Monotonic completion index, starting at 1 (independent of sortSeq).
Dispatch→applied round-trip latency, ms.
ms inside the backend sort_splats_by_depth call, or null if unknown.
Worker-side overhead around the kernel (workerMs − kernelMs), or null.
Round-trip minus worker time (Comlink + clone + queueing), or null.
Element count in the applied ordering (points / segments / splats / triangles), or null if unknown.
One depth-sort completion in the dedicated MONOTONIC event stream (issue #711). A completion is recorded once per APPLIED ordering by the depth-sort coordinator. Unlike the aggregate 'Depth Sort' profiler root, this stream retains an individual event for every application, so several completions between two consumer polls are sampled separately. Consumed by the orbit perf bench, which drains every completion since its previous poll.