1. What cost the most: the probe’s variant case. I wrote it with string literals, so it failed for a reason unrelated to #1127. Isolating that took four nikaia run probes (p2 to p5) and a re-check with built strings. The fix itself took one build cycle.

2. What I looked up that the brief could have handed me: only small things. The facts set_positions is keyed by the arm’s body, and nothing in Rust builds CheckFacts, so a new field touches only check_facts.nika. The checker leaves or-pattern names untyped (pattern_parts), so only the lowering needs the per-alternative sets. The brief’s two suspected locations were both right, which saved most of the search.

3. Checks:
   • Useful: the old-binary probe up front, the build cycle and fixpoint, the scratch project with the new test plus the 12 related narrowing tests under both settings, the fails-before run, clippy, pre-push-check and badges.py.
   • Wasted: one probe whose .to_upper() does not exist.
   • Gaps I should flag:
     • I ran clippy with --profile fast, not plain cargo clippy --all-targets -- -D warnings.
     • I did not run rustup check.
     • My grep for related it tests was cut at head -10, so it tests on #886/#902 may exist that I never ran.
     • I took the disk figures after committing, so they are only in my answer, not in the report.
4. One change that would save the most: have builder.md tell builders to build test-probe strings from a call (.to_uppercase(), f"{i}"), not from literals alone. A String that is only ever filled from literals can be lowered as &'static str, which hides or mimics other defects. The Traps view-checker entry makes the same point for views; a general line like that would have saved my four detour probes.

This site is open source. Improve this page.