Left: the guideline as published, line by line. Right: the rubric compiled from it. Hover a row to see the exact line it comes from.

Every image is built from the rubric: the label is written first, then the region is painted, then it is verified. Higher levels add more hazards, safe look-alikes that should not be flagged, and harder conditions.

The same labeled hazards, grown into walkable worlds. Walk them yourself, or drop a model into the agent environment: it moves, looks and flags hazards through the room the way a clinician on a home visit would.