to_string_of_a_view_fills_a_string: a loop variable over a ref Vec[String] has owned kinds but is a view. I only found out from the 10-minute tests/compiler run, so that run, clippy, the it set and the pre-push check all ran twice.docs/adr and docs/history before I found docs/specification/adr/adr-282.md.Kinds and TierLocal structs.Binary and Cast give owned_kinds(), as do the scalar payloads in bind_a_payload).for loop binds its locals.The brief could have given the ADR path and said “number values are owned kinds already, the fault is viewed() in the to_string rule”. It had that rule’s line but sent me to “where the local’s kinds are formed”, which was not where the fault was.
lower-std fixpoint, tests/compiler at both settings, the it guard set with text_tiers:: and views_kept::, clippy and the pre-push check.it and pre-push on the wrong fix. Also an extra fixpoint cycle, one nikaia run on a test file inside the project (refused), and a cat that lost $S because I backgrounded an S=… && chain, the trap builder-facts already names.to_string_of_a_view_fills_a_string, which it mentioned only as a note). builder.md should say to copy those tests into the scratchpad and nikaia run them right after the first build, beside the new probes. That takes seconds, where waiting for the full tests/compiler run took 10 minutes.