On the point about what constitutes "an error" in these discussions, I agree that it is the ES counter that is used.
A few years ago, I saw the same kind of thresholds as they applied to 21CN ADSL2+ (attached). However, in that picture, the threshold was termed "MTBE" (mean time between errors), and was set (on the speediest/least-stable option):
DLM intervened when MTBE was less than 60 seconds
DLM de-intervened when MTBE was greater than 600 seconds
DLM stayed put if MTBE was between these values.
An "MTBE of 60 seconds or less" is equivalent to the Zen page stating "no more than one error per minute".
When you look at the error thresholds in terms of "time between errors", you can see how the increments in ES are best used toward calculating the MTBE (by inverting), rather than using increments in the CRC counter.
Note that the BTW 21CN error thresholds are altered 10x between service options, but the Openreach GEA error thresholds only change by 2x.
Note too that the resyncs are handled on 21CN as MTBR.