OK. So I've set the timeout to be 5 seconds, because setting it to 0.0 (which says just fetch whatever is available) seems to cause data to glitch in the ways shown in the two attached plots above. This could be a digitizer issue, it could be my code, I don't know. But positive timeouts are working for me.
More importantly, I've set the number of samples to fetch to be a third of the actual record length, which has also led to more reliable actions. When the number was too small I get throughput problems (which means I definitely get missing data from chunk to chunk) and if it's too big I get some other glitches.
I seem to be getting more consistant behavior when every fetchbinary16 command is grabbing the same number of samples, then when it is allowed to be variable (depending on the whims of what's available).
Hope that helps someone else out there.
