Finding Pawn Breaks automatically

To really understand a chess concept, I think that it's necessary to see a lot of examples. It can be especially useful to compare situations where a specific idea worked to similar situations where it's the wrong approach to the position. The only problem is that it's often difficult to find many examples. One would need to go through many games which is very time consuming. This is where filtering in chess databases can come in handy.
These filters can be very helpful, but there are always more things one would like to search for. In this post, I'll focus on pawn breaks since they are such an integral part to the strategy in a given structure and difficult to understand without many examples. There are also often multiple different breaks for both sides, so it's interesting to look at all different options to get a better understanding of certain positions.

Defining Pawn Breaks

The first step in trying to find something is to specify what exactly one is looking for. For this post I define pawn breaks as pawn moves that attack an enemy pawn which is blocked by a friendly pawn.
This might not be the most exact definition. I went with this one since I think every move that fulfils these criteria will be deemed a pawn break. The only question is if there are other moves which would also count as pawn breaks. Let me know if you have a different definition of pawn breaks.
Using a simpler definition has the big advantage that finding pawn breaks is quite straightforward. I first found all the blocked pawns in a position and then looked at each game move to see if this was a pawn move that attacks a blocked pawn. If you're interested in the details, you can find my code here.

Examples

I think that looking for pawn breaks is especially useful when studying specific structures. So I looked at the pawn breaks in the Benoni structure where White has pawns on c4, d5, e4 and Black on c5, d6 and e5.
There were of course many position with the usual breaks, f4 and b4 for White and f5 and b5 for Black. But I found it particularly interesting to see some other breaks for both sides, like g4 for White:

Black has also the plan to push on the kingside and break with g5 or g4, like in the following position:

Constraints

As always when doing little projects like this, I added some constraints to keep it small. I only looked for pawn breaks and filtered the games based on the structure beforehand. This lead to many breaks where the structure has been transformed before the break was played, including endgames like the following example:

These examples are also very interesting, but aren’t directly linked to the opening in question.
So when only studying a specific opening, it might make more sense to limit the search for breaks to a specific structure or specify that it has to happen in the first 15 or 20 moves.
I also only looked at the moves played in the games and not at any engine evaluation. This was mainly done to save time. One could check each break with the engine or one might even take a position and "ask" the engine if the best move is a pawn break. The implementation of the part of the pawn breaks would be the same, one would just analyse the position with Stockfish before and after the break.

---

If you enjoyed the most, you might also like this post about finding ideal squares for pieces with Stockfish.
Let me know what you think of the idea and if you would change anything about the way I search for pawn breaks. I'd also like to hear what other themes in chess games you'd like to search for.