1. Most time: the nikaia test --both-settings --project tests/compiler run, which took over 10 minutes in the background. The second most was the gate plus clippy, about a minute of the gate’s tests plus clippy’s build beside it. The fix itself was 5 lines and took one build cycle. Turns went only to reading, the one cycle and the long runs.
2. Looked up: I read the whole of docs/builder-facts.md (the one read the brief asks for). I also looked up the Pattern::Guard field names in ast.nika (cond, expected) and how tier_expr is called on a match guard. The brief gave the line number for the call but not the variant’s shape. One line in the brief, Pattern::Guard { cond: Expr, expected: String? }, would have saved that grep. No ADR or converter was needed.
3. Checks run:
   • The probe on the old binary (needed: it showed the test failing).
   • One build cycle with two lower-std runs (needed).
   • The probe on the new binary.
   • The six-guard gate with clippy beside it.
   • pre-push-check.
   • The full tests/compiler run at both settings.
   • A separate sequences:: run, to see the guard’s figure. It was wasted: the gate run had already reported it ok.

   The full tests/compiler run is required for check/ changes but arguably oversized for a text_tiers.nika fix. Running only the grammar_* tests plus the new one would have given the same signal much sooner. nikaia test has no name filter, though.

4. One change: let scripts/commit.nika handle a rejected push. Either it does a git pull --rebase itself when the tree allows, or the brief states plainly that a rejected push counts as done. As it is, my run ends “PUSH REJECTED” even though the work is finished, and the report must carry Pushed: rejected. Second choice: state in builder.md that the tests/compiler run is needed only when a change can alter a lowering beyond the touched pass. A fix in one tier pass could then run just the tests that match its name.

This site is open source. Improve this page.