[Eval] Report failures and unknowns in results table

[Birfy]

Summary

Implement a bounded reporting slice for issue #13 in the existing evaluation table.

• Keep every result as an attempt and count errors separately.
• Preserve the legacy score, scored-trial median time, steps, decisions, calls, and scored-trial cost metrics.
• Add all-attempt time and cost aggregates with known-value denominators and unknown counts; include recorded data from failures.
• Score only non-error trials with explicit rewards, preserve real zero, and show missing rewards as unknown.
• Update the eval guide to explain semantics and denominators.

Verification

• Updated onto latest main 90246d6b317964674412c43d0dee966a01ca515c; exact PR head is 91e1d54027b0c1db2772accdf0a249b37777534e.
• Local ruff check, ruff format --check, and git diff --check passed in the project’s installed Python 3.13 environment.
• UV_CACHE_DIR=/tmp/system1-agents-uv-cache uv run pytest tests/test_evals_table.py -q: 8 passed.
• UV_CACHE_DIR=/tmp/system1-agents-uv-cache uv run pytest tests/test_replay.py -q: 17 passed, 6 skipped (existing optional tests).
• Ran evals.table.main() without mocks over a temporary write_job fixture. The CLI printed a row preserving a genuine score, elapsed time, and cost of zero.
• Added replay regression coverage for explicit zero elapsed time and missing elapsed time/timestamps. The replay API retains its numeric zero fallback; the table separately preserves missing duration as unknown.
• Exact-head Actions run 37948219165 for 91e1d54027b0c1db2772accdf0a249b37777534e: full, core 3.11, core 3.13, and Windows core 3.11 jobs passed; browser job skipped.
• No model calls or evaluation runs were made.

Related: #13. This is a reporting foundation only; it does not implement the browser/desktop task suite or claim its acceptance criteria are complete.

---

Note

^{Cursor Bugbot is generating a summary for commit 91e1d54. Configure here.}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-09T15:06:32.635675Z	91e1d54	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 8df79c6. Configure here.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 8df79c619b

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".