If data is split across multiple files that are sequentially numbered, then it is foolish to beleive that every file exists. Some files can go missing, or be corrupted.
I learnt the hard way that one version can be corrupted, and the other version is fine, and it's all too easy to replace to good version with the corrupted version
Yeah it can change because a fixed sampling rate would make your analysis too easy
Did I expect the sampling frequency to be some multiple of a microsecond? Yes. Was it, in reality? No. In some random subset of the data, it can be 1.0004 microseconds.
Ha ha jokes on you who needs metadata.
If you find a channel that is labelled "temperature", do not assume that it measures temperature. It could be literally anything.
Just for fun, the data collector can decide to add some data channels. Or throw some away.
Remember that channel called "temperature"? It can be accurate in the first half of the data, but wrong in the second half.