fix: a body that stops mid-read is transient, and a round that dies still says so

Round 5 died on `http.client.IncompleteRead(0 bytes read)` and left no record
at all. Two defects, both ours.

The retry classified only connection-level failures. IncompleteRead and
RemoteDisconnected arrive while the *body* is being read, after the connection
succeeded, so they are not URLError -- they escaped both the retry and the
error wrapper and walked out of the round. The network is allowed to be
unreliable; losing a paid round to it is not the provider's defect.

And the exception path was wrapped around the stages that come *after* the
proposal is applied, because those are the ones with something to revert. The
stages before it had already spent money, and a round that leaves no record is
a round nobody can account for. Every outcome is journalled now, with the
exception type in the reason.

Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Claude-Session: https://claude.ai/code/session_01T7XD7x5KonMdmGRkp63EQG

src/openultrasast/learning/rounds.py

@@ -406,10 +406,25 @@ def run_learning_round(
         spent = spent_usd() if spent_usd is not None else 0.0
         return not within_budget(spent, cost_cap_usd)
 
-    before_scan = scan_factory(config)
-    before_train = _score(batch, before_scan, family)
-    before_hold = _score(holdout, before_scan, family)
     record = _round_record(round_number, family, None, taxonomy, config)
+    try:
+        before_scan = scan_factory(config)
+        before_train = _score(batch, before_scan, family)
+        before_hold = _score(holdout, before_scan, family)
+    except Exception as exc:
+        # These stages run before anything is applied, so there is nothing to revert — but they have already spent
+        # money, and a round that leaves no record is a round nobody can account for. Measured: round 5 died on a
+        # network read during its first stage and vanished.
+        _finish(
+            journal,
+            directory,
+            record,
+            outcome="reverted",
+            reason=f"stage_failed: {type(exc).__name__}",
+            cost=spent_usd() if spent_usd is not None else 0.0,
+            trajectories=trajectories,
+        )
+        raise
     if _over_cap():
         return _finish(
             journal, directory, record, outcome="reverted_cost", reason="cost_cap_exceeded", cost=spent, trajectories=trajectories
@@ -551,7 +566,7 @@ def run_learning_round(
             version=version,
             trajectories=trajectories,
         )
-    except Exception:
+    except Exception as exc:
         # A stage that raises has already spent money and already changed the tree. Both facts belong in the
         # journal before the exception leaves, or the next round starts from a state nothing describes.
         snapshot.restore()
@@ -560,7 +575,7 @@ def run_learning_round(
             directory,
             record,
             outcome="reverted",
-            reason="stage_failed",
+            reason=f"stage_failed: {type(exc).__name__}",
             proposal=proposal,
             cost=spent,
             trajectories=trajectories,

src/openultrasast/provider/openrouter.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import http.client
 import json
 import os
 import re
@@ -20,10 +21,14 @@ _T = TypeVar("_T")
 
 
 def _is_transient(exc: Exception) -> bool:
-    """A network failure worth retrying (rate-limit, 5xx, connection error, timeout)."""
+    """A network failure worth retrying: rate-limit, 5xx, connection error, timeout, or a body that stopped.
+
+    `http.client.IncompleteRead` and `RemoteDisconnected` arrive while the *body* is being read, after the
+    connection succeeded, so they are not `URLError`. Measured: one of them ended an evolve round that had already
+    been paid for."""
     if isinstance(exc, urllib.error.HTTPError):
         return exc.code in _TRANSIENT_STATUS
-    return isinstance(exc, urllib.error.URLError | TimeoutError)
+    return isinstance(exc, urllib.error.URLError | TimeoutError | http.client.HTTPException)
 
 
 def call_with_retry(
@@ -108,8 +113,8 @@ class OpenRouterChatClient:
 
         try:
             return call_with_retry(_do, attempts=self.max_attempts, base_delay=self.retry_base_delay)
-        except (urllib.error.URLError, TimeoutError, json.JSONDecodeError) as exc:
-            raise OpenRouterError(f"OpenRouter chat request failed: {exc}") from exc
+        except (urllib.error.URLError, TimeoutError, http.client.HTTPException, json.JSONDecodeError) as exc:
+            raise OpenRouterError(f"OpenRouter chat request failed: {type(exc).__name__}: {exc}") from exc
 
     def complete_json(self, *, model: str, messages: list[dict[str, str]], timeout_seconds: int = 60) -> object:
         message = self.complete_chat(model=model, messages=messages, timeout_seconds=timeout_seconds)
@@ -151,8 +156,8 @@ class OpenRouterEmbeddingClient:
 
         try:
             response_payload = call_with_retry(_do, attempts=self.max_attempts, base_delay=self.retry_base_delay)
-        except (urllib.error.URLError, TimeoutError, json.JSONDecodeError) as exc:
-            raise OpenRouterError(f"OpenRouter embedding request failed: {exc}") from exc
+        except (urllib.error.URLError, TimeoutError, http.client.HTTPException, json.JSONDecodeError) as exc:
+            raise OpenRouterError(f"OpenRouter embedding request failed: {type(exc).__name__}: {exc}") from exc
         return parse_embedding_response(response_payload)