[th1nhng0]

Thanks, these are all excellent suggestions.

The "distance to goal" constraint is a clever idea, though it's tricky to implement because the tunnels act as teleporters, so a simple distance metric doesn't apply. I'd likely need to pre-calculate a shortest-path graph first.

I'm benchmarking Huub and Pumpkin now. My main goal is raw performance, as a friend's imperative brute-force solver (https://github.com/FoxtrotOnce/RailboundSolver) solves most levels in under 2 seconds. It's a high bar to meet with a declarative model, but I'm keen to see how close I can get with these optimizations.

I've also just pushed some initial benchmarks for Chuffed vs. OR-Tools to the repo if you're interested (https://github.com/Th1nhNg0/railbound_cp/tree/master/docs/be...). Thanks again for the great feedback!