The more I think about it, the more I can see a similarity to Bald_Eagle1's problem of some years ago.
In his case, one of the copper conductors of the pair in the "tail" cable (which links the pole top DP to the underground D-side cable) had fractured. The PVC insulation held the two ends close together and the circuit "sort of" worked, albeit with degraded performance. The circuit's performance was better in mid-winter rather than mid-summer due to the change in the PVC insulation's properties with temperature. (Softening.)
Essentially the circuit behaved as if there was a parallel resistor/capacitor shunt connected in series with one leg of the pair . . . which acted rather like a high pass filter. The defect was also vibration sensitive. So whenever the pole was climbed by an Openreach technician, the modem - DSLAM resynchronised.