Sugestion: alternative eval display in % of win probability

[Mike S.]

Alternatively, or in addition to the traditional pawn units, or centipawns eval display it would be nice to have an evaluation display with an estimated winning probability in percent, based on the engine evals.

The % values could either be provided by the engine using it's internal conversion (eval to %), or the interfaces could use generic, updatable/adjustable conversion tables for engines which don't supply % values themselves.

(I am aware that current protocols don't support that yet.)

By that, engines could continue to use the traditional way of expressing positional factors by values from the material scale - if all engines even do that still, I don't know that - while at the same time, users could be presented with such winning probabilities in % which seem somewhat more telling to me.

Actually I wonder why this hasn't been introduced decades ago already...

[zullil]

Reminds me of this post: http://talkchess.com/forum/viewtopic.ph ... 63&t=39963

[fern]

For me such a device is very easy to implement: just put at the end of the PV line number 100% to express the likelihood of victory for the engine.
Or, if they prefer, 0% of likelihood for me to even get a draw.
If you prefer to add some complexity, they can put a 50% when there is no moves in the board and forces are equal.

Always mathematically inclined regards
Fern

[bob]

Alternatively, or in addition to the traditional pawn units, or centipawns eval display it would be nice to have an evaluation display with an estimated winning probability in percent, based on the engine evals.

The % values could either be provided by the engine using it's internal conversion (eval to %), or the interfaces could use generic, updatable/adjustable conversion tables for engines which don't supply % values themselves.

(I am aware that current protocols don't support that yet.)

By that, engines could continue to use the traditional way of expressing positional factors by values from the material scale - if all engines even do that still, I don't know that - while at the same time, users could be presented with such winning probabilities in % which seem somewhat more telling to me.

Actually I wonder why this hasn't been introduced decades ago already...

It is not so easy. A +3 might be a win at long time controls, but might not even draw at very fast time controls. How would you ever calibrate something like that with the wild variety of hardware options there are???