Acceptance run 250 (PR 164, 4d42d25c) passed every stage -- the four boots and test-32 (39 programs, 0 failures) -- and was marked failed by its last step, Keep the logs and DoE CSVs (actions/upload-artifact@v3, .gitea/workflows/acceptance.yml:70). Gitea's artifact service timed out finalizing the upload, then answered 500 on four retries: Finalize artifact upload failed: Artifact service responded with 500. Run 252 uploaded the same paths and succeeded, so it is intermittent.
The step uploads all of logs/ and experiments/bare_metal/runs/ -- 2,620 files, every run in the tree's history, not this run's -- which is the likely reason the finalize request timed out. The nightly (nightly.yml:91) and the release do the same. Those logs are committed to git already; the artifact is a convenience copy, yet a failed upload turns a passing gate red.
Done when
The uploads keep only what this run wrote (as scripts/nightly.sh publish does, -newer build/nightly/.started; or logs/<TS>/ of this run)
Decided: a failed artifact upload fails the gate, or is a warning that cannot turn a passing acceptance red -- and the decision written in the workflows' comments
Seen: an acceptance run whose upload is a handful of files, not thousands
🤖 Filed with Claude Code -- found checking PR 164's acceptance runs (Captain Bob, 2026-11-12: "file the artifact issue").
### Found
Acceptance run 250 (PR 164, `4d42d25c`) passed every stage -- the four boots and test-32 (39 programs, 0 failures) -- and was marked failed by its last step, **Keep the logs and DoE CSVs** (`actions/upload-artifact@v3`, `.gitea/workflows/acceptance.yml:70`). Gitea's artifact service timed out finalizing the upload, then answered 500 on four retries: `Finalize artifact upload failed: Artifact service responded with 500`. Run 252 uploaded the same paths and succeeded, so it is intermittent.
The step uploads all of `logs/` and `experiments/bare_metal/runs/` -- 2,620 files, every run in the tree's history, not this run's -- which is the likely reason the finalize request timed out. The nightly (`nightly.yml:91`) and the release do the same. Those logs are committed to git already; the artifact is a convenience copy, yet a failed upload turns a passing gate red.
### Done when
- [ ] The uploads keep only what this run wrote (as `scripts/nightly.sh publish` does, `-newer build/nightly/.started`; or `logs/<TS>/` of this run)
- [ ] Decided: a failed artifact upload fails the gate, or is a warning that cannot turn a passing acceptance red -- and the decision written in the workflows' comments
- [ ] Seen: an acceptance run whose upload is a handful of files, not thousands
🤖 Filed with [Claude Code](https://claude.com/claude-code) -- found checking PR 164's acceptance runs (Captain Bob, 2026-11-12: "file the artifact issue").
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Found
Acceptance run 250 (PR 164,
4d42d25c) passed every stage -- the four boots and test-32 (39 programs, 0 failures) -- and was marked failed by its last step, Keep the logs and DoE CSVs (actions/upload-artifact@v3,.gitea/workflows/acceptance.yml:70). Gitea's artifact service timed out finalizing the upload, then answered 500 on four retries:Finalize artifact upload failed: Artifact service responded with 500. Run 252 uploaded the same paths and succeeded, so it is intermittent.The step uploads all of
logs/andexperiments/bare_metal/runs/-- 2,620 files, every run in the tree's history, not this run's -- which is the likely reason the finalize request timed out. The nightly (nightly.yml:91) and the release do the same. Those logs are committed to git already; the artifact is a convenience copy, yet a failed upload turns a passing gate red.Done when
scripts/nightly.sh publishdoes,-newer build/nightly/.started; orlogs/<TS>/of this run)🤖 Filed with Claude Code -- found checking PR 164's acceptance runs (Captain Bob, 2026-11-12: "file the artifact issue").