Results for Toyocom model TCO-999, 20MHz (ISD4-2A-4592) order by LBNL, Jan 2002.
The oscillators were measured for at least an hour, with frequency samples taken every five seconds.
The frequency samples were analyzed in blocks of 100 samples, and blocks of 500 samples,  using our standard Allan Variance code.
 
 Sample Total samples in data file Best Allan Variance for blocks of 100 samples Worst Allan Variance for blocks of 100 samples Best Allan Variance for blocks of 500 samples
1 of 5 2000
1000 (after days of burn-in)
1266
17751		 1.49807e-09 
3.80823e-11
4.02409e-11
3.70821e-11 		 2.60542e-07
3.16235e-10
5.01130e-11
6.43210e-11 		 6.74659e-09
6.37598e-11
4.34813e-11
4.53838e-11
2 of 5 first 5000 of 10,000
last 5000 of 10,000		 5.64249e-11
4.43473e-11		 1.73850e-09 
7.79216e-11 		 6.35023e-11
5.54132e-11
3 of 5 1000 3.53343e-11 4.48254e-11  3.82697e-11 
4 of 5 1000
2000		 7.46437e-11
5.81559e-11		 1.06238e-10
7.58172e-11		 9.19596e-11
6.77010e-11
5 of 5 2000,
2000
first44839 of 45839
last1000 of 45839		 6.20969e-11
5.34274e-11
4.36180e-11
5.97650e-11		 2.18640e-10 
9.93111e-11
6.11854e-10
7.29004e-11		 8.15994e-11
6.32800e-11
5.27664e-11 
6.53183e-11

Evidence supports the assertion that oscillator short term stability improves toward an asymptote, as oscillators accumulate more operating time.

Statistics are too small to make a good estimate of 'yield', where yield is defined as delivering performance adequate to the needs of IceCube.

Additional samples are on order by Wisconsin.   #1 and #4 will be burned in more, and retested,  while we wait for them to arrive.

Data files available on request. Send me an email.

Measurements and analysis by Gerald Przybylski, Lawrence Berkeley National Laboratory, for IceCube.