Yes it’s categorizing based on the score loss for the move
Excellent <= 0.2
Great <= 0.6
Good <= 1.2
Inaccuracy <= 4.0
Mistake <= 10
Blunder > 10
7 Likes
What made you land on those numbers specifically?
Acts of divination by folks smarter and better at go than I
3 Likes
funny how the only “blunder” i made in my match today i was already ahead by 43 points. An 11-point mistake didn’t matter at that point, my opponent resigned shortly after:
I don’t have a better suggestion as of now but maybe in the future this might support the idea of incorporating win rate into the categorization. I’m satisfied with the current system so far.
3 Likes
0 blunders, easily. 0 mistakes? I don’t think so. I think it’s hard to do 0 “mistakes”/inaccuracies against a complete beginner almost as much as against a professional player. because there’s so many options for good moves that it’s a waste of time to find the best possible one. Playing against a pro player, your options for good moves are much more limited, which might actually help avoid mistakes in some cases by making them more obvious.
4 Likes
It’s still a mistake though in an objective sense.
Even if you’re winning the game from move 10, and you ask someone “what could I improve on?” would it be better if they said “nothing because you won the game”?
I think I’d prefer the categorisation to be objective on what could be better pointwise, than to use Katago’s winrate, which we know isn’t the same as a human winrate from a given position, to factor into the decision.
If I intentionally make a mistake to simplify the game, the I can ignore the categorisation when it comes to reviewing the game, but I think it’s more confusing to pretend I played perfect say if I play moves that don’t change the winrate, until they do suddenly change the winrate.
3 Likes
The word “blunder” is kinda weird tbf >___>
I think it’s more commonly used to mean “losing the lead” instead just “big mistake”. If you’re winning by 2–3 points, and then make a mistake worth 4 points in the end-game, that is a blunder. But if you decide to play a non-optimal move when you’re winning by a ton, its not really a blunder per se, but more like “being lazy” or “taking the easy way”
You might see a situation where you could kill some group with a ko, but if you’re already winning by a good margin, deciding to let it be and just playing simple yose is perfectly normal, and many would say its actually a good idea to avoid any unecessary risk.
But yeah, on the table it just means “mistake more than 10-points” - good to take it with a grain of salt ^^
6 Likes
I wonder if this indicates another kind of move we could tablulate, which is outcome-changing blunder.
That would probably get the name “blunder”, and “big mistake” would be … “big mistake?”
2 Likes
Even more as “easy” or “lazy” it can be the better move considering the increase of chance of winning by solidifying instead of the biggest possible score difference.
4 Likes
“major mistake” or “severe mistake”?
Or if we dont wanna repeat the word “mistake” twice, maybe “error” might work? No idea how well those would translate into different languages tho >__>
Indeed, hard to make strict categories for a game where everything is always so relative and situational ^^
2 Likes
So what if you actually gain points?
2 Likes
Those numbers aren’t absolute point change, they’re “how many score points does KataGo think this loses you compared to the optimal move”. You’ll never see a number above zero.
1 Like
I’d also say a blunder is not an absolute size of a mistake, but is relative to your level, being one far below it. You can self identify blunders.
5 Likes
Sure you will. At least if I understand what’s going on - Katago doesn’t try to maximize points, it tries to win. Occasionally you can find a move that gains points but is not optimal.
Here for example H4 apparently gains 1.8 points:
5 Likes
Ahh, neat! Appreciate deepening the nuance of my understanding. I believe my original point still stands, other than that error.
2 Likes
Yeah, and then you say “oh, lets use win-rate instead of score, because that always behaves itself” right until you realise that often it’s pegged at +/- 100% for most of the game 
1 Like
That’s the advantage of using points instead of win rate – it lets us continue to evaluate moves even when win rate only changes by 0.01% or less. But what if win rate is used first, and then in case of a tie, points loss is used to distinguish inaccuracies and mistakes…? like some sort of smart system that uses the best features of both metrics.
2 Likes
I think using score is fine, as is going from +50 to +30 counting as a blunder. Yes, maybe it was on purpose as part of simplifying a won game, but maybe it was not so worth highlighting. If you do a few more of them and throw your lead then that, rather than the 2 point endgame mistake which took you from +1 to - 1, can plausibly be considered the more important mistake to correct because if you stop throwing away 20 points in middlegame you can afford making 9 2 point endgame mistakes instead and still win.
1 Like