feat(frontend): overlaid agent stat polygon in the Eval section

[pfbyjy]

Stack 4/4 — based on #1100 (merge #1098 → #1099 → #1100 first; this PR then retargets and needs a rebase/update-branch).

Adds a Stat polygon card: every shown agent's polygon overlaid on one radar, so agent "personalities" read as silhouettes on shared spokes — a wide efficiency pentagon with a low score vertex is a budget generalist, a tall narrow score spike is an expensive specialist.

• Axes: score is absolute pass@1; cheap / light / fast / lean are cost, tokens, time, and steps per trial as efficiency relative to the best shown agent (min/value, outer edge = best). Axes nobody reports are dropped from all polygons rather than rendering misleading zeros.
• Joins the row's shared machinery: common legend, hide toggles (hiding an outlier re-normalizes the axes over the rest, same philosophy as the Pareto frontier), and hover cross-highlight — hovering an agent anywhere bolds its polygon and fades the others, which keeps the overlay readable at many agents.
• The axis tooltip reports each agent's raw value (actual dollars/tokens/time, with the ~/* estimate marks), so the normalized picture never hides the real numbers.

Same feature gate as #1099 (Eval toggle, allowlisted accounts only). With this PR the stack's combined tree is byte-identical to the original spike branch (verified by diff), rebased onto current staging.

Verified with tsc --noEmit, ESLint, next build, and rendered screenshots including the hover cross-highlight.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BzsxcETgi6UjLtCnety4sQ

---

Generated by Claude Code

---

Note

Low Risk
Frontend-only experimental analytics behind the existing Eval allowlist; no API or auth changes.

Overview
Adds a Stat polygon card to the experiment page’s allowlisted Eval graphs row: one radar chart with every visible agent’s polygon overlaid so cohort “shapes” are easy to compare.

score is absolute pass@1 (same estimator as the leaderboard); cheap / light / fast / lean plot cost, tokens, time, and steps as efficiency vs the best shown agent (min/value, outer edge = best). Spokes with no reported metric are omitted for all agents; hiding an agent in the shared legend re-normalizes the efficiency axes, matching the Pareto card behavior. Tooltips show raw values (including cost ~/* estimate marks), not only normalized radii.

The card reuses the row’s AgentLegend, hide toggles, and hover cross-highlight (bold polygon + dim others). It is lazy-loaded next to the Pareto frontier and solve grid when showEvalGraphs is on. CHANGELOG documents the stat polygon alongside the other Eval cards.

^{Reviewed by Cursor Bugbot for commit c9d278b. Bugbot is set up for automated code reviews on this repo. Configure here.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
oddish-app	Ready	Preview	Aug 6, 2026 6:58pm

[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 c9d278b. Configure here.}

[cursor]

Sub-cent costs show as free

Medium Severity

The cheap-axis tooltip formats per-trial mean cost with formatCostUsd, which floors values under half a cent to $0.00. That hides the real dollars the normalized polygon is based on, and can make two differently placed agents both read as free. The adjacent Pareto card already formats the same means with four decimals for this case.

Additional Locations (1)

• frontend/src/components/agent-stat-radar.tsx#L113-L118

 

^{Reviewed by Cursor Bugbot for commit c9d278b. Configure here.}

[cursor]

Risk: medium. Left a non-blocking comment — Cursor Bugbot reported an unresolved medium-severity finding, so this was not auto-approved. Human review is already requested; no additional reviewers assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}