All evidence records

reference · implementation

What Jev needs from a computer-use adapter

Supply explicit accessible controls and shared state.

TypeSafe vendor documentation; Chrome DevTools and Apple platform APIs. System One mapping is our interpretation.

Evidence confidence

moderate. The input contract is documented by its provider. The benefit of a particular adapter still needs a controlled task comparison.

What was observed

Turn a screen observation into a bounded next-action question.

Review the official state, fan-out and Jev 1.13 limitation documentation alongside browser and Mac accessibility APIs.

Baseline

No comparative experiment in this reference record.

Finding

Jev currently accepts text. Questions in a batch evaluate independently. A computer-use adapter must supply the relevant labels, values and relationships to every question, with exact checks implemented in code.

What the result does not establish

Documentation describes the model contract and known weaknesses, not reliable task completion. Accessibility coverage depends on the application. Page content can influence the model despite being labeled untrusted.

What we would test in System One

Our inference: bind computed AX names to observed nodes, include group and field state, filter irrelevant text, keep exact comparisons outside the model and retain a review option.

This recommendation is our interpretation of the study. Related research does not establish the quality of every Engine recipe.

Try a related workflow

Primary sources

Read this record in System One Bench. Source commits are pinned where available. Review dates describe our inspection, not the original run date.

Metric definitions and review method · Submit a correction or new result