At position Q16,R17,R16,Q17,P17,P18,O17, @aesalon started the conversation:
Here is a move that is good test of the move labels.
( FYI @GaJ )
Using these definitions.
Has it been proven to be inferior to the other choices?
Can it become the best move in certain situations?
If there is strong evidence that it is almost never the best move, can that be enough to call it a mistake?
Where does the fact that it has been played a few dozen times by pros and AlphaGo Zero was defeated by it fit into this decision, if at all?
What about the fact that it is extremely rarely a meaningful mistake for 99.9% of amateurs?
What’s interesting about this position? The given move is the most common one among pros, and it appears over 100 times in AlphaGo Teach. I’d be shocked if you could find any pro who thinks the end position is not joseki, which is required for the move to be marked GOOD or MISTAKE.
Is it ever the best move in a situation?
If you include AlphaGo teach with games from the past year, you get 7532 3-3 invasions. 163 play that move. 2.2%.
I am certain I’ve heard explanations that include “this is not joseki”, probably more than once. It is just glossed over because it’s been discussed enough. Plus, would one source explicitly stating that be enough?
Apples and oranges. A like comparison should be from the starting position, not several moves back. In a 12x12 square of the starting position, pros chose to stretch 88 times (52%), double hane 80 times (47%) from 2016-2018. AGT chose stretch 130 times (24%), double hane 393 times (73%). To answer your question, the answer is: at least pros and AlphaGo thought so that many times.
KataGo concurs. LeelaZero is a little less favorable. KG thinks the moves are equal to 1/10 of a point when I ran it. LZ thinks double hane is better than stretching by about 2.1% win rate in my test on about 10k playouts.
Do share one, if you have the time. Probably one source is enough. Pros sometimes use hyperbole to make their point though.
You know by now that I’m using the 2018-2019 database of 5000 games for that number. The argument will hold weight either way. 3-3 invasions are played a lot, that old joseki is not played. I gave you the database, you can check it out.
I also completely disagree with you that it’s apples and oranges. It’s apples to apples. The stretch “joseki” serves the same function as the ones following https://online-go.com/joseki/350. The double hane(a change of direction) is the orange.
Anyway, this was posted as a discussion.
Imagine that these conditions are true:
There is no situation the old joseki is the correct choice in.
There are very, very few situations where it is a meaningful mistake in.
Professionals no longer play it.
What would you label it as?
I thought this is an interesting test of the move types.