Semantic metadata

Artifacts can name a JSON metadata sidecar. AASG validates it before using a semantic region for crop/redaction or a gesture timeline for a video overlay.

Schema 1: image regions

{
  "schema": 1,
  "media": "home-light.png",
  "regions": {
    "setup-tips": {"x": 42, "y": 560, "width": 996, "height": 480}
  }
}

regions is required. Coordinates are non-negative integer pixels with a top-left origin; width and height must be positive. Schema 1 does not support gestures.

Schema 2: video gestures

{
  "schema": 2,
  "media": "onboarding-light.mp4",
  "coordinate_space": {"width": 1080, "height": 2400, "origin": "top-left"},
  "gestures": [
    {"type": "tap", "at_ms": 700, "cue_lead_ms": 120, "x": 540, "y": 1800},
    {
      "type": "swipe",
      "at_ms": 2420,
      "cue_lead_ms": 120,
      "duration_ms": 450,
      "from": {"x": 900, "y": 1200},
      "to": {"x": 180, "y": 1200}
    }
  ]
}

Schema 2 requires coordinate_space; its width and height are positive and its origin is top-left. Gesture events must be sorted by at_ms, and every coordinate must fit inside that space.

Gesture	Required fields
tap	at_ms, x, y; optional non-negative cue_lead_ms (default 0).
swipe	at_ms, positive duration_ms, from, to; optional cue_lead_ms.
drag	at_ms, at least two points; optional cue_lead_ms. Each point has offset_ms, x, y; first offset is 0 and later offsets strictly increase.

at_ms starts the visual cue. The wrapped test action begins cue_lead_ms later. Rendering rejects unsorted events, out-of-bounds coordinates, media-dimension mismatch, and gestures extending beyond the decoded video.