Whats the assumption we can potentially target as reason for the counter-intuitive result?
that data in pdf files are noisy and zstd should perform better on noisy files?
What's counter-intuitive about this outcome?
What's counter-intuitive about this outcome?