ops-dashboard · pi-gpt-5.5
pi · started 2026-10-08T14:50:37.302Z · horizon 420s · judge ai (openai:gpt-5.5), 5 frames scored
Area under curve
0.806
headline, 0-1
Tab title/icon
12.5s
browser chrome, not the page
First render
12.5s
anything on screen
First reviewable
27.8s
worth feedback
Final score
0.86
no regression
Correctness over time
Where the time went
98%
Model thinking 43.4sTool round trips 0.1sUnaccounted 0.7s
| Phase | Time | Share | Side |
|---|---|---|---|
| Model thinking | 43.4s | 98.2% | agent |
| Tool round trips | 0.1s | 0.2% | 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 | 1.6% | unknown |
| Wall clock (cold start) | 44.3s | 100% |
Agent or toolchain
98%
Agent (model + tool round trips) 43.5s
Toolchain (install + build + boot + paint) 0.0s
Unaccounted 0.7s
Accounted for: 98.4% 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. void: check already passed | 4.2s | -- | 5.3s | 1 (edit) | -- | -- |
| Add an eleventh depot row for Porto: 6,480 shipments, 93.1% on-time, $19.95 cost per shipment. | 4.3s | 4.3s | 5.5s | 1 (edit) | -1.2s | -- |
Judged frames
Only the heading/subtitle render; all required dashboard content is missing from view.
Polished header and KPI row are visible, but table, bars, and footer are absent from view.
Strong dashboard visible, but missing visible top-five bar section and footer.
Strong designed dashboard with accurate main content, but footer absent and volume bars not proportional.
Strong dashboard matching nearly all requirements, but volume bars are not visibly proportional.
Caveats
- Iteration "header-blue" is void: its check already passed before the prompt was sent, so it cannot measure this edit. Fix the check in the brief.
- 4 poll ticks were skipped because a capture overran the interval.
- 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.