Agents in this repository check their changes with platform/ci/run_workflow_locally.py --for-changes, which runs the GitHub Actions workflows a diff would trigger. To keep a long sweep from hanging, each workflow gets a 300-second budget. The Portfolio Pages workflow's "Test the site builder" step takes longer than that alone.
When the budget ran out, the step that was cut off and every step after it were marked FAILED. The later ones showed (0.0s) because they never started. On 2026-11-11, three run records reported losing time to this. Each time the output looked like a real defect until someone read the TIMEOUT: reason at the end of the line.
What changed
- A step the budget stops is now
UNRUNwith the reasonTIMEOUT: workflow budget of 300s spent .... The report then prints the command that reruns that workflow by name, which has no budget. - The sweep's last line counts these workflows apart from failures:
3/3 selected workflow(s) passed; 1 ran out of budget with no verdict.The exit code is still non-zero, so the sweep can't pass just because the budget ran out.
The second fix is smaller. run_venture_tests.py --changed diffs against the merge base with origin/main. In a shallow cloud clone there often is no merge base, and the tool used to crash with a CalledProcessError traceback. It now prints the fetch command that fixes this (git fetch --depth=40 origin main, then --deepen=500 if that's not enough), and the other option: pass the venture slugs by name.
Neither change makes a check more lenient. Both make the result say what actually happened, so a red line points at a real problem.