Question about about strange game results (non LG!) Hex, Havannah
7 replies. Last post: 2021-04-27Reply to this topic Return to forum
Miwarre ★ at 2021-04-25
I am collecting Hex games for study from LG (of course), BGA and PlayOK.
Both PlayOK and BGA (before 2019) do not record precisely the game outcome and just give a winner; in a sizeable minority of cases the stated winner seems in contradiction with the game flow.
For instance, this PlayOK game (hx535572) : Black is last to play and is clearly winning, still White is declared as winner in the TXT version (follow the “TXT” link to access it). Another: PlayOK game (hx535617) : this time White is the last to play and is winning, but Black is declared winner in the TXT ( a simple PlayOK example to check result notation).
In neither case the (declared) looser may have run out of time, being the last to move; I am not sure he could have resigned (no longer being on move), but in any case it should have a desperation move with a 60 sec time margin, not comprehensible in the context.
Out of the 18236 PlayOK Hex games of 2019, this happens ca 3700 times, slightly more than 20%: a bit too much to be an occasionally glitch. A few (fewer) cases also in BGA, but I do not have figures yet.
Anyone has any hint? Shall I dump the whole lot as unreliable? Thanks!
(Note that PlayOK games cannot any longer be accessed after one year, so the example, being end of 2020, will become inaccessible by the end of 2021)
Daniel Sepczuk at 2021-04-25
On playok there is a time limit for the game, so the defeats were on time
HappyHippo ★ at 2021-04-25
It does seem that playOK lets you resign even when you aren't on move, so it's possible, but I don't know why a winning player would do that.
Miwarre ★ at 2021-04-26
Thanks for the replies.
It is true that in PlayOK (as in most contexts) there can be defeats for time out, but not these: in all these cases, the defeated player was the last to play, so he DID play on time; in case, his opponent could have run out of time, but the opponent actually won.
While I have the attention of two highly experienced Hex players, what would you suggest as a “policy” for collecting games? Ignore P.OK completely and stick to LG (+ the games of the best players on BGA)? Ignore P.OK games with no swap rule, with these inconsistencies and perhaps played by guest players and keep the (few) remaining (all these conditions can be detected programmatically)?
Hex games on LG are probably of the best quality, but high level playing often runs above the head of learners, so a mix can be useful. Also keeping in mind that collecting BGA games, AFAIK, can only be a manual, slow and very tedious work (with a daily limit), while P.OK can be 'mined' mostly automatically.
Thanks for any thought!
HappyHippo ★ at 2021-04-26
What's the plan with these games? If you're doing stats of some sort, yeah you'll want to remove the problematic ones. For my database I removed games that ended on timeout, unless I could manually prove who the winner was. But if you're just studying them, I don't see why you'd need to remove these.
Miwarre ★ at 2021-04-27
Good question, HappyHippo! To which I do not have a definitive reply (yet?).
After decades (I frequented board games, mostly Go and Draughts, when I was young), I re-discovered Hex since a few months and I realised how much more it deserves to be known and played (I am very surprised that there is no 'official' Championship of any sort). So, several projects are popping in the back of my mind, from learning to analysing to developing tools. Most of them require (or can use) a game DB, so I started collecting games, noticing idiosyncrasies and inconsistencies and having doubts and questions.
As the primary goal is not set (yet?) or as there will probably be several main goals, I will probably end up keeping everything in a big, 'dirty', DB to be filtered for each specific purpose.
Thanks for the comments: discussing is always useful to clarify one's mind!
Tom Ace at 2021-04-27
Kevin Gorman's database is still online at http://hex.kosmanor.com/hex-bin/board