When the time for a move is up and the best move is selected it may be beneficial to continue searching a bit longer when the result isn't clear cut.
A possible strategy is as follows:
See the computer go mailing list archive for inspiration or other solutions. Some bots seem to do this a lot and with M being the amount of time allocated initially for the move the go as far as 3*M for N.
I tried a similar approach (return the move with the best win ratio), but it didn't improve strength. Maybe searching more is the key here.
I'm not sure to be honest. I've currently trying a few things and none of them seem to improve the bot. Even the large patterns (#298) isn't really working. I guess the time for easy wins is over.
Aaaaand it didn't help. 😁 It's probably better to just optimise the variables in time_control instead.