1. What cost the most: about 8 turns of tracing. The brief placed defect 1 in the emitter (emit_function.nika ~564). In fact the impl’s keeps column comes from crates/nikaia/src/contracts/keeps.rs::infer. That same column drives the checker’s kept_args fact (the &&*known at the direct call). After that came about 3 probe turns on an unrelated false acceptance (let f = h.f through a ref) before I found a test shape that avoided it.
2. What the brief could have handed me:
   • Where keeps is computed: keeps.rs infer, with an Item::Trait arm and impl methods through uses_of.
   • That call sites are decided by checker facts (kept_args in check_core_args.nika), not only by the emitter.
   • That method calls carry the resolved key in walk.method_callee; a bare callee is the method name, which matches several candidates.
   • That takes_a_handle’s duplicate (.clone() after the argument) already exists as the pattern to copy.
3. Checks:
   • Ran: the lower-std/build cycle twice, then nikaia run probes (mine, the trait probe, the free-function control rows), the guard set, tests/compiler at both settings, clippy and the pre-push check (all green).
   • Wasted: the first build cycle, because it keyed on callee and missed method calls; two probe runs on the field-move issue.
   • Running the gate in the background with a 600 s limit was risky: tests/compiler alone can take about 10 minutes.
   • The push was rejected and I did not rebase, per the rules.
4. One change: before writing a brief that confines a fix to tools/emit_*.nika, the architect should check where the deciding fact comes from: the contract column (keeps.rs) and the checker fact (kept_args). Here that check would have produced a brief allowing a ~10-line union in infer, so the whole issue could have closed in one run instead of partial. For builder.md: a trap line saying “fn-parameter run/keep is the contract’s keeps (contracts/keeps.rs), read by both the checker and the emitter.”

This site is open source. Improve this page.