Techniques¶

A reusable building block that studies share.


Beat matching	Pair beats seen by two signals one to one, using a delay window around the known level rather than nearest neighbours, which flip between beats when the lag is near half a beat.
Coincident jumps	Look for stamp jumps that appear in several independently clocked tracks at the same moment. A step of the common clock moves all of them; a device's own correction moves one.
Cumulative duration curves	Plot the number of gaps or segments longer than x against x on log–log axes, with segments cut by the start or end of a recording treated as censored; cliffs and steps mark mechanisms with a duration of their own.
Junction scores	Score the step across a record junction against a linear extrapolation of the samples before it, scaled by the signal's ordinary sample-to-sample steps.
Null controls	Repeat each test where the effect cannot be: at random times, at the same interval later, sign-flipped, or between two channels of one device. The effect is real only where it beats its control.
Phase-step detection	Fit a regular train before and after a candidate break and test for a step in its phase, accepting only steps without an intervening beat that could explain them.
Predict, then test	Turn what a source says into a prediction about the data (a step every 4,295 s, a skew between two channels, a sample lost every 35 minutes) and look for it, with a control; a source that is not borne out is set aside.
Remainder test	Fold each gap modulo the signal's message duration: whole lost messages pile up at zero, while clock corrections and partial losses spread over the whole interval.
Sawtooth fit	Model a clock as a straight drift that is pulled back at correction events, and fit the rate from whole teeth (the correction divided by the time since the previous one).
Template beat timing	Time each beat by fitting a template of the case's own beats, for sub-millisecond precision that does not depend on one fiducial point.
Units from drift	Infer physical units from their drift rates: fit Gaussian mixtures with a shared per-unit width and let a model-selection score choose how many units there are.
Windowed cross-correlation	Cross-correlate two signals' envelopes in sliding windows (for example 10 minutes, stepped by 5) to follow an offset that changes over a recording, keeping only windows whose peak is clear.