The read is scored using the metric d_err. This is a value in base pairs, measuring
the difference between the actual alignment and the correct alignment.
Positive d_err values mean the read is aligned too far to the right.
Under the "strict" scoring scheme the read POS has to exactly match the correct POS. Under the lenient scoring scheme, if the aligner has soft or hard-clipped any bases at the start of the read, the comparison point is moved to the right by that many bases, effectively checking if the aligner has correctly placed the read after the soft-clip.
By running a contrast analyses with these two settings it is possible to uncover problems with the soft-clipping algorithms used by aligners.
Reads from completely inside insertions, which are normally left unmapped by a linear aligner, can be mapped by a graph aligner if the insertion is known. If the aligner piles up these reads at the start of the insertion point, a reassembly variant caller like Haplotype caller can retrieve the insertion. For this reason, such reads are scored relative to the insertion location.