daily-fix: batch_judge tolerates transient 404 on poll retrieve

kind: infra#daily-auto-filed

track:

Highlight text on any card (body, plan, or an event) to anchor a comment, or leave a whole-task comment here. Mention @claude to summon a reply.

No comments yet.

Overview / Motivation

Auto-filed by /daily 2026-07-03 (backfill run 2026-07-04) from the day's transcript problem sweep.

Bounded transient-404 retry (backoff, N attempts over a few minutes) on the poll retrieve beyond the create grace, before declaring a batch lost.

Bug observed: 2026-07-03 #742: a transient batches.retrieve 404 MID-POLL crashed the driver (epm:failure infra); the batch was alive server-side and the 6th launch later harvested it 10/10 — a full relaunch cycle lost. #995's landed grace covers only BATCH_CREATE_404_GRACE_S after create, not a 404 hours into polling.
Why it matters: The Anthropic batch API demonstrably returns transient 404s for live batches; a single-shot read is too fragile for a multi-hour poll loop.
Confidence (emitter): medium

Primary target: src/explore_persona_space/eval/batch_judge.py
Grep the workflow surface for the pattern before editing and update every hit; list them in the plan.

scripts/workflow_lint.py default run passes; ruff on touched files passes.
This task was auto-filed by the /daily three-route classifier (route 2 — behavior/logic change, independent review).

Activity