@hypit/spatial

Pure two-dimensional video geometry. The package owns the final raster viewport, resolved picture-plane geometry, source extents and affine mappings. It owns no layer tree, media, Paint, motion, Timeline, detector, Provider or renderer policy.

The spatial values

• Canvas is the final raster viewport. A <space:Canvas> publishes .canvas for raster consumers and .bounds for ordinary geometry consumers. Its runtime value contains only width and height; the @1 coordinate convention is a protocol rule rather than four repeated constants.
• SpatialPoint, SpatialFrame and SpatialPath are plain resolved values in the program picture plane: top-left origin, x right, y down, square pixels. They carry no Canvas id, parent pointer or construction history.
• IntrinsicExtent is the natural width and height of a source or local picture plane.
• ContentFit is an authored mapping policy.
• SpatialMap2D is an affine map from a local plane into the program picture plane.

Canvas is deliberately not a container. Visual components own clipping, Paint, draw order and motion. A Frame may extend outside the Canvas and may be reused anywhere its numbers mean the same thing.

<space:Canvas id="vertical" width="1080" height="1920"/>
<space:Frame id="safe" within={vertical.bounds}
  left="6%" top="5%" right="94%" bottom="92%"/>
<space:AnchoredFrame id="portrait" within={safe}
  x="100%" y="100%" width="320px" height="320px" anchor="bottom-right"/>

Frame, AnchoredFrame and AspectFrame are three constructors for the same runtime SpatialFrame. Other layout packages can publish that same Type without modifying this package.

An author or project component that already knows an affine relation can publish it directly:

<space:Map id="turned" xx="0" xy="-0.5" yx="0.5" yy="0" tx="920" ty="180"/>

This is the open value below fitting conveniences, not a scene node. It carries no source, Frame, Canvas identity, clipping policy or time.

Content fitting resolves a mapping

resolveContentFit(frame, extent, fit) returns a SpatialMap2D. It does not invent a second FittedContent model. Consumers may apply the map to source-local Points and Paths, and may derive the mapped source bounds when they need a rectangular optimization or diagnostic. That bounding rectangle is not the mapping and cannot replace it after rotation, skew or reflection.

ContentFit.sizing selects the scale before alignment:

| Sizing | Result | | --- | --- | | contain | Preserve aspect and fit both dimensions inside the destination | | cover | Preserve aspect and cover both destination dimensions | | fit-width / fit-height | Preserve aspect and match the named dimension | | native | Keep the source's pixel dimensions | | scale-down | Use contain while limiting scale to at most 1 | | stretch | Match destination width and height independently |

framePoint selects a normalized point in the destination and contentPoint selects the source point placed there. offsetPx adds an explicit displacement. bounded constrains the translation to the available placement range; free preserves the authored translation. Clipping is still the visual consumer's decision.

The Recipe decoder exposes fit, frame-x, frame-y, content-x, content-y, fit-offset-x, fit-offset-y and fit-constraint. Visual Track uses it, and project components may use the same pure functions.

Evidence over time is a peer package

Frame-indexed observations do not belong to static geometry. @hypit/region-evidence relates prepared regions to a Timeline and resolves them into picture-plane Frames. Detection, identity association, interpolation and source-to-picture projection remain explicit preparation or project-package work. This keeps Spatial reusable and prevents a central scene or evidence registry.