A extremely difficult blind spot

[hhfhhfhfh1]

Dear lightvector:
I was just doing a selfplay between weight b28s7332 and b18s9996 with komi 3.5 and found this position

I set the engine only analyze N8

I also use 18b to analyze it ,however, after 1M visits ,it did not found the blind spot as well
so I directly use it to analyze N8:

the winrate looks better.
the sgf i paste here:

199.txt

[hhfhhfhfh1]

Addition:xingzhen can find this rather quickly

[michito744]

@hhfhhfhfh1

If it is 28block, it will find N8 instantly. The weighting was almost equal to M3.
This type of oversight is not uncommon, so it is good to provide information about newer versions whenever possible.

[hhfhhfhfh1]

@hhfhhfhfh1

If it is 28block, it will find N8 instantly. The weighting was almost equal to M3. This type of oversight is not uncommon, so it is good to provide information about newer versions whenever possible.

The upper two image is analyzed by b28s7332 while the last one is analyzed by b18s9996 ,actually ,I was not sure whether katago's analyze on it was accurate , because I use b28s7168 b28s7332 b28s7628 and b18s9996 ,they all gave different winrate on those positions.

[michito744]

@hhfhhfhfh1

b28c512nbt-s733 is incompetent regardless of the value of elo. I have actually found examples where problems that were solvable in the 18block era are unsolvable.

b28c512nbt-s765 clearly shows qualitative improvement.

[lightvector]

Thanks for reporting! This will probably also join into a pile of test positions and possibly hint/sgf positions for future training.