Salute!
BJ and mm43 have cracked the code.
The info I can find on the DFDR's reveals very slow sampling rates, and I am not sure if the data is time-stamped at the sensor or by the "black box" when it receives the data.
My background was in military testing, and we sampled the digital data at 50 hz or better. Our analog data from accelerometers and pots were recorded on old-fashioned tape! However, the digital stuff was parallel on the tape (7 channels), so we had a mechanical means of synchronizing the data.
As the referenced report by BJ states, the sampling rate was too slow to get an accurate replay of the rudder reversals by the 587 jet.
For my business we needed to see very good parallel data streams of control position, force inputs, and aircraft/missile attitude, gee, etc.
The serial data recorder using a single memory device has to use various update rates depending upon the parameters of interest. And I agree with the observation that filtering the data as it is being recorded is not a good thing. Our data reduction facility could filter as necessary, but we captured the "spikes" and such as raw data.