← Workbench

Synthetic mechanics evaluation

These are self-authored fictional cases, labeled synthetic_unverified. Passing them does not establish hiring quality, fairness, model performance or production readiness. No model runs here.

Browser replay

Running synthetic cases…

CaseExpectedObservedStatus

What is checked?

Evidence extraction and score checks in 3 golden and 2 regression fixtures. CI also compares 36 Python/browser combinations and tests the human decision gate. Golden labels have not been externally validated.

Failure cases include a self-reported skill without supporting text and a profile without any evidence. All scores stop at HUMAN_REVIEW; a human must supply reviewer, action and reason.