In this game, the goal is to trigger the longest chain reaction. Once you click on a cell, you can no longer click on any of them before they all stop moving. The chain reaction and the soothing sfx make it a relaxing game. You can play the game is various places, for instance here.

For my last submission, I spent roughly a week instrumenting and patching libTAS in order to bruteforce 1.3 million possibilities in 5500 grids. But we're mid-2026. AI is now sufficiently advanced to be able to fully reimplement the game in C with a few prompting (or, to be fair: a LOT of prompting and frustration about how stupid it is despite its capabilities), along with a complete bruteforcer.

Thanks to this, I was able to bruteforce over a billion grids in less than 24 hours. While the last bruteforce barely found a solution hitting the 4000 cap, I found many solutions over 5000, and a best solution of 5800. In order to find the best grid, I used the strategy of making a few attempts on each grids, then performing a full search for the grid when one of them is below a certain threshold (2000 or 3000, I did some tweaking during the bruteforce). I ran several bruteforcers in parallel tweaking these parameters (from one doing an extensive search on all grids to one aggressively cycling through grids).

The resulting solution is an elegant grid with almost no unreachable cell which finishes visually pretty well sorted.

The bruteforcer code is on Github, along with documentation regarding the RNG used by Ruffle.

There should be roughly 2 billions possible grids (2^31) due to the way the seed is constructed. While I used a time-efficient strategy to cycle through grids to quickly find the most promising ones, an extensive search of the whole address space should be doable in reasonable time (several days/weeks of computation). I doubt, however, that the best result would be significantly higher than the current solution.