I'm imagining that the byte-aligned LZ matches (e.g. one match emitted for timestamp and another for the remainder of the log-line) in the first pass are using

I'm imagining that the byte-aligned LZ matches (e.g. one match emitted for timestamp and another for the remainder of the log-line) in the first pass are using relative offsets, so they are actually exposing a deeper pattern to the second pass.