Change search behaviour based on time control and speed

[scarlet canyon]

#sf-general message Just leaving this here.
I'm guessing it is not currently done to simplify things but I think it could be a good idea at some point to avoid having to compromise strength at short time controls to improve it an longer ones and maybe it'll lead to some gainer, who knows.

[upbeat vector]

improving stockfish at STC or LTC while knowing it doesn't gain at higher time controls is just bad - because in realistic usage you gain nothing while you make sf stronger at shorter time controls thus making it harder to improve

[scarlet canyon]

Realistic usage can vary tho
And maybe it helps when sf is low on time?
And it doesn't necessarily have to improve at shorter tcs, maybe it helps at longer ones too to take this into account, idk tho, it's just an idea

[upbeat vector]

varying stuff that is known to scale non-linearly depending on time left was tried a lot

it fucking sucked at longer TC

like you would expect it to be better at STC and neutral or maybe even better at longer TC

but in fact it was better at STC and just worse at longer TC for w/e reason

[polar fractal]

Depending on how the patch is written, it could scale to TCs beyond fishtest TC. For example putting + depth in condition, as opposed to adjusting when depth < constant. Though of course, it still has to perform well at stc through vltc.

[upbeat vector]

also needless to say that all this patches eat insane amount of resources because they still need to be checked at VVLTC to see if they fuck up scaling

[harsh roost]

Theoretically if lc0 is stronger than SF at longer time controls, you could just make an engine that switches between lc0 and SF, right?

[knotty kraken]

SF can use the cached results of previous calculations, so switching between probably would make it weaker

[plain temple]

Except lc0 isn't stronger at any time control