Ctrl Alt Muse

Apprentice turns a demonstration video into a checklist that can check later recordings of the same hands-on procedure. A user uploads a teaching video, reviews the generated steps, and uploads another recording. Apprentice looks for visible evidence of missed steps, incorrect order, wrong-object interactions and unusual timing. Findings link to timestamps and object overlays, with a downloadable audit report. Our design principle is: models propose, pixels ground, code decides. Meta models interpret the demonstration and locate objects; deterministic rules decide whether the visible evidence satisfies each step. When evidence is missing or ambiguous, the result stays unconfirmed. The prototype targets single-camera tabletop procedures, with potential applications in assembly, workshops and procedural training. It includes private accounts, budget-controlled processing, protocol-validation checks and evidence-linked reports. Real-world accuracy has not yet been established on a labelled test set.