Where are the 10-13 games? They're -4 under Stevens so far, which is far back as this current squad can really be traced.
538's RPM method is fine as a preseason projection, and in fact, I had the Celtics winning 33 games preseason using my own version of RPM. The purpose of this "toy" is to track performance for a large over/under pool I'm in.
However, with respect to PKB's question of how much weight 10 games has? The answer is a lot! Tangotiger summarized this here, and I have posted similar results for my own method. To summarize, using this "toy" on past seasons, at this point in the season in previous years, the R^2 between the projected rest of the season and the actual rest of the season is approximately 0.5. I'm using an SRS-based method however, rather than RPM one, which is why OKC grades out so poorly. It's literally just my own version of SRS, put into a logistic regression to get probabilities of winning any individual game. Then I plug that into the remaining schedule, and project out the rest of the season. It's not back-to-back adjusted yet, although the home court advantage weighting should mostly capture that.
Now, I haven't done any testing as to how that compares to RPM-based projections. My guess however is it holds up pretty well, as it has the advantage of "seeing" parts of the season RPM hasn't seen, which for young teams especially is going to be important, given how RPM handles rookie/sophomore seasons. It's certainly done fine in general however.
EDIT - it however just a toy. It took a couple hours to run the regressions and build. A subjective analysis will likely beat it, even if preseason RPM projections will not.