1. Time and turns.
   • The Walking struct would not build in a test. I tried five variants, each a lower-std round of up to several minutes, before probing the rule once in a scratch file. That probe showed a struct with ref fields cannot be built from a local or a ref parameter, so the send.rs tests were blocked.
   • The bootstrap loop cost more. The old binary refused Item::Test. The index lowered.rs held another agent’s stale describe line, which stopped the build. text_tiers.nika was being edited by another agent while I worked. I had to hide the test, lower, build, restore and lower again.
   • One check call was left blocked on stdin because of a stray cat >.
2. What the brief could have handed me.
   • Parsed is not Clone, so the file has to be re-parsed with refine_with.
   • The tools module needs the ShownBy* imports before a lowered assert compiles.
   • The seam: lower_one plus the with_tests_as_functions pattern in modules.rs.
   • The bootstrap order for a new lowering feature.
   • Ref-field structs cannot be built in tests. I had already hit the same NK1106/NK2302 refusals when I wrote my first threads test, so I should have checked the first test’s shape before step 1.
3. Checks.
   • Useful: the flip test, cargo test threads_lowered, clippy and pre-push.
   • Wasted: the four failed lower-std rounds on Walking variants, and a cargo check --release that told me little.
4. One change. The brief should have told the builder to probe the first test’s shape (build Walking from locals) in a scratch file before starting step 1. A line in builder.md would also help: when the shared tree holds another agent’s broken .nika or lowered.rs, say up front how to bootstrap around it. Otherwise the builder rediscovers the same constraints mid-run.

This site is open source. Improve this page.