Your firmware has been absolutely invaluable during the month of hell recently. It has shown exactly how bad and how widespread the problems have been.
As for cutting out some candidate explanations: it only applies to one line and not to others; it’s upstream-only.
I need to look at the numerator and denominator of a ratio here, I think, and I have been lazy in not doing this earlier in this case. I need to look at the attenuation, power output and the noise level in the high and low parts of the square wave once more. If we assume for the moment that it is not the case that the output power of a transmitter is going up/down, then either the attenuation could be going up/down or the absolute noise level could be going up down. For the former, I would assume a change in resistance due to a ‘semi-contact’ make/break - I don’t have the words I need here. For the latter, interference would seem more likely, but then we are stuck with explaining how it can only be one line.
Since line 2 seems to have taken on the upstream-waveform of line 3, while line 3’s upstream has been cured, I do wonder if someone possibly swapped E-sides when they fixed the recent month of hell. Would that be a possible candidate explanation?