ezmsg.nwb.clockmodel.SHARED_MODEL_JITTER_FLOOR_PERIODS

SHARED_MODEL_JITTER_FLOOR_PERIODS = 1.0

a target must also be jittier than this before a sibling’s clock is borrowed.

Serves two purposes. It keeps the ratio test meaningful when the cleanest member is perfect – a stream whose stamps were regenerated as i / fs self-fits to ~1e-14 s (or exactly zero), and a pure ratio test against that either rescues everything or, with a > 0 guard, rescues nothing precisely when the reference is most trustworthy. And it stops a large relative difference between two already-clean streams from triggering a rescue that cannot help: sub-sample jitter is not corruption, whatever its ratio to a cleaner sibling.

One period separates the observed populations with room to spare: clean recordings self-fit to 0.3-0.5 periods, while a stream whose stamps were written at the wrong intra-chunk rate lands at ~37.

Type:

Absolute jitter floor for the shared model, in sample periods