1. What cost the most: waiting. The full it binary took about 15 minutes (934 s) and tests/compiler --both-settings about 10. My old-vs-new check comparison over examples, packages and tests/{build-time,guide,language} hit the 20-minute background limit. Its output went through grep, so nothing was saved and the whole run was lost.
2. What I looked up that the brief could have handed me:
   • That nikaia check FILE inside tests/compiler checks the whole project, so a probe file has to be copied out first. One wasted call.
   • That nikaia test has no name filter, so “run the 22 assert tests by name” cannot be done literally.
   • That no tests/compiler test sets unproved-claims through a nikaia.toml (a grep).

   The brief’s locations and its decision were exact. The fix took one edit and nothing beyond it.

3. Checks run:
   • Useful: the probe on the old and new binaries, including the refuted, carried-back and proved variants; the old/new compare of the 22 assert tests (quick and conclusive); tests/compiler (484 passed); the filtered it run, which found 3 of the 4 failures in 20 s; the full it run, which found the fourth.
   • Wasted: the project-wide compare (killed, no output); the first proj.sh attempt (check needs a FILE argument). clippy 1.97 passed and 1.99.1 was not run; I judged the lowered change too small to need it.
4. One change that would have saved the most: Step 2’s filter should name prove_through_loops and unsafe_follows_switches. Better still, the brief could say: “a precondition fix changes what is refused, so grep -rln 'assert(' crates/nikaia/tests/it/ and run those modules”. Then the 4 refusals show up in one minute, not after a 15-minute full run. The brief should also say ahead of time what to do with a true unproved call in a Rust test: edit its program, or stop. That would have let me finish done instead of leaving develop red.

This site is open source. Improve this page.