OK. Well. I don't know if something changed or what it truly was. But I can see that there is something going. Good thing I have a large test bed. I ran through a ton of clips checking this out and found a few interesting things. There now appear to be 2 ranges of values used to indicate what it is, video header or body and audio header or body, and what is going on with it...
The old set worked in this range:
This was the norm for every clip I had older than about 5 months. But then I started grabbing clips from "The Dead Zone" on usa during the marathon. So no more than 2 weeks ago and for almost 2 full weeks 1 episode a day. On the 8 I got thanks to tennis, does anyone really watch that anyway? , I got 3 with this range scheme and 5 with the new one...
VIDEO HEADER RESETS AT > 00 CB 10 --> 0 BB 20 == diff of FF0 (4080)
VIDEO BODY RESETS AT > 31 8e 6c --> 20 50 40
VIDEO HEADER RESETS AT > 01 4f c4 --> 01 3f cc == diff of FF8 (4088)
VIDEO BODY RESETS AT > 6d 7d 9c --> 5c 30 40 == diff of 11 4d 5c (1133916)
So 2 totally different ranges. I don't get it... But these are the only 2 ranges I have seen so far. The rollback point changes by a few bytes in either direction for both header and body but it stays in roughly the same range. So... if that remains the case we can still use this technique the fixing of the -1x4 problems...