Going back to the data collected so far with PEAC16, I did some adjustments to get this graph:

This is the rate (in %) of errors missed by the GrayPar unit, depending on the error burst's length.

So the peaks of misses are around 33% at 6, 12 and 17.

Burst lengths 1 to 4 and 8 to 10 are below 12.5%.

13-14-15 are not bad either.

I also traced the adjusted rates of error discovery at the PEAC level:

so "most" of the errors are indeed found after 8 cycles but that is still not a guarantee. An adaptive algorithm is required for the retransmission, to move the retransmit window and compare previous data, looking for whether they match.