hesela.devInterleaved parity
Interleaved parity assigns separated data positions to independent parity groups so that a short contiguous loss can be distributed across those groups.
The XOR construction reviewed in FAST 2010 groups every m-th data sector with its parity sector [1].
In Hesela's 16+2 byte example, even and odd positions form two groups of nine. All 17 adjacent pairs are recoverable, but only 81 of all 153 two-position erasure sets are. Losing positions 0 and 2 leaves two unknowns in one group. These are exact constructed-pattern counts, not fault probabilities [2].
Known erasure locations and correct, version-consistent survivors are assumptions. Two parity bytes do not by themselves establish an MDS any-two-loss guarantee [2].