TRF: Total number of sequences with repeats removed from file

Could someone please clarify something from the kneaddata log file? I’m using version 0.12.0 and I’m confused (and alarmed) at the number of reads/sequences removed from my metagenomic files by TRF.

Input 1&2 read count: 70,179,888
Trimmed paid 1&2 read count: 65,368,820 (93% of input reads)
Repeats removed 1 sequence count: 305,115 (0.43% of input reads)
Repeats removed 2 sequence count: 308,169 (0.44% of input reads)

Despite this seemly huge drop in the number of reads/sequences, the file sizes of the trimmed.fq’s are only slightly bigger than the repeats.removed.fq’s.

So my questions are: is the count of sequences given from TRF the number of reads it has removed? Is there a big difference in meaning between reads and sequences? Does TRF only remove short reads, hence the file sizes remain similar despite the huge number of reads removed?