Watch a real automation run — step by step

This is one recorded workflow, compiled once, then replayed three ways. Press play, or drag the slider, to step through the actual screenshots the replayer saved. Watch it heal when the interface changes, and stop when something is wrong.

What you're watching. A nurse's triage task in a demo EMR was recorded once and compiled into a deterministic workflow. Below, the same compiled workflow runs against three versions of the app. For each step the overlay shows how the target was found (the resolution “rung”), whether it had to heal, and whether the expected screen actually appeared afterward (the postcondition). Every frame is a genuine screenshot from the run — nothing here is staged.
after
run frame

Honest by construction. The app is MockMed, a synthetic stand-in for a real EMR — the data is fake, but the record → compile → replay pipeline and every frame are real. The replay is deterministic and model-free: model_calls = 0 on all three runs, $0.00 per run. The theme-drift run heals via the geometry and ocr rungs and writes each fix as a reviewable diff; the surprise-modal run halts rather than report a save that never happened. Across all runs: 33 real steps captured.