Another project I completed during my time at Next Gen Stats was identifying pull blocks in the run game with the tracking data.
For context, a pull block is when an offensive lineman moves across the formation to block a gap different from the one he started in and next to different players than he started next to (there are lots of nuances to pull blocks that make a simple description a bit challenging).
I created a rules-based approach using custom thresholds and football logic to identify when an offensive lineman was pull-blocking. Functionally, the main things to look for are the relative direction of movement from a lineman to the rest of the offensive line as well as the expected direction for a zone-blocking scheme. For example, I was working primarily with the Buffalo Bills tracking data as they utilized pull blocks often, particularly with their tackles which is fairly unusual. If Dion Dawkins is pulling from the left side, you will be looking for an inverse to check if Spencer Brown is pulling from the right side.
I validated the approach based on PFF labeling for pull-blocks for the 2024 season and also utilized the film room from NFL Pro to verify that all the plays I was identifying were actually pull blocks. I actually managed to find multiple plays that were pull blocks but were not properly identified by PFF, highlighting the benefits of a rules-based approach for a simple concept like this.