Yuck!
I think this is the very first time you have shown the SNRM v time plot corresponding to an HCD event.
In all cases where I have witnessed a bad SNRM v time plot, similar to that you have shown, the fault has been a bad joint. HR or semi-conductive. The cure has always been to identify the joint and re-make it.
I really wish that there was some way to apply 95V AC ringing current on your three lines. In their normal state they just have the constant exchange battery DC voltage present. I wonder if that constant DC whetting voltage
without the occasional overlay of the AC ringing current actually accelerates joint faults?
I remember reading about how, in the early years of the 20th century, before there was the National Grid, some local electricity supply companies actually provided a DC mains supply. One artefact of a DC mains supply was that light bulbs would fail far sooner than the equivalent bulbs used with an AC mains supply. One trick to extend the working life of the bulbs with a DC supply was to turn the bulbs around in their sockets once a week (or was is once a month). Hence, by analogy, my suspicion is that the occasional application of ringing current to each of your lines may minimise the bad-joint effects. Something for A&A to consider, perhaps?