Home | About | Tournament Winning Decklists | Forums

OCTGN Matchup Analysis, Part 3

Originally published at: http://stimhack.com/octgn-matchup-analysis-part-3/

Discuss the latest StimHack article here.

Weird, it’s not on the main page

Huh, I see it on the main page.

Looks good! While I was able to see some of these trends while looking at them with the other data posted, it’s nice to see them spelled out. I look very forward to seeing what other data you put up (and the day when decks like GRNDL Supermodernism or TWIY get enough plays to be doable. Someday…).

If I’m not mistaken, in the Jinteki section, the big picture is showing the HB: Etf picture in stead of Jinteki

1 Like

Fixed. Thanks for pointing that out!

Unfortunately OCTGN data is not accurate because there have been numerous times people have quit before I have a chance to either score my last agenda as the corp or finish off a winning run as the runner. OCTGN will not let me finish the game once there is only 1 player left. This literally just happened right now when I rezzed a SanSan at 6 pts. and the runner posted GG and left the game before the agenda could even hit the table. Also there have been numerous games where I’m coasting to a victory and the game disconnects and stops working for my opponent when they come back.

That is totally a reason not to do any more analysis.


There are two kinds of disconnects: intentional and unintentional. There is no particular reason why unintentional disconnects would affect you – or anyone else – only when winning. I think it is extremely unlikely that unintentional disconnects bear any relationship to game state, in which case losing the random subset of games where there was an unintentional disconnect doesn’t affect the integrity of the remaining data.

As to intentional disconnects, what bias do you suggest that they introduce? If intentional disconnects occur when one player is clearly going to win, then a disconnect artificially depresses the winrate of the “winning” player, their ID, and their faction. But for this to introduce an overall bias, there would need to be some pattern to when people disconnect – that is, people would need to disconnect more against particular factions or IDs. For instance, if you join a game, see your opponent is playing Andromeda, and immediately disconnect. If this were indeed the pattern of disconnects, the net effect would be to slightly suppress the winrate of the more successful IDs and slightly exaggerate the winrate of the less successful IDs.

However, there are only 8,306 games (of 211,854) where either player conceded. I actually removed these games from my analysis because it was simpler than assessing the “legitimacy” of each win by concession. Unless intentional disconnects comprise a much, much larger share of games played than concessions do – which I doubt – then I think intentional disconnects are unlikely to leave significant biases in the remaining data.

That said, if enough OCTGN players want to spend some time recording the frequency of intentional disconnects, we could compare the rate to the concession rate as a check. I’d have to look at the monthly active players to get some idea of the required sample size.


Also, with regard to disconnects, as long as the frequency is fairly stable over time, the conclusions regarding change from one datapack to the next should not be affected even if the accuracy of the absolute win rates is slightly diminished.