1. What cost the most: the named-variant shape. The brief named “tuple and named variant patterns” but gave no probe for the named one. It still refused after my first build, and only then did I find that declared_of skips named variants entirely. That cost a second build cycle of about 3 turns. The 10-minute tests/compiler run was the longest wait, but it was necessary. Writing the for probe with a bare => for x in .. arm also cost one turn (for is a reserved word there).
2. What I had to look up:
   • How #1063 (ac156203) made for elements work: TierLocal.of through walked_struct/element_struct. The brief could have named TierLocal.of and struct_of_local as the mechanism to copy.
   • That TextDeclared.variants holds only tuple variants.
   • Whether bind_static_payloads runs for every arm (it does).

   All were quick greps; the brief’s Locations were accurate.

3. Checks:
   • Run: the probes on the old binary, then the new one; the unfixed suite binary for the before-fix check; one lower-std/build cycle per change; tests/compiler --both-settings; the packages/yaml tests; fmt::every_nika_file; cargo fmt; clippy; pre-push-check; and cargo test -p nikaia-std --lib text_tiers::.
   • Wasted: the unit-test run, which matched 0 tests because there are no lib tests for a .nika module. Probably also the separate nikaia fmt diff pass, which caught one trailing comma.
   • Skipped: clippy 1.99.1, because it isn’t installed and needs a cold build.
4. One change: a brief that names a pattern shape should carry a probe for each shape it names. Here that means the named-variant probe, plus a note on which of them go through declared positions (TextDeclared). Builder.md could also say “probe every shape the brief names on the old binary first”, which it nearly does already, but a ready-made probe per shape would have saved the second cycle.

This site is open source. Improve this page.