FG Efficiency Study - Applying statistics to all of the myths

[Sevs]

Wow, Instosis is still way higher than any of the old decks even without skips ... now imagine it with skips ... :O

Well ... three gods in fact already have their skips. Then it's the deck-creator himself who has played those
games and I am guessing he didn't "play out till 95% sure to quit" most of those actually hopeless matches as we have done here.
Lastly, 200 games still imply an open ending ... in either direction.
Just give a kirch a go at the deck and he will surely put the nail in that coffin. 

Naa but seriously, the deck is terror ... living proof that the age of heavy control FG-decks is over.

That is true but as a result of 300 games before these stats I have found that the winrate of those gods is less than 5% so starting the game on its own implies a >95% chance of losing

[Polari]

All un- or partially upped decks other than the lance deck are unaffected as are upped LA, Voodoo decks and Instosis as the new champion. To me that sounds like enough decks to form a solid baseline. The shard decks that are still worth playing can be reintroduced as we get updated results for them.

As far as I can see that would be the Rol/Hope variations ... without the shards. 
CCYB might be able to compensate a bit by just switching a couple shards over to sanctuaries.

Ghostal definitely works with 0 shards, Mono-AEther takes a hit but I'm already planning to test it with SoSas, and as you said, rainbows can soften the impact with sanctuaries. SoSas could actually work in pretty much any deck that used SoGs as more of a way of buying time for their main plan (like various dune scorp decks) instead of looking to establish total control with them.

[Jangoo]

I added the rest of the decks to the top-posts and the skip-stats/farming-guide is now complete:

- Top12 upped decks with skip stats
- Top2 partially upped
- Top2 unupped

I also added a "legend" (red colouring of FGei-total), so those decks tainted by the SoG-nerf are now easy to identify.


Outdated or not, it is quite interesting to see what the skips did to the decks performance:

Each upped deck gained from 25%-44% FGei by implementing skips and the outcome is
that more or less all the upped decks perform similarly well when permitted to play to their strength.
Funny enough, Fast-Draw-Ghostal, being omnipotent against many gods, only gained some 15% and
has the largest set of gods that barely didn't make the list of playable gods (yellow gods).
The breach from Top1 (Mono-Aether at 7229) to Top10 (Perpetual Light at 5846) isn't all that great if you ask
me, so that must have been an outstanding job in selecting those decks eligible to be tested in the first place. 

Now the partially upped and unupped decks gained largely from implementing the skips and they also have
the very most skip-gods. Whereas the upped decks skip around 9-13 gods, these decks flat out say no
to more than half the gods and get performance gains as large as 60%!
My personal winners here are Flay'em and unupped Liquid Antimatter who can make you ~4600 electrum/h
when skipping the right gods.

Fun Facts

There is just one deck tested where it is advisable/profitable to play against Hermes Graviton defends his position as sucky god by having only two decks advised to play him Incarnate is the most soft-hearted FalseGod: Just one deck is advised to skip him

[Higurashi]

The breach from Top1 (Mono-Aether at 7229) to Top10 (Perpetual Light at 5846) isn't all that great if you ask
me, so that must have been an outstanding job in selecting those decks eligible to be tested in the first place. 

Especially considering two things: I've found the time per game way too high for Perpetual Light from the start. Compare with Dragon RoL/Hope, and consider that Angels end a game faster in my experience.
And secondly, I updated Perpetual Light with 1.29. Removed the SoG 's, as it definitely seemed more profitable through gained speed.

I look forward to seeing how big a hit MA has taken from the SoG nerf. In my experience so far, it hasn't been all that bad.

[Jangoo]

Yeah, I find that the skips worked out pretty well concering the time/game for all the decks.
When everybody was claiming to play the games "much faster" with a certain deck before,
this was probably due to the subjectively perceived speed when playing the deck like you
should, meaning, when not fiddling around with hopeless matchups and crappy draws and
just skipping those parts right away.
Skips take down the win-rate in any case but boost your time/game from around 25% (e.g.
Fast-draw) to as much as 44%! (Mono-Aether) ... time is of the essence for an efficient deck.

Perpetual light came down from a bummer 3:19 min to 2:23 min, which may be realistic but
still doesn't quite compare to Turbospeeds 1:44 min or Ray of Lulz' 1:55 min.
So, according to the data, one cannot really say that Angels finish games faster than dragons
or crusaders. With ~5% headstart on the other two decks in win-rate though, the Angel-deck
relatively bets on the second efficiency-condition, not on speed.
To be really a 100% sure of that, Jenkar and kirchj33 would have to play Perpet-light and RoLulz
to supplement Onizukas games and create actually equal data-sets though. 

Fun Facts

With 1:21 min/game, Shak'ars Revenge is the fastest FG-deck tested With 4:17 min/game, IgotP-time is the slowest 1:33 min/game and 1:36 min/game, make Mono-Aether and Liquid Antimatter come in second and third Fast-draw-Ghostal takes 2:58 min and has a unique standing by taking almost twice as long as any other deck in the Top6

[Higurashi]

Obviously the reason I commented on Perpetual Light's speed is because we have only had one tester submit results, and that those aren't in line with my months of experience. The higher winrate makes sense, but it's due to the increased speed of damage in a limited amount of turns (which is enough to kill an FG). If that's true, it should also be faster in time/game.

[The_Mormegil]

Obviously the reason I commented on Perpetual Light's speed is because we have only had one tester submit results, and that those aren't in line with my months of experience. The higher winrate makes sense, but it's due to the increased speed of damage in a limited amount of turns (which is enough to kill an FG). If that's true, it should also be faster in time/game.

It's possible that, while it wins faster, it loses way longer. For instance, it loses by deckout far more than a deck with Dragons (of course), and that kind of loss is very time-consuming.

[Higurashi]

Good point. If you're not doing enough damage early enough, you'll deck out more against Bond FG's. Testers didn't quit early in the first batch, if I recall correctly. That makes the results very inaccurate then :/

[gumbeh]

Funny enough, Fast-Draw-Ghostal, being omnipotent against many gods, only gained some 15% and
has the largest set of gods that barely didn't make the list of playable gods (yellow gods).
The breach from Top1 (Mono-Aether at 7229) to Top10 (Perpetual Light at 5846) isn't all that great if you ask
me, so that must have been an outstanding job in selecting those decks eligible to be tested in the first place. 

Oh, thank the Great Biscuit in the Sky! For awhile, I had been repeatedly, repeatedly, repeatedly warning people that we would encounter such an effect (specialists rocketing up in performance with skips, and generalists gaining little). I gave up, though, when it was obvious that popular sentiment was entrenched in saying "Yep, Fast-Draw Ghostal is clearly da BEST!" At least I was able to commiserate with Kirch regarding lower-than-advertised effectiveness with Ghostal.

Whoever was crowing the praises of Ghostal, I don't remember your names anymore but dammit, folks, I TOLD YOU SO!

Among the older contenders (not counting Instosis and other new faces), I'd be happy to test Liquid Antimatter again against today's slightly different FGs, or take Mono-Aether for a spin if re-assessing it is a priority. I have the cards for Instosis, but I'm sure other people are eager to test it so I'll leave them to it.

[Jangoo]

Obviously the reason I commented on Perpetual Light's speed is because we have only had one tester submit results, and that those aren't in line with my months of experience.

Good point. If you're not doing enough damage early enough, you'll deck out more against Bond FG's. Testers didn't quit early in the first batch, if I recall correctly. That makes the results very inaccurate then :/

Aaww, don't say that Higu or you gonna make me cry. But:

The case of Perpetual light is very interesting concerning study-design, deck-proposition (specialists vs generalists) and skip-implementation, so let me go into detail:

First of all, I agree with you for the most part since you are referring to the three main critical issues with the studys test-design here.
The first two of them have been explained in detail HERE (http://elementscommunity.org/forum/index.php/topic,25609.msg393369#msg393369).
At least for me, the third issue became more and more evident only as the study progressed. Here is a quick roundup:

1. The darn "95% sure" and the unfortunate games against doable/non-skip-gods that stay in the sample even after skips
Testers were supposed to quit a match when being 95% sure they won't win it anymore from day one.
As far as I know, there hasn't been a first batch where nobody did this ... but perhaps the interpretation
of it has indeed shifted as the study went along?
Either way, there are various obvious cases that players just shouldn't fail to play to:
Realizing when even an entire field of Archangels/Dragons won't ever beat FQs/Elidnis' bond-healing
should be one of those cases. You see that one coming from miles away ... but who knows if the testers actually did?

2. The type of testers
This whole know-when-to-quit-policy is of course quite shaky and, in its practical use by the testers, most
likely always far from what the deck-creators themselves would like to see ... for each and every deck tested.
As explained in that post linked above, this study wasn't designed as a 100% highperformance study.
If that were the case, you could just have each deck-creator submit his top-stats in a couple hundred
games and you are done. (But we pretty much already have that with everybody claiming their deck to be the
bomb and boasting 120% winrates in their deck-threads, don't we?  )
I think it is a given that pretty much nobody will live up to the standards of the deck-creators themselves but
Onizuka certainly is a good player. The question is now ... are Jenkar and kirchj33 even "better"? 

3. Number/Type of testers and spread throughout the decklist
Now the real critical point is when "better" players sort of "race" against "worse" players on different decks.
This is even more so the case when the "different" decks are comparable, e.g. both are Rol/Hope variations.
I guess we didn't really anticipate how much difference there will between the various players performance.
And we also expected to gather a bit more games and to get at least two guys on each deck.
Having to make due with who is willing to test what for how many games, in this study, unfortunately lead
to some decks being tested by only one guy, and others by three, some decks by "better" players
and others by "worse" players, some decks getting a nice 800 games and others only 150 ...

While personally I still think 1. and 2. are justified and worked out well, 3. is definitely a must to improve.

---------------

Now for the really interesting part: Why didn't Perpetual light perform as well as Turbospeed?

There are three ways to explain this, all of which probably bear truth:

1. The testers weren't as good

See above for that part and:
With Jenkar having been a king-of-the-hill player, it is quite possible he boosted Turbospeed.

Concerning the deckspeeds, I also drew some numbers from the statsheets for the three Rol/Hope-decks:
Mind that these numbers may of course imply either or:
Either the deck just is slower/faster or it were the players who made it that way ... who knows ...

Perpetual Light:
Average loss time against FQ: - (straight 100% win-rate over 11 games  )
Average loss time against Elidnis: 4.02 min   (Turbospeed: 3.85 min, Ray of Lulz: just one loss so it would be ridiculous to quote that here ...)

Average win-time
Turbospeed: 3.75 min
RayofLulz: 4.27 min
Perpetual: 4.24 min

Average loss-time (lost games against non-skip-gods / all lost games: skipped and non-skipped)
Turbospeed: 2.05 min / .65  min
RayofLulz: 2.42 min / .76  min
Perpetual: 2.42 min / 1.12  min

So, yeah, Turbospeed came in quite a bit faster in all instances, especially in its loss-time just like Mormegil suggested!
Note that those Ray of Lulz-games are also done by just Onizuka, so to some degree it has got to be
something about the way the Perpetual deck works as such.
I don't find that loss-time against Elidnis as a bond-god to be that much different for the two decks though,
so either everybody blew quitting against bond-gods on both decks or that's just how long it takes. 
That very last number (1.12min loss time for Perpet) on the one hand supports Mormegils theory, on the
other hand, it takes us to the explanation number 2 and 3:


2. Skip-implementation and small samples

The exact procedure I used for the skips is explained HERE (http://elementscommunity.org/forum/index.php/topic,25609.msg440843#msg440843).
The quintessence, showing a little bit in the case at hand, is that decks with smaller samples don't profit as much from skip-implementation as decks with large samples. The reason for this is that it is very hard to make a good call when there is only a few games played against a god:
Just a single game goes lucky or not and the gods fate is sealed.

For example, Perpetual light ran in an FGei(c) of ~4880 over 5 games against Obliterator with 2 games won.
At an overall FGei(cn) of 4602 for the deck definitely a god you should play right?
Now imagine one of those 2 wins barely went bad for whatever reason: Booom -> FGei(c) of ~2200 and Obliterator is a clean skip-god ...

Because small samples just aren't very reliable, gods with few games get somewhat lower thresholds to stay in as playable because
you never know if those crappy couple games were just plain bad luck. So for decks with large samples, like Turbospeed, it has been
a clean 80%-or-bust call in, I think, all cases.

Now to think that small-sample decks got entirely dragged down by loads of substandard gods who stayed in would be wrong.
In fact, even here the choice was mostly very obvious and it has only been about one "stayin"-god per small-sample-deck as with Perpetual light:

Perpetual light:
FGei(cn): 4602
80% threshold: 3682  (15+ games)
70% threshold: 3221  (10+ games)
60% threshold: 2761  (5+ games)
Rainbow: 3452

So with only 7 games played, Rainbow stayed in. In fact, he would have only had to make the 2761 to get a chance ...
Fyi, if Rainbow were declared a skip-god the skip FGei would go from 5846 to 5934.
When comparing the two decks, you could also look at this inversely and consider the three yellow gods Turbospeed has:
Yellow here means they made the 70% theshold but over as many as 20-41 games they'd just have to make 80% to stay in.
Now picture Turbospeed had only been played over some 300 games and those gods had still made 70% ... they'd have stayed
in and Turbospeeds FGei would be notably lower.

Small samples just kinda suck for the skip-decisions or vice versa: Large samples permit very clear and more effectful decisions.


3. Perpetual light is relatively a generalist or: The deck simply isn't as good

Honestly, I have played a lot of Rol/Hope a longer time ago and I have fiddled with Angels and Nymphs
(Crusaders weren't out yet) and what can I say: Dragons get the job done best imho.

But personal impressions and the "good deck - bad deck" theory aside, here are some more numbers  :

Skipped gods:
Turbospeed: 13
RayofLulz: 12
Perpetual: 9

As explained under 2., except for Rainbow, this spread is not due to the way the skip-gods were
chosen at the small-sample-decks disadvantage. Perpetual light simply has a much broader proposition.
It is, relatively, a generalist which is also reflected in its somewhat higher win-rate.
If you can't skip as many gods because they are all somehow viable = if you don't have many peak-gods
and rockbottom-gods, then skipping won't give you as much a boost of course.
This, in two ways, also explains the 1.12 min taken for each loss, even with skips:
If you play against each and every joe in town, then you wind up spending more time on duking it out
until you finally lose indeed because your win-condition simply isn't as pointed as with more specialized decks:
 
- A specialists straight skip equals 1 second of lost time
- A generalists failed attempt to win another mediocre matchup is always much more

Yes, the time to lose could be reduced if the player is highly experienced and knows exactly when
to quit against all those mediocre gods ... but it's never going to be 1 second ... ever.
It's not only faster but also easier to make the quit-call with a specialized deck, which, somehow,
makes the "easy" deck better ... at least if you want to recommend it as a grinder for the general public.

Fun Facts

The most impressive FGei(c) was brought in by Liquid Antimatter: A roaring 15032 elec/h against Akebono Second and third are Turbospeed Rol/Hope with 14357 against Divine Glory and Liquid Antimatter again with 13378 against Obliterator The highest per-god score performance has Liquid Antimatter: 2202 score/h against Obliterator Second and third are Liquid Antimatter with 1891 against Akebono and Mono-Aether with 1764 against Neptune

[JonathanCrazyJ]

sooo... when's instosis gonna get on the list?

EDIT: I see it has stats there, just isnt in the OP, im a noob lol.

[Sevs]

sooo... when's instosis gonna get on the list?

EDIT: I see it has stats there, just isnt in the OP, im a noob lol.

1) cause it might get nerfed.
2) I didnt follow their policy of no skipping. I might redo those 3 skips to follow the guidlines.
3) It would help reinforce the stats if someone else adds 100 games or more