THJ for the bench!

[pompelmo]

Lots of problems here.

The LAC series didn't add anything to the negative data for KP/Maxi because the data being used is regular season (a much larger sample size).

The data doesn't somehow narrow in on KP/Powell's start against Cleveland vs. KP/Maxi's start against NO.  It tracks every moment of every game of the entire season.  KP/Powell wasn't just better when they started together.  They were better when they played at any point in any game together.

Sorry for not making this clearer. What I mean is the opposition that KP/Maxi and KP/Powell pairings face also matter. If you play Maxi against the more threatening matchups every time and get slightly negative +/-, while managing Kleber's minutes against lesser opponents, it is only natural for him to appear worse on the accumulated data. These stats make the assumption that either the quality of the opposition is always the same or it evens out in the long run with large enough data (which I pointed out in my above post). It doesn't.

Before the second game of the play-offs, Carlisle said Kleber didn't start much during the regular season because they prefer to manage his minutes unless absolutely necessary. Not because he is the worse option. I am pretty sure he would have still started in Clippers series and play the majority of minutes even if Powell was available. 

As you said, Dallas certainly have better data available which also takes matchups into account rather than relying on stats derived from basic on/off data. I would expect nothing less. What I say is you are reaching conclusions on incomplete data; similar to saying Joe Harris and Stephen Curry are similar quality three point shooters because their three point hit rate is close at 43%. 

Let's assume that it is the only data available to the public. This is not the stat that the Mavs would rely on.

They know Curry hits 41,5% of his almost 5 pull up threes a game while Harris can hit only 1 in 5 attempts which he takes only 1 per game (probably all heaves). They know Steph Curry can hit over 35% of his even tightly and very tightly contested shots while Harris excels only on open and wide open threes. They wouldn't confuse Harris with Curry.

Publicly available stuff is pretty basic and loaded with lots of noise. 

By the way, I don't think they ever thought "let's just ignore this data we have collected from the Clippers series because these were not regular season games" either. Separating regular season and play-off data is alright for the history books, but doesn't make much sense for the analytics team to do so.