Repository navigation
v0.195.35
·
109 commits
to main
since this release
Measured, 0.195.34: TIMINGS preflight=3s push=9s pr=4s checks=432s merge=7s verify=9s tag=2s release=224s npm=90s total=780s release-clock=242s, shipped from the merge to the published release in 242 s on the first push.
- Fixed, listed first. 0.195.34 deleted the model-cost advisor but left
bench/router-replay/, whose runner imports the deleted hook, sonpm run bench:router-replay:checkcrashed with a missing module. The bench measured only that advisor's tier advice, so it is deleted with its three npm scripts. No released check ran it: it was a final gate of the assurance-differentiation loop profile, which is a development tool. - That profile's required routing gate is now
routing-replay(npm run reproduce:routing, the offline routing replay that already fed the profile's measured routing gate). It exits 0 when every control behaves and 1 when one is broken; both directions were run for this change. The replayable fixtures drop the bench's corpus and baseline, and the assurance baseline inventory citesscripts/routing-replay.mjsas routing evidence in their place. - Two leftover mentions of the advisor are gone: the README's list of what moved to the capability reference, and a comment in the routing economics module.
- The generated engineering-excellence inventory lists one bench directory fewer.