I gave the new J-Field update a try today, and didn't notice any unusual variety counting. However, I also don't have any profiles which use less than 2 engines during Phase-2, so I may have missed anything new in that regard. Still, I wouldn't expect a single engine to increase variety much in the way the video shows. So I'm a bit stumped there.
However, if the main point was a sudden "jump" of variety from 1.9 down to 1.4, then yes, I
believe that's been a typical result of a "rejected bucket" for some time, since each group/bucket gets its own variety counter during Phase-1. In that case, Nate's got it. The spread plots in the video may very well be indicative of an imminent bucket switch depending on the confidence guard settings, environment, etc.
For example, if things aren't going well in the current bucket during Phase-2, it'll flip to a better bucket and restart Phase-2*, counting up from the variety values accumulated in that bucket during Phase-1.
Sort of like:
Bucket # | Phase-1 variety value | Phase-2 variety increase, at the time of bucket rejection | Phase-2 variety increase, at the end of phase 2 |
---|
1 | 0.3 | 1.6 (you see "1.9" variety at the moment this bucket is rejected) | n/a - bucket #1 rejected, and bucket #2 chosen automatically |
2 | 0.2 | n/a (not rejected) | 1.2 (you see "1.4" at the end of phase 2 in this "good" bucket) |
*I could be wrong in stating that it re-starts Phase-2. It may just "transfer" all of the Phase-2 data into a more appropriate bucket but still use it upon switching to that bucket, instead of discarding it. I just don't recall off the top of my head.
Indeed the recent changes to variety algorithms might make some of this more apparent, or change it in some other ways - or maybe not. Details beyond my pay grade, and I have yet to fully investigate.
Either way, the variety increase with 1 engine does seem a bit funky. If I'm drinking one kind of soda, no matter how many cans I crack open, I'm not increasing my variety of soda!