1. What cost the most. Mostly waiting on the tests/compiler --both-settings run plus the yaml package tests, which were one background call. The only wasted work was a probe round. My first two negative probes (neg, neg2) passed on the old binary too, so they proved nothing and I had to write four more. One reason was that I guessed wrong which refusal applies.
2. What the brief could have handed me.
   • The brief’s own example for step 2 (“a view of a by-value parameter, NK2302”) does not hold. On the old binary, a by-value String viewed into a pushed element is accepted. With two ref parameters the refusal is NK2106, not NK2302.
   • A negative probe the architect had checked as refused today would have saved that round.
   • The brief’s Locations were exact, so I had nothing to search for.
3. Checks.
   • Useful: the old-binary probe matrix, one build cycle that needed no second fixpoint build, the new-binary probes, tests/compiler --both-settings, packages/yaml, fmt::every_nika_file, and pre-push-check.
   • Wasted: the first pre-push-check, piped to tail, printed nothing, so I ran it again to get the exit code. Also the two negative probes that were never refused.
   • Skipped on purpose: clippy and cargo fmt, because no Rust changed.
4. One change. In the brief template, give every “stays refused” step a probe that fails on today’s binary, with the code it fails with. Or have builder.md say: “a negative probe must be refused on the unfixed binary first, or it proves nothing.”

Also, a correction to my report: I wrote Gate: green for the checks I ran, but the full suite is the driver’s. The disk figures (3.7 GB cache, 7.5 GB free) are only in my final message, not in the report’s lines.

This site is open source. Improve this page.