fix: back off after ops claim errors

Assistant: codex
Assistant-Model: gpt-5.6-sol
Assistant-Session: 01a02b6f-7db1-7222-918b-e813a6bda38d

rein_aharness/claim_loop.py

@@ -42,6 +42,7 @@ logger = logging.getLogger("rein_aharness.claim_loop")
 class ProcessResult:
     claimed: bool
     empty: bool = False
+    retry_full_interval: bool = False
     run_id: str | None = None
     approach: str | None = None
     ok: bool | None = None
@@ -104,7 +105,11 @@ def process_one(
     try:
         claimed = client.claim(limit=1)
     except OpsRunError as exc:
-        return ProcessResult(claimed=False, reason=f"claim error: {exc}")
+        return ProcessResult(
+            claimed=False,
+            retry_full_interval=True,
+            reason=f"claim error: {exc}",
+        )
 
     if not claimed:
         return ProcessResult(claimed=False, empty=True, reason="queue empty")
@@ -396,8 +401,13 @@ def run_claim_loop(
             break
         if max_iterations is not None and iterations >= max_iterations:
             break
-        # Sleep full interval only when empty; short pause after work
-        sleep_for = interval_seconds if result.empty else min(2.0, interval_seconds)
+        # Empty queues and upstream errors both use the configured backoff.
+        # The short pause is only for a cycle that actually claimed work.
+        sleep_for = (
+            interval_seconds
+            if result.empty or result.retry_full_interval
+            else min(2.0, interval_seconds)
+        )
         stop.wait(sleep_for)
 
     logger.info("claim-loop stop iterations=%s exit=%s", iterations, exit_code)