First of all, could you please stop referring to abandoning a lost game as skipping. For the purposes of this thread, a skip only refers to quitting a game at the stats, based on the name of the opposing god (as pointed out by Keolino) and I wouldn't want people to get confused.
What Zawadx meant, wast that a god with a high amount of losses will end up marked as a skip in the recommendation, so the duration of those games doesn't matter afterwards, because they are set to 1 turn (or whatever turns to skip is set at).
Though, as pointed out by CuCN, the turns to lose does influence the place where it ends up in the skip order, so it might change the recommendation in the first place.
All right, now that that is clarified:
Yes, your point is valid, factoring in turns to lose would increase the accuracy of the tests, IF we can figure out a meaningful reduce tester bias from the TTL recorded.
Your point is that winrate and turns to win are not objective either, so there is no reason record those and NOT TTL. But I do disagree with you here: to get the most objective measurement possible, one should try to reduce the subjectivity in the data wherever possible. Currently there isn't a more objective way to get more accurate winrates are TTWs aside from getting much more testers playing a lot more games per deck (I will get back to this later).
Turns to Lose on the other hand is much more dependent on play style, as pointed out already. The only way to get accurate data on this would be restrict the style of all testers, to ensure the data from the different decks and different testers is accurately comparable, which raises a few other problems:
First, what would be the 'correct' style. Playing until 0 hps, or until you 'know' you are going to lose, there is good arguments for both: the former because it's less subjective, the second because it more accurately represents the behavior someone farming.
Second, how to classify that style. Playing until 0 hps can be influenced my making 'mistakes' during play, like forgetting to play the next sundial/SoSac with PDials. Playing until you know the match is lost is even more questionable, because how would 'lost' be classified: until you are absolutely certain you can't win (aka, a miracle played with not enough damage out until deckout to try again), note, this hardly ever happens, at least for most decks. So you want to classify how to be 'reasonably' certain, after which subjectivity from testers comes in again, and much larger subjectivity than in the TTW.
The real problem though, is what Keolino pointed out last: I want to make the data gathering as easy and unrestricted as possible, to (hopefully) maximize the number of testers and tested games, which should help reduce tester bias, and yes, that would include reducing errors in TTL tracking. But, the way I see it, the real problem with any restriction imposed during testing will likely put of potential testers.
The way I have the study currently set up (or at least tried to), is that someone farming can take stats as a side activity, without having to change their preferred play style or skiplist, and the data is completely usable. The only limitation is that a deck or modification needs 3 played games against each god to get they own separate listing, but stats without those 3 played games are still included in the overall data and count towards the stats of the modification and deckgroup, the modification (or deckgroup) is simply hidden until it meets those requirements.
Damn, playing melancholy, I'm hatin' it. I'll finish the tests (3 per god, I'll do tons of skipping in order to test the ones I lack) and then forget about the deck. Never, ever gonna play it. It's not even a valid deck for a FG prediction, I'll just delete the code.
Nothing against the creator of the deck, it's only that I hate decks THAT susceptible to starting hands.
To help you out games required to complete the data set for Melancholy: Decay (3), Eternal Phoenix (1), Ferox (1), Fire Queen (3), Gemini (2), Incarnate (1), Jezebel (1), Morte (1), Octane (1), Osiris (2).
