ops-dashboard · pi-gemini-3.8-flash
pi · started 2026-10-08T14:21:44.861Z · horizon 420s · judge ai (openai:gpt-5.5), 4 frames scored
Area under curve
0.895
headline, 0-1
Tab title/icon
44.0s
browser chrome, not the page
First render
44.0s
anything on screen
First reviewable
44.0s
worth feedback
Final score
1.00
no regression
Correctness over time
Where the time went
97%
Model thinking 3m 10sTool round trips 6.0sUnaccounted 0.7s
| Phase | Time | Share | Side |
|---|---|---|---|
| Model thinking | 3m 10s | 96.6% | agent |
| Tool round trips | 6.0s | 3.1% | agent |
| Dependency install | 0.0s | 0.0% | toolchain |
| Build | 0.0s | 0.0% | toolchain |
| Dev server boot | 0.0s | 0.0% | toolchain |
| First paint | 0.0s | 0.0% | toolchain |
| Unaccounted | 0.7s | 0.4% | unknown |
| Wall clock (cold start) | 3m 17s | 100% |
Agent or toolchain
100%
Agent (model + tool round trips) 3m 16s
Toolchain (install + build + boot + paint) 0.0s
Unaccounted 0.7s
Accounted for: 99.6% of wall clock.
- Excludes 15.0s of deliberate idle observation (settle window); the curve still covers it.
Iteration: prompt to visible change
Follow-ups were injected into the same live session.
| Edit | First visible change | Correct change | Agent done | Tool calls | After agent | Broken for |
|---|---|---|---|---|---|---|
| Make the page heading blue. | 13.3s | 13.3s | 21.8s | 2 (bash, edit) | -8.6s | -- |
| Add an eleventh depot row for Porto: 6,480 shipments, 93.1% on-time, $19.95 cost per shipment. | 10.3s | 10.3s | 25.8s | 2 (bash, edit) | -15.5s | -- |
Judged frames
All requested dashboard elements are visible, accurate, aligned, and deliberately styled.
All requested dashboard elements are visible, accurate, and well designed.
All requested dashboard content is visible and professionally designed.
All requested dashboard elements are visible, accurate, aligned, and deliberately designed.
Caveats
- judge: openai:gpt-5.5 does not accept a sampling temperature, so these frames were scored at the provider's default rather than at 0. Two runs of the same page can be scored differently, and that variance is inside the AUC. result.json records the temperature as null rather than as the value that was asked for, so p2p compare will not pool this run with one that was scored at a fixed temperature.