Union-box pair tests performed after pruning, the sweep's own cost.
Bodies whose extent resolved and were therefore compared.
Every emitted intersecting proxy pair, larger fraction first, then larger shared volume, otherwise sweep order under finite comparable grading metrics. NaN grading does not establish that numeric ordering.
Bodies whose extent the record does not resolve, so nothing compared them; elements in record order, then populations.
What one whole-building overlap sweep measured, and what it cost.
comparedcounts active union-box candidates after X pruning. It excludes shared batch staging/tessellation, body resolution, sorting, active-list prune/splice work, part-pair volume folds and result sorting; it is not total operation count or elapsed cost. Current sweeps share one element part table and world staging rather than restaging per body. Population lookup can still be quadratic and explicit populations add stored-transform work.