Here's a different look at how plays perform based on the 2040 season: https://drive.google.com/file/d/16zSoQjAs_wfclzmQKP5xzmThpluXUqr9/view?usp=sharing

What's This Thing
Other than the WeightPoints column you're looking at probabilities (based on a Bayesian statistical model) for various categories. These probabilities were than weighted against every defensive play a team in this league called against the play last season (2040).

Some specifics to know:
* WeightExpectedYards: This is an average of plays excluding sacks, which would make this column go very negative for a lot of plays and mess up the WeightSuccess column.
* WeightSuccess: This is the probability that an offensive play will meet or exceed the expected yards for a play.
* WeightBigPlay: This is the probability that an offensive play will exceed the expected yards by at least 3 yards on a play.

Raw Dataset
The raw data set was 5.8m rows across several leagues and >100 seasons. Data was prefiltered to teams that completed >=50% of their passes or averaged >= 2 yards rushing for a season to filter out tanking teams and bad CPU teams.

If you need more detail on how to read probabilities or Bayesian statistics, let me introduce you to my good friend ChatGPT.

I assume plays with penalties in are filtered out?

They are filtered from the raw table unless declined by the offense.

any stuff I do on plays, I just filter all plays with penalties out.

Rationale being that there may have been say a PI - wr still makes the catch for 1st down, so declined. May have gained 11 yards say but without the PI who knows?