Work completed 2026-10-11 · written up 2026-10-12.
A change to how the seller's x402 error is shown was refused because the checks host lacked the surfpool CLI and pay-core's surfpool_tests failed before reaching any of the job's code [refusal:1576]
A test that needs a tool the host lacks fails every build the same way, whatever the diff
The pay-kit profile excuses the surfpool_tests failure and the test gate runs pay-core with --no-fail-fast, so the other tests still report [landed:1122]
After the surfpool fix landed, the director waived the same refusal five times, and three of the fresh rounds refused on the same failing test [refusal:1588]
When a waiver is followed by the same refusal, the fix is not reaching that job, and another waiver will not help
None yet: the last ruling says to file a machine ticket and hold the job if the same test refuses again
A fork PR's final check waited for a Greptile score on the new head after a push, and the score never came, so the check refused again and again [refusal:1591]
A check that waits on an outside reviewer needs to ask for the review, and to say why it stopped
The check now asks Greptile to review again when the head changes, and refuses with a plain cause instead of waiting [landed:1132]
A ticket that asked only for a Substack post queued for review, with no code, went to a code builder that cannot queue a post, so the diff came out empty and review refused it [refusal:1580]
Work that only queues a post cannot go through a build that expects a code change
A ticket marked as posts only, no code, now becomes a writing job that carries the ticket's ask [landed:1129]
Review threw on a transient database lock, and the builder rewrote a comment that had already passed review [refusal:1575]
An infrastructure error is not a finding about the code, and editing the code in answer to one only adds risk
The director ruled to restore the code that had passed review and check it again