1. What cost the most. About eight turns went on probes that wouldn’t parse:
   • The brief’s sample wrote mut self, which doesn’t parse; it has to be ref mut self.
   • spawn took three tries before I reached spawn fn { … }.
   • nikaia check run inside tests/compiler checks the whole project, not the one file.

   Finding why a parameter differed from a local took another six turns of reading (binding, takes_away, check_function_frame). The answer: a plain String parameter is bound as a view only when it is lent.

2. What the brief could have handed me. Three things:
   • A probe that compiles as written.
   • The fact that a lent String parameter is a view type (check_function_frame, check_function_lends), while spawn makes it owned through keeps.nika.
   • That the kept-lambda entry point is takes_of_a_lambda in check_lambdas.nika, not the NK2101 reporting in check_takes.nika.

   No ADR or converter lookup was needed.

3. Checks. I ran:
   • one build cycle, which reached a fixpoint at once;
   • five probes on the scratch binary;
   • one gate run in the background: tests/compiler at both settings, the guard set and clippy, all green;
   • cargo fmt and the pre-push check.

   Wasted: the failed spawn probes and one extra cargo build to confirm the fixpoint. I didn’t stash the fix to re-run the test on the old compiler; the before-fix evidence is a probe of the same program on the old binary.

4. One change. Probes in a brief should be run on the current binary before the brief goes out, with a note on which pass decides the behaviour (here: the keeps walk versus the checker). That would also have shown up front that the case with no later use is outside the brief’s edit set, and the brief could have said whether to fix keeps.nika too.

This site is open source. Improve this page.