Responsive captures
360 / 768 / 1440
Meta
Llama 4 Maverick
Frozen one-attempt benchmark evidence. The live build is reproduced from the untouched model worktree using deployment-path adaptations recorded in its manifest.
Total39.25/100
Raw39.25/100
Automated26/60
Visual8.25/20
Code0/10
GateFail
Interaction evidence
Run facts
- Status
- completed
- Duration
- 0.2 minutes
- Cost
- $0.0034 USD
- Agent steps
- 2
- Candidate
- C-1156
Failed automated checks
- F02A[F02A|6][GATE] filtering controls cards, URL state and direct loadinghard gate
- F02B[F02B|2] browser back and forward restore filter state2 points
- F03[F03|4][GATE] a keyboard-operable service button reveals supplied detailhard gate
- F04[F04|4][GATE] dialog traps focus, closes with Escape and returns focushard gate
- F05[F05|4][GATE] the native-select form has errors and deterministic successhard gate
- R02[R02|5] long copy is not clamped or clipped and missing media has a fallback5 points
- R04[R04|5] mobile navigation is present and keyboard usable5 points
- A02[A02|4][GATE] whole page remains accessible while the modal is openhard gate