chore(evals): standardize model judge config

[aryeko]

Summary

• add the standard manual pointwise model-judge config alongside the deterministic default config
• add standard eval:judge:* helper scripts and pin eval-kit to v0.1.3
• update eval docs so users validate/run pointwise judging through evals/eval-kit.model-judge.config.json instead of editing the default config

Verification

• pnpm install
• pnpm format
• pnpm eval:doctor
• pnpm eval:list
• pnpm eval:validate-fixtures
• pnpm eval:judge:doctor
• pnpm eval:judge:list
• pnpm eval:judge:validate-fixtures
• pnpm check
• git diff --check

Did not run eval:judge:coverage or provider/model calls.

[aryeko]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ca33058465

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Omit the nonexistent generate run from reports

When following this new no-generation flow, the user is told to judge an existing candidate and not enable generation, but the report command below still passes --generate <run-id>-generate; compileReport then unconditionally reads that supplied generate run's report.md (evals/adapter.mjs:355-358), so the documented flow fails unless the user happened to create an unrelated generation bundle. Please either omit --generate from this manual pointwise-only report command or add a matching generation step/config.

Useful? React with 👍 / 👎.