[TheGreatAxios]

Summary

• Replace evaluate(name, options) with runEvaluation({ name, data, task, scorers }).
• The spec is an Evaluation. The saved comparable result is an Experiment at format 2. Format 1 is rejected.
• Report.checks and calibrate's public signature stay. The JSONL expect field stays for a later rename.

Verification

• bun run check exited 0 (typecheck, oxlint, oxfmt, 1 unit test, 53 e2e tests).
• Review of 0f62ede found no defects.

Fixes CL-11496

[linear-code]

CL-11496

[corbits-triage]

Triage: Pass

No changes needed from you. A maintainer will review this pull request.

[TheGreatAxios]

Review · Approve

Replaces the eval entrypoint with one runEvaluation call (task, data, scorers).

Findings

No findings.

src/evaluate.ts is gone. In-repo call sites, examples, and the README use runEvaluation. The remaining evaluate import is @corbits/system-one in src/jev.ts, not the old entry. Passing a previous Report as compare is a type error (Report.cases is not an Experiment). bun test e2e src exited 0 (54 pass) and bun run typecheck exited 0.

[TheGreatAxios]

Superseded by the 0.2.0 release: #32

Do not merge this pull request separately. The release branch cherry-picks the feature stack (except the 0.1.3 bump on #22) and sets @corbits/evals to 0.2.0 with @corbits/system-one ^0.4.0.

[TheGreatAxios]

Superseded by the 0.2.0 release: #32