What a SplitMergeClaim is actually resting on. Three named evidence
profiles rather than a numeric score, because the difference between them is
a difference in KIND of evidence, and averaging kinds into one number hides
which one was missing:
verified — the pieces sit inside the whole's own extent, every
participant carried a EntityFingerprint.volume, and the volumes sum
to the whole's within DiffOptions.splitVolumeTolerance. The only
profile where the geometry was measured rather than bounded.
extent — the pieces sit inside the whole's extent and their boxes cover
it on all three axes, but a volume was MISSING somewhere (and never
refuted). The weakest profile: an axis-aligned box cannot tell a genuine
split from a redesign in place, nor a filled volume from a perimeter
enclosing an empty middle.
displaced — the pieces moved out of the whole's old extent, so
containment says nothing. Accepted only on complete volumes within
tolerance PLUS congruent sorted extents PLUS a pairing that is unique in
both directions. Lower confidence than verified because position, the
strongest single piece of evidence, is exactly what is absent.
What a SplitMergeClaim is actually resting on. Three named evidence profiles rather than a numeric score, because the difference between them is a difference in KIND of evidence, and averaging kinds into one number hides which one was missing:
verified— the pieces sit inside the whole's own extent, every participant carried a EntityFingerprint.volume, and the volumes sum to the whole's within DiffOptions.splitVolumeTolerance. The only profile where the geometry was measured rather than bounded.extent— the pieces sit inside the whole's extent and their boxes cover it on all three axes, but a volume was MISSING somewhere (and never refuted). The weakest profile: an axis-aligned box cannot tell a genuine split from a redesign in place, nor a filled volume from a perimeter enclosing an empty middle.displaced— the pieces moved out of the whole's old extent, so containment says nothing. Accepted only on complete volumes within tolerance PLUS congruent sorted extents PLUS a pairing that is unique in both directions. Lower confidence thanverifiedbecause position, the strongest single piece of evidence, is exactly what is absent.