calor evaluation
Run the experimental type-system research evaluation workflow. This surface is for experiment governance, not the published Calor-versus-C# dashboard.
| Command | Purpose |
|---|---|
calor evaluation registry | Query docs/experiments/registry.json |
calor evaluation registry-validate | Enforce the registry's append-only invariant |
calor evaluation ab | Compare paired baseline and candidate runs and emit a decision memo |
registry requires a query mode and may require a hypothesis, tuple, or user.
registry-validate requires base and head registry files. ab requires
baseline and candidate run files, a primary metric, and a promotion threshold;
it also accepts repeatable no-regression guards, direction, bootstrap resample
count, and random seed.
Use calor evaluation <command> --help for the complete experimental option
surface. For the public language benchmark, see calor benchmark
and the methodology.