download opening book here and put it in testing/book
suggested improvements:
- hardbound =
time_left - 10msmove overhead. alloc_time =min(soft_target, hardbound). (suggested by a member of the stockfish discord server) ✓ done - softbound for elo gain (must be smaller than hardbound)
softbound is the budget per move you want to spend; checked once at each iterative_deepening iteration boundary, to stop launching a new depth if you can't realistically finish it.
hardbound is the time you must not exceed; checked every N nodes (2048 in this
engine) via the canceled flag, to interrupt the current iteration if it runs
past budget.
softbound < hardbound. They give different signals: soft says "we got a decent result, don't push further"; hard says "stop right now or we lose on time".