And since it looks now we're heading for a third example case today, I'll post the full results up as this time I finally did capture everything in the US=608 case for comparison.
It looks to be the case that after booting and the usual initiation and negotiation sequences that the router picks a set of US bits-per-bin that are all around 1 bit too high to arrive at US=608k with a reported initial SNR margin in "6dB" and will then go over to a new plan with US= ~400k with roughly the same _shape_ to the low bin allocations but with most of the values being 1 bit lower. So not as if there's some single frequency noise source appearing or as if there are a lot of high-noise frequencies that shift up and down, but it suggests that bitswap can't deal with it and it's noteworthy that the US SNR margin does not show a higher figure in the first state than after the change (although this router doesn't report fractions of 1dB).
I realise that I need to look this up. I'm assuming that bitswap doesn't permit the US-DS partition to move? In any case, it's true that in the previous cases the US goes down by 200k and the DS sync rate has gone up by 32k or 64k or so. It would be hoped that the router would display the appropriate number of US bins to show where the partition really is, and reflect the fact that that particular line is on the IPStream Max Premium service with its advertised higher cap on the max US sync of 832k compared to the more usual US cap of 448k for the home/home-office service.
I wonder if I’m seeing the effects of for having had this line too long through too much history, for having had it in operation since 2004 first in the days of pre-Max fixed 500k service, then getting a free upgrade to SOHO IPStream Max, then choosing oi move it over to the IPStream Max Premium service, and am wondering if I needed to ensure that it got pushed back into the 10 day training period so that the changed US-DS partition’s effects can be retested fully.