RunWave screenshot artifacts currently include the overlay grid used for playtester targeting. That grid is useful for the agent while choosing actions, but downstream verifier/judge consumers need clean visual evidence of the game.
This matters for benchmark tasks that use RunWave during verification: the playtester can use grid-marked screenshots to act, but the screenshots passed to a VLM judge should not contain grid lines, labels, or margin markings. Otherwise the judge sees RunWave UI artifacts rather than only the submitted game.
Observed impact:
- Saved screenshots/captures can include the red grid and row/column labels.
- These artifacts may be reused by task verifiers as judge evidence.
- Grid-marked evidence can distort visual scoring, especially for reconstruction/gameplay tasks where screenshots should represent only the submitted game.
The expected behavior is that verifier-facing screenshot artifacts are clean, while grid-marked screenshots remain playtester-only evidence.
RunWave screenshot artifacts currently include the overlay grid used for playtester targeting. That grid is useful for the agent while choosing actions, but downstream verifier/judge consumers need clean visual evidence of the game.
This matters for benchmark tasks that use RunWave during verification: the playtester can use grid-marked screenshots to act, but the screenshots passed to a VLM judge should not contain grid lines, labels, or margin markings. Otherwise the judge sees RunWave UI artifacts rather than only the submitted game.
Observed impact:
The expected behavior is that verifier-facing screenshot artifacts are clean, while grid-marked screenshots remain playtester-only evidence.