Waymark

Waymark helps a person revisit an indoor route through visual landmarks instead of invented GPS coordinates. An operator teaches a sequence of places; Meta Muse Spark Open reads signage and static cues, and Waymark stores reference images in a topological route graph. When revisiting the space, Meta compares the current image with learned references. A deterministic local gate requires the expected checkpoint, all learned signage (excluding fixture watermarks), model support above a threshold, and two distinct image observations before advancing. Wrong-wing or uncertain views hold the location; closing a taught link replans over known links or explicitly reports no available route. The interface shows the evidence, current checkpoint and decision trail, and exports the route. Intended direction: offices, venues and airport interiors where GPS is insufficient. This entry is a working landmark-route prototype, not a tested navigation aid: no SLAM, metric floor plan, obstacle avoidance or turn-by-turn guidance. The demo uses original illustrated scenes and real Meta inference; second-view scenes are crops, not independent physical captures. Mentra glasses were unavailable, so the included USB camera adapter has not been tested with the new workflow. Eleven deterministic checks pass, with actual model results published separately. New during this event: Meta adapter, structured observation prompt, visual reference matching, graph/planner, verification gate, evidence export, UI, replay fixtures, tests and demo. Pre-existing: the author's Mentra Look camera transport module, clearly identified in README. No earlier assistant or game features are presented as new work.