If I might chime in... I the computer can play itself (if this is not already possible, I don't think that would be too hard to implement out-of-repo), a test-case could be created to check for a valid end-state and just let the machine run until it hits that scenario.

If this this could be done without invoking the UI, the amount of times the tests could run would significantly increase.
That would also give an indication whether the UI is involved, i.e. if the scenario never occurs, the error is bount to be in UI related code.

If I might chime in... I the computer can play itself (if this is not already possible, I don't think that would be too hard to implement out-of-repo), a test-case could be created to check for a valid end-state and just let the machine run until it hits that scenario.

If this this could be done without invoking the UI, the amount of times the tests could run would significantly increase. That would also give an indication whether the UI is involved, i.e. if the scenario never occurs, the error is bount to be in UI related code.

The Extreme AI, which I believe was used in this case, is actually implemented in Java in our RoyalUrAnalysis repo. Therefore, it has been tested extensively in a similar simulation of playing it against itself (e.g. to determine whether the first player gets a big advantage).

Therefore, I think it would probably be a bug in the interface between the client and the AI. That would require us to write the tests to run in JavaScript, unfortunately. However, I think something like this getting me to write some JavaScript tests would definitely be an improvement!