feat(metrics): Introduce VerdictBadge component

[kaapi-agent-bot]

Issue

Closes #270

Summary

• Introduce VerdictBadge component to represent judge-metric scores.
  • Good: 4–5
  • Needs Refinement: 2–3
  • Needs Improvement: 0–1
• Badge rendered in:
  • MetricsOverview (run summary scores)
  • ScoreDisplay (evaluation run list cards)
  • DetailedResultsTable and GroupedResultsTable (per-trace metric cells)
  • Avg Correctness column in CategoryMetricsTable
• No badge for cosine-similarity scores or unscoreable entries.
• Verdict logic handled in app/lib/utils/evaluation.ts, using existing color tokens.

Checklist

Before submitting a pull request, please ensure that you mark these task.

• Ran npm run dev and npm run build in the repository root and test.
• If you've fixed a bug or added code that is tested

Original PR description

Summary

Judge-metric scores were shown as raw numbers with no indication of whether they were good, mediocre, or poor, forcing readers to remember the cutoffs. This adds a VerdictBadge component that bands a numeric score into a traffic-light verdict:

• Good: 4–5
• Needs Refinement: 2–3
• Needs Improvement: 0–1

The badge is rendered next to:

• Each run-level metric card in MetricsOverview (the run's summary scores)
• Each score chip in ScoreDisplay (used on the evaluation run list cards)
• Each per-trace metric cell in DetailedResultsTable and GroupedResultsTable
• The Avg Correctness column in CategoryMetricsTable

Cosine-similarity scores (detected by name match /cosine/i, and the dedicated avg_cosine column) and categorical / N/A / unscoreable entries render no badge, per the issue.

The verdict logic (getVerdictBand) and labels live in app/lib/utils/evaluation.ts, reusing the existing status-success / status-warning / status-error color tokens from globals.css rather than introducing new colors.

Assumptions

• There's no backend verdict field on scores today (confirmed by reading app/lib/types/evaluation.ts) — the band is computed client-side purely from the numeric value using the cutoffs given in the issue.
• Non-cosine numeric scores (TraceScore/SummaryScore with data_type: "NUMERIC") are assumed to be on the 0–5 judge-metric scale described in the issue. If a future metric uses a different numeric scale, it will need to be excluded from banding the same way cosine is.
• There's no single backend-provided "overall run-level verdict" field, so the run-level surfaces (MetricsOverview, ScoreDisplay) get a badge per metric rather than one combined badge — this covers "run-level" per the parent issue (Evaluation 2.0: Transition to v2 API #265) until a dedicated overall-verdict field exists.
• Cosine detection is done via a case-insensitive "cosine" substring match on the score name, since there's no structural flag distinguishing cosine from judge metrics in TraceScore/SummaryScore.

Closes #270

[coderabbitai]

Important

Review skipped

Auto reviews are limited based on label configuration.

🏷️ Required labels (at least one) (1)

• ready-for-review

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Advanced

Run ID: 80021c1d-c9a9-46ee-838b-cbde87ccb32e

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}