Measure and improve
`kusaw metrics --range today` prints finished runs, autonomy, stage times, throughput, and outcome counts for one factory. Outcomes settle over days: a landed change younger than three days is still settling, and the outcomes panel uses a 30-day window.
kusaw metrics --range today prints finished runs, autonomy, stage times, throughput, and outcome counts for one factory. Outcomes settle over days: a landed change younger than three days is still settling, and the outcomes panel uses a 30-day window.
Scorers are an explicit score step in finally. Sampling is fixed from the run id and the scorer name, so a replay is scored the same way. Scorer cost is charged to the finally allowance.
Self-improvement drafts proposals. A person approves each one. Retro is the finally step that drafts them. A proposal may change a scalar in kusaw.yaml, a step's instructions_file, or a file in a declared skill folder. It does not edit repository code. Drafting has its own daily budget.
Benchmarks replay a task and compare variants. A replay is correct when every repository is verified and the held-out checks pass. Gates in a replay are passed automatically and are not counted as a human interaction. kusaw bench report --markdown writes the table for you to keep. kusaw bench report --tasks prints pass@k, pass^k and spread; see Reading reliability. The factory does not switch a default because a variant won.
Routing and reviewer-model overlays stay opt-in until that evidence says otherwise. See recipes.md.
Recipes
These overlays are opt-in. Neither one replaces the default pipeline. The factory does not switch a default on its own.
Benchmarks
Benchmarks replay a captured task against baseline and variant overlays. A replay is correct when the clean-checkout regrade is pass (`grade_clean`). The in-place grade (`grade_in_place`) is recorded next to it. `kusaw bench run` queues those replays (default `--repeat 3`). `kusaw bench report --tasks a,b --variants baseline,cheap` reads reliability from the latest batches. `kusaw bench report <batch> --markdown` writes the comparison table for you to keep.