The Mavericks are scoring high in a bunch of the stat-heavy models. I don't have my databases in any shape to mess with, but I expect they're actually pretty noisy. You have one year of Doncic, KP's data is from some time ago, they overhauled the roster at midseason last year and many of those parts (or their FA replacements like Delon Wright and Seth Curry) should be also noisy.
OKC is probably a tad high because nobody wants to predict Paul and Gallinari deals (and maybe also Adams and Schroeder ones) and there's not much to replace them with. Zach Lowe pointed out that a lot of teams might just wait on the Thunder and try to pluck the players in the offseason, unless it's a contender trying to get Paul/Gallinari/Adams as the "one last piece" addition ala Gasol last year.
The Nets and Spurs were both really noisy datasets coming into last season, but everything worked out, and with Carlisle and the whole staff still in place it looks really similar to those two situations.
For what it's worth, Pelton's system put the Mavericks in the playoffs but he said everyone from the Blazers down to Suns (in the combined prediction ranking) were sneaking in enough times in the simulations to suggest it could be whoever gets "hot" enough (going like 10-5 over a span) to end the season.