AI now makes use of matches that don't have roster information (typically no replays)
Use roster info from the match if available.
Use roster from the same tournament if other matches available.
If tournament in last 30 days -> use ShiftRLE active roster list.
If team with the same ID (ID is unfortunately attached to org, not roster) has played in last 90 days, assume the roster stayed the same. (When roster info from match available: Update an org's roster if they play with a new roster twice in a row or the org hadn't played for 90 days.)
If data has one minor inconsistency (e.g. series length doesn't match actual game count), the inconsistency is squashed. If there is a big issue or more than one, the data still gets thrown away.
Fixed a bug where newly formed teams attending international tournaments weren't correctly assigned their region. This mainly affected Gamers8 and happened to mainly cause EU to be lower rated than they should've been.
Player uncertainty now correctly updates with score scaling.
Reduced overall uncertainty slightly -> Every rating is slightly less recency biased and updates slightly slower
Made the 2 week rating trend 11 days so results on Friday aren't still overlapping with the Sunday of the regional 2 weeks before.