Chunks decoding to fewer bytes stay on the main thread: their blosc (+ delta)
costs less than the round trip. Measured (Chrome, M4 Max, 6 s playback of the
tp50 gate store — ~49 chunks/tick, median 208 B compressed): offloading
every chunk RAISED main-thread busy from ~20% to ~28%, while the codecs it
removed had cost ~3 ms in total. The Luxar writer targets 64 KB chunks, so
real payload chunks clear this comfortably.
Chunks decoding to fewer bytes stay on the main thread: their blosc (+ delta) costs less than the round trip. Measured (Chrome, M4 Max, 6 s playback of the
tp50gate store — ~49 chunks/tick, median 208 B compressed): offloading every chunk RAISED main-thread busy from ~20% to ~28%, while the codecs it removed had cost ~3 ms in total. The Luxar writer targets 64 KB chunks, so real payload chunks clear this comfortably.