LeelaZero always assumes a game is played with 7.5 komi. KataGo respects what you tell it. This means in a game with e.g. 0.5 komi the LZ analysis is invalid and vastly different to KataGo and leads to confused users. For example Go (Baduk, Weiqi) Players on Facebook Public Group | Facebook .
If the komi is not 7.5 then I would suggest either displaying a warning on the LZ analysis, or simply not using LZ at all as KataGo is superior in pretty much all regards now. 6.5 komi Japanese is pretty close so maybe that’s ok to use LZ and it’ll be only a little wrong.
The thing about win percentage is that its foundation in any particular position is much more obscure to the user than the estimated score.
It’s the difference between saying “I have £5000” and “I have enough to get by”.
I also suspect that people who use score estimation rather than winrate are less slavish to the bot, because five points sounds a lot less intimidating than 95%.
I expect that score estimation is a lot more useful in endgame as well, when the typical user would rather have a -1 +1 -1 +1 style of display rather than a \ / \ / rollercoaster graph.
I don’t know if this is the best way to use it, but I use win percentage when looking at joseki suggestions in AI review and cross-referencing them with the OGS Joseki Explorer to decide whether or not to learn them, and score evaluation otherwise.