Talk:Solutions to puzzles

Removal of puzzle 5 (done)

I propose the removal of puzzle 5 unless someone can prove the solution is good (because the bot disagrees strongly). At the start, KataHex thinks Blue has only a 35% win rate, with either 1. b8 or b10. Move 8 is a mistake; Red f9 instead has a 98.6% win rate. Move 22 at j2 is only 3% for Red; KataHex suggests i6 which is 99% for Red. Hexanna (talk) 01:01, 7 May 2023 (UTC)

• For posterity, this puzzle has been removed. The puzzles have been re-numbered so the current "puzzle 5" is fine. Hexanna (talk) 15:45, 7 May 2023 (UTC)

• And in case posterity would like to see what the removed puzzle was, it is still in the history. Selinger (talk) 16:20, 7 May 2023 (UTC)

Yes, I wondered about that too. Arek's a strong player, but with 23 moves in the proposed solution, and nothing approaching a proof that the solution works, it doesn't really qualify as a puzzle. It's more like "here is a game that I won". Selinger (talk) 02:16, 7 May 2023 (UTC)

Also, the solution is definitely wrong; the Mohex dfpn solver can prove that 22. i6 is winning for Red. Selinger (talk) 03:17, 7 May 2023 (UTC)

The new puzzle 8

The new puzzle 8 strikes me as another open-ended question similar to the removed puzzle 5. Can it be simplified to get some kind of essential point out of it? We don't have any criteria for what counts as a puzzle (as opposed to a tricky situation from a game where computer analysis showed what the winning move was). But I think one possible criterion is that it should be simple enough for a human to analyze it with sufficient effort at least in principle, and ideally, that the answer is something that can be explained. My point is not that the puzzle is too hard (I am fine with hard puzzles), but that the answer seems to be something that must be verified by computer even after somebody tells you what the answer is. Selinger (talk) 14:11, 8 July 2023 (UTC)

The difference with the old puzzle 5 is that puzzle 8 clearly has a unique solution. I added it because I found the answer elegant. The board position is messy but is representative of a "real" game. It might be hard to analyze with strict mustplay analysis (like puzzle 1 by Tromp), but I think with sufficiently good intuition you can "guess" the right answer. KataHex's policy net (without search) has the correct answer as its top choice by far, with 55% probability. I'm not very invested in this puzzle though, feel free to remove or simplify it. Hexanna (talk) 14:56, 8 July 2023 (UTC)

Here's a first attempt at simplifing the puzzle, in the style of Seymour. Blue to move. The unique answer is still g12.

Still, the puzzle is very hard because of the immense number of possibilities. Mohex takes the better part of an hour to verify it, and longer to verify that there are no other solutions. The difficulty of Seymour's puzzles is mitigated by the fact that he provides a perfect opponent, which we do not have here. Some very good player might be able to guess the answer, but I doubt that any human player can verify the answer without the help of computers. Selinger (talk) 17:51, 9 July 2023 (UTC)