Verified independent pairs retained for review and distributed point application.
Geometric containment with a one-original-reference-pixel boundary tolerance.
Entire selected plan boundary in reference pixel coordinates.
Stable opaque identifier; do not use credential-bearing URLs.
Spatially deduplicated accepted correspondence count.
Independent accepted count divided by the footprint's tentative pair count; independent of search-area size.
Median forward residual in original reference pixels.
Selected similarity, affine or projective model in pixel space.
Footprint clipped to the search rectangle, in reference pixels.
Clipped plan area divided by full plan area, calculated in query coordinates.
95th-percentile forward residual in original reference pixels.
One-based position in descending evidence quality.
Fraction of full query region with available reference data; alpha is grid sampled.
Versioned ranking measure, not a correctness probability.
Largest footprint disagreement between spatial split fits, in reference pixels.
Fraction of sampled query edges supported by local reference edges.
Query evidence hull area divided by observable query overlap area.
Descriptor pairs of the job whose reference point lies inside this footprint.
Row-major homogeneous matrix mapping query edges to reference edges.
Explanations of ambiguity, extrapolation or missing reference data.
Validated placement with independent, spatially distributed evidence. The selected query region defines its footprint; partial candidates can rank above complete ones. Geometry, data availability and feature support are separate measurements. Small residuals alone do not imply acceptance or survey accuracy.