- Biggest cost: I ran the fmt check only at the end. Applying the draft, writing the test, blessing it and fixing the comments all came before the
fmt::every_nika_file scan (about 50 s), which then failed on #1015. Running nikaia fmt on a copy of the draft right after applying it would have shown the blocker in the first five turns. The fast build at pre-flight was needed: the binary was older than 262d5fa6, and I could only tell from the rebuild because commit times are rebase times.
- What I looked up that the brief could have given:
- Whether #1015 was still open. The brief listed the open defects #1038 and #1037 but not #1015, though builder-facts mentions it.
- What fmt does with a hand-broken rule line. It moves the continuation to one tab, which ruled out the only layout fix.
- The pattern of the earlier package commits: they include
nikaia.contracts, nikaia.derived, nikaia.lock and nikaia.proofs along with the source.
- Checks run:
- Useful: the two pre-flight guard tests (via
nikaia run copies), nikaia check, three package test runs, the fmt scan, pre-push-check, and four small fmt probes that isolated the trigger.
- Wasted: one package run that only printed error messages for inspection. The message edits and the final package run after comment-only changes were also moot once fmt blocked.
- Not needed: the #1037 test on its own, since the package tests cover it.
- The change that would have saved the most: add to the brief’s pre-flight, and to builder.md’s step for starting from a draft: “after applying a draft, run
nikaia fmt on a scratch copy of every changed .nika before writing anything”. The architect checked check and test on the applied draft but not fmt. A pre-flight that runs the whole package gate (including fmt) on the bare draft would have stopped this run, or the brief itself, at once.
This site is open source.
Improve this page.
