Preprocess status=error triggers force retry and 0.3s polling loop — investigate backend + SDK handling

[u2135]

Summary

In some environments preprocess GET returns status='error' (e.g. status_detail mentioning internal server error). SDK logs preprocess_error_retrying_with_force, POSTs {"force": true} again, then continues polling.

Current behavior (SDK)

• run_preprocess treats failed / error as recoverable once: second POST with force: True, then loop continues.
• Between polls, non-terminal states sleep 0.3s (time.sleep(0.3)).

Questions

1. Backend: Why does preprocess return error intermittently or always in affected deployments? Fix server-side if misconfiguration or bug.
2. SDK: Should repeated error after force fail fast instead of polling until timeout/raise?
3. SDK: Should status_detail be logged for operator visibility?

Acceptance criteria

• Reproduce with backend team; capture raw GET response for preprocess endpoint.
• If backend fixed: confirm intercept latency drops to expected (near proof/query path only).
• SDK: decide on fail-fast / backoff / max wait after error post-force (separate small PR).

References

• src/provably/handoff/_preprocess.py — run_preprocess, log preprocess_error_retrying_with_force