Uri Blass wrote: ↑Sun Jan 12, 2020 6:20 am
Note that for me more interesting then centipawn is analysis like
white depth 20 lose against black depth 10
white depth 21 draw against black depth 10
white depth 21 lose against black depth 11
...
The idea is that if white want a draw simply play comp-comp game at small depth to see if white can draw.
If white can draw or win then increase black's depth by 1 in the next game
If white lose then increase white's depth by 1 in the next game.
The problem with that is that after all the data that you get, the question is, did you catch the mainline? Because if there's some line by white that can draw all the games of lower depth, or a line by black that can beat all the games of lower depth, but you miss them, then any data that you get back will be irrelevant.
I'm curious, how much depth does Stockfish need to switch to 36...Rxd4 ? Because if it's very high depth, then the tool would waste time increasing white's depth so it can draw, and increasing black's depth so it can win for irrelevant lines which would clearly be much slower than anything else that finds Rxd4 faster.
If Rxd4 is the only move that wins (for the sake of discussion) but white needs very high depth to draw against (say) Rxb7 then you could spend days playing games of increasing depth for white building data for irrelevant positions, while it's clear instead you should have found Rxd4 ASAP (anything with low depth without this move would only be wasting your time.)