From 2 phone conversations with Stephen Osborne, 2/2/99:

For each path:

T == t_1 + t_2
T_i == 1/N sum ( T )

(An average total transit time in counts is generated as the sum of the
times in each direction, divided by the number of samples-per-output.)

delta_T == T - T_(i-1) >? T_threshold

(The difference between each 200 Hz T in the current sample is checked as 
being within the spike parameter of the previous output sample's average T.)

Note that:
1. Because the sum in both directions is tested, a real change in temperature
is weighted twice as much as an instrument spike.
2. If more samples are averaged (10 rather than 20 samples per second output),
a real temperature ramp will induce a larger delta_t, and thus be more
sensitive to false spiking.
3. For a 15cm path sonic, a cycle slip = 1.25 C.