs):
+        del conn
+        if kwargs["analysis_run_id"].endswith("1"):
+            raise RuntimeError("malformed provider reply")
+        delivered.append(kwargs["analysis_run_id"])
+
+    monkeypatch.setattr(analysis_run_worker, "deliver_queued_analysis_run", fake_deliver)
+
+    last_id = await analysis_run_worker.consume_analysis_run_stream_once(
+        _TwoRunsValkey(),
+        _Pool(),
+        last_id="0-0",
+        tepp_client=TeppClient(),
+        adjudication_client=NullAdjudicationClient(),
+    )
+
+    assert last_id == "2-1"
+    assert delivered == ["00000000-0000-0000-0000-000000000002"]

tests/test_estimate_llm_channel_weights_script.py

@@ -8,7 +8,10 @@
 
 from __future__ import annotations
 
+import pytest
+
 from lineageweave.adjudication_client import judge_prompt, parse_confidence
+from lineageweave.http_client import HttpClientError
 
 import scripts.estimate_llm_channel_weights as script
 
@@ -31,7 +34,8 @@ def test_shared_judge_prompt_and_confidence_parse_round_trip() -> None:
     assert "Record B: Follow-up record" in prompt
     assert parse_confidence("0.85") == 0.85
     assert parse_confidence("confidence: 0.4 maybe") == 0.4
-    assert parse_confidence("no number here") == 0.0
+    with pytest.raises(HttpClientError):
+        parse_confidence("no number here")
     assert parse_confidence("1.7") == 1.0