If you were to run that high an amount of extra land cards - would it be an idea to maybe go by the idea many pages ago with dredge knight? - having lightning axe to discard lands - life from the loam to get them back?
Maybe, but Knight is pretty slow as is, slowing it down further with Loam means you don't want the red cards that provide aggression, which pushes you back to Bant.
The more I use it, the more I'm liking Blighted Gorge. I won a round against Affinity today because I was able to Gorge an Etched Champion. I had no cards in hand so my opponent moved in on a Ravager to pump him up and end the game, and in response I tutored up Gorge and killed it.
I really love his way of thinking! What i'm now really really interested in, is how are you currently building your deck using all this data?
What does your deck look like?´
How many lands are you running?
What creatures and how many?
I have several updates planned to this approach, each deck requires it's own simulator to run because the strategy (basically just the algorithm for what casting priority cards have) changes by cards included and goal. Right now I'm writing one in a different style for Legacy Burn (I'm trying to empirically answer the question of Sensei's Divining Tops or not, and if so how many) which lets me make several enhancements, most notably results going to a database rather than a spreadsheet (and ultimately a web form for anyone to easily query results), the ability to have some minor interaction from opponents (blockers, removal spells, maybe counterspells... think a generic midrange deck or a control deck, aggro vs aggro is oddly enough the hardest to do), alongside changes to manabase construction for fetches vs non fetches. Those enhancements will all be moved over to the program for this one as well but it takes time.
My Burn one will probably take another 1.5 weeks depending on how much time I spend on it, then time to run games (though I can do that part while writing the other one), so we're talking probably 1-2 months to make the updates to the Knightfall program that I wish to make.
Anyways on to the deck, I just played a tournament Saturday, I don't get many opportunities to play Modern lately. I did absolutely horrible in it but that was due to making bad plays rather than most deck choices themselves being bad. Here's the list I went with, I'll save the tournament report because it was extremely bad (we're talking potentially the worst I've ever done at an event in several years), but I will comment on some card choices and what I learned. Note that I played 61, the 61st is probably Pillar of Flame.
So here's what I learned on the various card choices. I went with 23 lands+4 dorks because it lead to a higher number of fast starts, going purely by the numbers I posted above the fastest decks went all the way up to 28 lands with 0 dorks, but 4 dorks over 4 lands was only about 0.1 turns slower while having a healthier mix of combo and non combo wins, and I thought the versatility was worth the speed. Additionally, I couldn't think of another 4 utility lands I wanted. Hierarch makes colored mana which is important but the utility lands were colorless... I think the higher land count with fewer dorks is more defensible in 3 color land. With 4 colors you just can't afford for any of your first 3 land drops to be colorless. The low number of bears is another reason I used 4 dorks, I was already skipping 2 drops so I needed 3's to cast. Anyways onto specific cards:
1. As the simulator suggested, Blighted Gorge was fantastic. Tutoring removal is great and mana acceleration or Knight combo mana is just enough to get you to activations. Gorge won me 2 games (once tutored, once drawn) and could have won me another had I sequenced my turn better. I think this is a 4 color only option, Naya tries to close the game out too fast and Bant can't pay the red. Similar logic should hold true for a version using black with Blighted Fen.
2. Cathedral of War was horrible, I was experimenting with various land toolbox cards and unlike Gorge this one didn't work out at all. The manabase can handle the second colorless land but this isn't the one you want it to be. Kessig is just so much better, and any option needs to not come into play tapped.
3. Stirring Wildwood and Horizon Canopy rounded out the land toolbox and were fantastic.
4. The colored mana sources didn't work. I took way too much damage from my lands (especially mid combo), and the colors didn't come together. I needed more fetches because twice in 6 rounds my land drops demanded I get a Steam Vents which isn't in there, and not being a Forest/Plains is a bad include anyways. I also felt like I needed more basics, and more fetches. I'm not sure if this is an excuse to goto 3 colors, to add more lands up to the suggested 28ish, or both. More fetches or Ghost Quarter seem important though, Knight just wasn't getting fed well enough.
5. Meddling Mage was better than Voice in the field I faced, I wished the numbers were reversed. A split is definitely right though so that Collected Company has the potential to hit what you need.
6. I'm rethinking the whole Retreat plan, I only comboed once and it honestly wasn't even all that good when I did. Every other time my life total was too pressured to combo.
7. Celestial Flare is something of a pet card, when I played against Bogles round 2 and got a free game 1 win off of it I was happy it was there. My opponent actually got a judge to deck check me on it he was so surprised. That's why I like the card though, no one ever expects it and it is by far the best edict in the format. Plus, once you show it a single time it's a very easy card to bluff in future games which makes it the perfect game 1 card.
Now for the sideboard
8. Retreat to Coralhelm was a waste of a slot, I never once brought in the third copy. In hindsight I'm not sure where I ever would. Wanting the extra copy would imply my opponent isn't pressuring my life total and doesn't have some way to interact with a Knight. If that's what's happening, why not just have another threat instead?
9. Worship was... interesting. It has a great interaction with Geist of Saint Traft and while it prevented me from losing several games it didn't actually provide me with an opening to win the game. I drew it in every round I brought it in, and every time I played it (which was most games as I drew it a lot) the game went to a draw.
10. Destructive Revelry should have been a Reclamation Sage, I just didn't remember that card existed until the middle of the event.
Ok, it's time for more spreadsheet spam from my simulator. I took a bit of a break before posting these results. The were compiled over 10,000 games per deck, play/draw, and nearly 41,200 total decks analyzed for just over 820 million goldfished games analyzed, hopefully it gives some insights. Links are screenshots of the top 25 decks in each category. If anyone wants to look at them in spreadsheet form for their own information here's the raw data, just open with a spreadsheet and separate columns by tabs.
Draw https://www.dropbox.com/s/tflyunjsslh2zpf/KnightfallWide10000gamesDraw.txt?dl=0
Play https://www.dropbox.com/s/02o0xyn0xz77n6h/KnightfallWide10000gamesPlay.txt?dl=0
I did something different with this data and included in each deck between 5 and 6 extra removal spells. The simulator never actually cast these since there wasn't a target to do so, but they do still count as cards in deck construction. They're labeled in my data as Path but it could just as easily be Condemn, Oust, Celestial Flare, or anything else, so between these and burn spells you're looking at 9-12 removal spells per deck, which should be somewhat accurate.
The first category I want to show is the fastest win on the play. For those who didn't read my previous posts on the subject, the reason I'm posing the top several decks rather than the top deck is because variance is a thing in Magic and it's important to look at the similarities/differences of several top lists. http://imgur.com/DARTkYv
Because of my sort parameters on these, the 5 path decks will naturally be faster than the 6 path decks which should only make sense. I'll save the spam and not post the top 25 with 6 paths but instead I'll just write down the results. On the play the average win turn goes up by 0.035 turns and the cut is a bear. On the draw the average win turn goes up by 0.04 turns and the cut is a Retreat to Coralhelm.
Next I want to talk about what deck combos early the most often. Nothing is able to turn 3 a consistent amount, with the highest on the play combo being 509 times out of 10,000 games but there are decks that win early and often. For this I'm going to use the sum of the decks turn 3 and turn 4 wins. These decks are slightly slower than the fastest decks but not by much. They have a high rate of comboing to win the game (around 29%) but the turn 3's don't show up too often, it's mainly a list of what does the best on turn 4.
Play http://imgur.com/4qoIcjM
Overall, I think it's most useful to look at the fastest decks and then to look at the decks most likely to combo if it's a bad matchup. A good game 1 configuration probably involves maximizing for the play and games 2/3 involve whatever you'll be on in that game.
Last, I want to talk about land counts. If you look at these results, alongside my previous results the first thing that will jump out at you is that the counts are very high. I've checked and double checked why they're so much higher than the traditional amounts and I'm pretty sure at this point that it's not a flaw in my logic, it's just something revealed through testing. I'm not really sure how I lean on that right now. In the play data, you can drop 4 lands for 4 mana dorks and see about a 0.1 turn slowdown in your games, but the starts are much faster and they appear to be less reliant on the combo.
Here for example are the top 25 fastest decks on the play with 4 mana dorks included. http://imgur.com/L9Mey6K
What this makes me think more than anything is that utility lands are better than we usually give them credit for being. Horizon Canopy seems way too good to pass up now, and if you have red even Blighted Gorge looks pretty solid given how much this data suggests burn spells are what you want.
My main thought is that a deck like this is poise
Hi all there's a very interesting article on modern nexus right now. It goes through how eldrazi aren't an unbeatable deck and they use knightfall as an example combo deck to beat it. This was just a brief summary and I would suggest everyone read it and come back with comments and suggestions. Here's the link http://modernnexus.com/seeking-company-broken-metagame/
I had similar thoughts which is why I'm now playing the deck. Where you want to be in the format (and this was true pre ban as well) is aggressive with a combo win.
Anyone should be able to download that (unless Dropbox gets mad at me), open it up with their spreadsheet program of choice, delineate the columns with tabs, and have all of the results. Then you can sort by whatever metrics you wish. I will note that you'll have to make one extra column, the WinTurn column gives a total of the wins so in order to get the average you need to divide it by the total number of games, 10000 in this case. It's a small annoyance but one that had to be done since my program wasn't printing the results correctly otherwise (and it also saves a small amount of CPU power while running the simulation). There's also a handful of decks that are repeated twice in there (bonus points if you can find them... they're somewhere in the 26 land decks), this is because my computer froze halfway through the simulation and then again a few minutes later once it started running and I had to restart it at that point.
Anyways, this one took approximately 48 hours to run and is on the play. I'm doing another now as I type this which will be on the draw which I suspect will give very different results. A few days ago I did a smaller run to compare play/draw and the results were interesting (those were posted above). This current run is with Path to Exile included along with a couple other removal spells. My goal was to try and figure out what the correct cuts are for more removal without sacrificing speed. I went very wide in looking for results here which is why it's taking so long to generate answers.
The possible deck pool was:
Land - 20 to 30
Manadork - 0 to 4
Nacatl - 4 to 4
Bears - 0 to 6
Goyfs - 4 to 4
Geist - 0 to 4
Retreat - 2 to 4
CoCo - 2 to 4
Bolt - 4 to 4
Shock - 0 to 2
Path - 5 to 6
You'll notice a few cards which were set to 4 as a constant and some others that I made 2-4 of's rather than testing the full range. That was a time saver at the possible cost of some accuracy, if I hadn't done that the run time would have been 6750 days (18.5 years) compared to 2 days.
In the end 41,149 decks were tested which means 411.5 million games of Magic systematically analyzed. As expected the 5 path decks were faster than the 6 path decks (which is just logically correct so helps prove the simulator is running properly). I won't post screenshots this time, I think it's a bit redundant with me simply posting the full data (if you don't know how to work a spreadsheet but want to look at it send me a PM and I'll explain, it's pretty easy to do) but I will post the fastest deck of each for some comparison, plus with such a large dataset I don't think just the top 25 lists do it justice. These tests are on the play, previous results suggest that on the draw you want to combo more, cut lands, and bring in burn. Essentially the strategy on the play is to win through board presence while the strategy on the draw is to use the combat step less.
I'm very surprised to see the land counts being what they are. I would have thought something along the lines of 26 lands with 2 manadorks would work better than 28 lands but those configurations are very far down the list. Here's the highest performing 26 land/2 dork list.
26 Land
2 Manadork
4 Nacatl
3 Bears
4 Goyf
3 Geist
4 Knight
3 Retreat
2 CoCo
4 Bolt
5 Path
This configuration did have a 1.5% turn 3 win rate, but the average game was 5.2002 which is a good deal slower than the other options at 5.1201 and 5.1564.
One more interesting list is the best performing turn 3 list. I think this is a metric worth looking into further because in bad matchups where the game plan may be to combo as quickly as possible as our method of racing, this is a configuration we should keep in mind. The fastest turn 3 configuration was able to turn 3 in 509/10000 games or 5%. It is
25 Land
4 Manadork
4 Nacatl
2 Bears
4 Goyfs
0 Geist
4 Knight
4 Retreat
4 CoCo
4 Bolt
5 Path
Obviously going beyond 4 mana dorks would help here, but my constraints set 4 as the maximum. This is a configuration I'll look into optimizing more later on.
I said that would be the last but I have one more I just thought of which is closely related to the turn 3 metric, and that's which deck put together the combo most often (remember that my simulator is capable of playing scry 1 so that aspect of Retreat to put things together is being utilized, and it can play multiple scry 1's in a row... though it has no idea how to play a scry 2+). So here's the deck that comboed the most (keep in mind that it may have comboed the most because it couldn't quickly win on the board).
30 Land
4 Nacatl
1 Bear
4 Goyf
4 Knight
4 Retreat
4 CoCo
4 Bolt
5 Path
The average win turn on this setup was 5.2291 with a combo rate of 3301/10000 games or just over 33%. I suspect this one is variance however because of the 0 manadorks. The next 3 highest combo decks all played a few.
@ Aazadan: I'm concerned that your model is telling us to play way too many lands. I feel like it is demonstrating a systemic problem - I don't have the computer/math skills to offer more though
I have wondered the same thing, I haven't posted most of my results for that very reason. Assuming my model is correct (and I don't see why it wouldn't be... for a goldfish, real games may require more interactivity though) I think that what is really being illustrated here is the power of utility lands, it's very possible that the community as a whole has underrated just how many lands we should run, and just how good utility lands are in order to get spell like effects while hitting land drops. Based on my results I actually just picked up some Horizon Canopys for myself, and I'm even considering grabbing a couple Blighted Gorge (I already have the other Blighted) on the theory that having more burn post board is actually a good plan (pure speculation on my part, Blighted Gorge is going deep but I do love a land toolbox). Also keep in mind that I haven't calculated the correct total with fetchlands which should in theory change it by 1 land or so. Calculating fetchlands is on my list of things to do but I need a narrow range of other cards to test first.
Also, keep in mind that what's optimal in a goldfish isn't necessarily optimal elsewhere. That's why I added removal even though it makes a pure goldfish worse. I want to see just how much things slow down by adding those cards. Anything that reduces speed by less than the value the extra card gives should be worth it, but that won't necessarily show up in a speed test.
What I find interesting here is that on the draw there are considerably more turn 3 wins though the number of combo wins remains fairly steady. The general strategy for an optimal kill speed looks to be cutting a couple of lands and increasing threats. If nothing else, there's probably a valid strategy in cutting a land for an additional CoCo or good 2 drop in sideboard games when you're on the draw.
I have some more results from my simulator. Tried to do a really indepth 100,000 game run with wide parameters which had a several day runtime and about 80% through the test I was looking at the results and they didn't match expectations, so I went to look at the code and found a pretty major logic bug with Retreat to Coralhelm, which was causing it to be under represented in decks.
As a result I fixed the bug and ran a shorter 1500 game test on the play, and will have another one on the draw later tonight. And after those results I'm going to add in some extra removal that slows the deck but gives more interaction.
Screenshots of the best decks will have to wait until later tonight where I have access to a computer with a high enough resolution to take them but for now I can report. The results are starting to look rather in line with what has been seen in well performing lists.
4804 decklists were analyzed, the fastest deck has a win turn of 4.948 (that previous bug really sped things up) and the most common numbers of each card among the top lists are
Lands - All over the place, 25-29, with 27 being most common
SmallOne (Manadorks) - 2 in the top lists 2-3 in the top 25 lists
LargeOne (Wild Nacatls) - 4, this was set as a constant so 4 is the only possible value
SmallTwo (Bears) - 3-4, 4 is most common
LargeTwo (Goyf) - 4, this was set as a constant so 4 is the only possible value
Geist (Geist of Saint Traft) - 3-4, 4 is most common
Knight (Knight of the Reliquary) - 4, this was set as a constant so 4 is the only possible value
Retreat (Retreat to Coralhelm) - 2-3, 3 is most common
CoCo (Collected Company) - 3-4, 3 is more common in the top 10, 4 is more common overall in the top 25 or even top 100.
Bolt (Lightning Bolt) - 4, this was set as a constant so 4 is the only possible value
Shock (Shock) - 0-2, it's split pretty much 50/50 but decks are either all or none, very few 1 of's.
So, putting that all together that in my opinion locks the ideal skeleton to something like this (I should note that several of my top performing lists came to 61 rather than 60 cards, though not all of them)
By CMC
Land - 26
1 CMC - 12
2 CMC - 8
3 CMC - 11
4 CMC - 3
By Type
Land - 27
Creatures - 22
Spells - 11
By Role
Mana - 29
Beaters - 19
Spells - 11
Flex - 1
Looking at it this way, the most interesting thing I found is that Collected Company seems to be working optimally with 22 good creatures, which makes me think that shooting for 27 creatures at the cost of removal would be a disadvantage because the fastest lists don't need that many hits, though something like Fiend Hunter/Banisher Priest to be both a creature and removal is probably pretty excellent.
This is so, so, SO cool.
Can't find the topic in the forums, but is this simulator downloadble and how tough is it to add more creatures/cards to the deck?
I'm mostly thinking about, could one drop the geist to 3, add 2 loxodon smiters and drop the bolts for maybe 3 paths? that would ofcourse make it harder to hit 20 easily.
And the SmallTwo, is that just a 2/2 ? or could it be a pridemage or voice maybe buffing us below the 4.0148?
This is the thread that I was referring to, there's not much discussion on it though sadly. Here is the script as it currently sits in my Dropbox, though the exact script that I have running on my spare computer right now is slightly different because the parameters were changed. It's easy enough, just have python 3.5 installed and run it via command line or use an IDE like PyCharm or PyScripter and run it there. For reference my runtime for the above with a 2.9 ghz quad core cpu and 32 gigs of ram (not that it was ever using more than 4... it was probably running on just one cpu core too) was 15 hours. My goal is to keep the runtime around 10 hours per iteration on my computer. With the above settings 5220 decks were tested at 2700 games per deck.
Geist could be dropped and replaced with Smiter easily enough, it's just a matter of changing it's power from 6 to 4 and removing the legendary tag. Bolts can become Paths but the optimal deck will run 0. This is because of the current approach which is just a goldfish. Bolt can be used as an interactive spell because in this case it's always being aimed at a persons face, but you could change it to hit a creature at the cost of your clock easily enough.
The SmallTwo is technically a 2/x for 2 mana. It could equally be a Meddling Mage, Grizzly Bear, Pridemage (without Exalted), or Voice of Resurgence, it could even be an Aegis of the Gods, as a goldfish the only thing that really matters is the power and mana cost. Making something more interactive than that is extremely difficult because it gets computationally complex though it is on my to do list. Fortunately, I have a pretty light semester right now so I have considerable time to work on projects like this.
The more complex something is the longer the runtime gets. Adding the logic from my previous version of the simulator to the Knightfall specific version for example dropped my runtime from ~0.2 seconds for 10,000 games to ~2.0 seconds for 1000 games (a 100 fold slowdown). This means that the more detail you add to your card types the higher your runtime goes, which means the fewer results you can obtain in any given time frame so you gain uncertainty due to the very high variance in Magic. On the other hand if you use very generic cards with a large number of games things become overly simplified and while you eliminate variance you add uncertainty in the cards acting in the ways you imagine they will. Ultimately you have to take one form of uncertainty or the other, I'm aiming for something of a middle ground leaning more towards simplicity as that's far easier to implement.
Adding a new card increases runtime between 100% and 500% depending on how many you're running. Which is just simple math to prove/explain: The number of possibilities (not all will run because they don't meet the total card criteria) would be for example 14-30 lands (17 possible land counts), 5 possible geist counts (0, 1, 2, 3, 4, 5), 5 possible bolt counts, and 3 possible retreat counts (0, 1, 2) so that's 17*5*5*3 or 1275 possibilities. Adding a 1 of path is 0-1 possibilities, which makes that 17*5*5*3*2 or 2550. As you can see, it skyrockets very quickly, the possibility of a second 1 of would double it again, or the possibility of a 2 of Path would be a 50% increase. This is why it's stepping in values of 2 in order to find ranges. That constrains a 0-4 to 0, 2, 4 or 3 possibilities. If you do that to the above that gets you 9*3*3*2*2 or 324 possibilities, and then running it a second time for the inbetween values is another 324 for 648 possibilities total rather than 2550. This reduces the base runtime but the exponential increases are still there so it's a very real constraint that always has to be kept in mind, otherwise it spirals out of control from the range of what I can do on a spare desktop PC to something that would need a supercomputer.
Currently I have 11 card types which is very high, reducing that by 2 cards from 11 to 9 would cut the run speed to nearly 1/10 what it is right now (1/9 actually). This is just an unavoidable fact of the computational complexity of Magic which I explained above, it is an extremely complex game. Because of this, making an exact optimal deck isn't actually possible (and the optimal deck would vary based on what you're playing against anyways), instead the goal is to break cards down by role with groups of similar cards and to make a skeleton such as wanting a mana curve of x 1/2/3/4 drops, y ramp, and further subdivided into mana dorks, combo pieces, CoCo, hits for CoCo, and so on. Once that is done, you have to go back to more traditional deck building methods.
Well, as I mentioned a couple pages back I've been working on trying to find a more definitive number of cards to play of each card type in this deck to make something of a more accurate deck skeleton. To this end, I've been building a Magic simulator that can be given a range of cards of each type to play, and then it plays some predetermined number of games with every possible deck configuration and records various metrics on them. I've been pretty happy with the more simplistic simulations (I have a thread on them in the main Modern forum) so I added a few more card types and have begun something a little more complex.
My simulator knows the following 11 cards, and was given the following constraints (you'll note there are some auto 4 of's, this is due to previous results of more simplistic decks indicating this is right, and I could save A LOT of processing time by doing so). Cards are only available in groups of 2, so 0-4 cards means only 0, 2, or 4 cards can be tried. The followup which is running now will then use for example 2, 3, and 4.
Land - 5 color land, 16-30
SmallOne - Birds of Paradise, 0-8
LargeOne - Wild Nacatl, 2-4
SmallTwo - Grizzly Bear, 0-8
LargeTwo - Tarmogoyf, 4
Geist - Geist of Saint Traft, 0-4
Knight - Knight of the Reliquary, 0-4
Retreat - Retreat to Coralhelm, 0-4
CoCo - Collected Company, 0-4
Bolt - Lightning Bolt, 4
Shock - Shock, 0-2
The simulator is programmed to recognize the combo and to prioritize the combo over other lines if it comes up. It also recognizes the legendary rule for Geist, and CoCo has a priority on cards (Knight if you can combo, Geist if you lack one, Knight, Goyf, Nacatl, Bear, Birds). For complexity reasons there are only 2700 games being played with each deck though the next iteration should be able to manage 10,000 of each (more possible decks means fewer games of each deck can be run) since these results narrow things down a bit.
Linked is an image of the top 25 deck lists given these parameters (I would have attached but the forum is claiming my 260k image is over 1 GB and thus is too large to attach). Of particular note is the #1 build which comes in at an average win turn of 4.0148. I am optimistic that with some fine tuning of the card pool and a more tuned mulligan strategy it will be possible to break the turn 4 barrier on average. All games are on the play, once I've discovered an optimal play deck I'm going to attempt to build an optimal draw deck.
I'm going to try to get some more indepth results as well, but my hope is that this gives people a better starting point to building the deck as it's currently all over the place.
I've got a small but high prize local tournament coming up in a few weeks (think like Worlds or the SCG PC but on a much smaller local level) and this deck is definitely on my short list of things to play. It does pretty much everything I'm looking to do in the format which is cast efficient 1 mana removal, edicts, slam Geist, play big creatures, threaten combo wins out of nowhere, and offer an above average ability to play some one of's (I very highly value putting the fear of the unknown 1 of into my opponent in these types of events... and word spreads across the room quickly).
In the videos I've watched I like the way the Zoo variant plays, it's aggressive and gets a lot of extra percentage points from unknown information. For example, if your opponent kills the Goyf they have to risk your KotR killing them, but if they don't kill the Goyf it's going to kill them eventually. I like having diverse threats like that.
When I've tried some sample lists online though before making my own I keep running into a problem that I didn't see in the decks video primers. The manabase is simply costing me games. It feels way too fragile as a 4c manabase but I don't see an alternative. If you want to pressure the opponent Nacatl puts you in red which also then lets you pick up Bolt for free, while Geist and the combo put you in blue. I haven't read this entire thread but I've read about the first 10 pages along with the last 30 and I haven't seen any manabase discussion at all other than on Sejiri Steppe and Kessig Wolf Run. It seems like people are just throwing some lands together and calling it done.
So, on that subject what have people thought about the various land counts? Collected Company really wants around 27 creatures and 3 CoCo (more on that below) brings you to 30 cards. 22 lands makes it 52 which leaves 8 other spells. That setup seems to interfere with the combo though. If there's only 8 other spell slots you have to sacrifice either removal or combo potential. Has anyone been able to work in Ghost Quarters or Horizon Canopy? What fetchland balance is optimal? Obviously some green fetches are good because all of the real T1 plays involve green mana but beyond that it becomes more of a question. Is Wooded Foothills really offering anything over Flooded Strand for example? Arid Mesa and Windswept Heath are clearly the best ones because they gets all 4 colors without having to run a Steam Vents, but is it right to be running say 4 Wooded Foothills over Misty Rainforest? Or even over Scalding Tarn?
Outside of the manabase when I build or tune decks I like figuring out a deck skeleton based on what breakdown for the various slots is optimal. That will take me awhile to decide on for this deck, but in the meantime I have some thoughts on the three drops that might be worth discussing.
With Collected Company, I've seen a lot of lists running 4 but I'm not so sure that's correct. I was playing with a bunch of percentages today in a spreadsheet and basically what I found was that the sweet spot between deck slots invested and consistency was at 3 copies of CoCo attempting to find 2 of's. This invested the minimum number of resources to have a reasonable change of finding the card. It seems to me that a lot of these lists are running more 3 drops than the format really allows you to run, so I think making a few trims here to get a better curve could improve lists as the ~13 3 drops definitely feels like too many.
For example the three/four drops could look like
2 Geist of Saint Traft
3 Loxodon Smiter
3 Knight of the Reliquary
3 Retreat to Coralhelm
3 Collected Company
That cuts 2 slots without appreciably hurting the chance at a combo, which then frees up two slots further down the curve for either more aggression or protection.
To put this in perspective a 2 of with 3 CoCo's is as consistent as a 3 of without any CoCo and a 3 of with 3 CoCo's is as consistent as a 4 of without any. The pattern holds for 1 of's as well. Basically you can think of 3 CoCo as being worth +1 copy of every other creature in the deck. The 4th CoCo can push that further but because of how math works, the second CoCo has less impact than the first, and the third has less than the second, so really that final copy isn't very good. If people wish, I can post the spreadsheet I built on this stuff later.
Private Mod Note
():
Rollback Post to RevisionRollBack
To post a comment, please login or register a new account.
Maybe, but Knight is pretty slow as is, slowing it down further with Loam means you don't want the red cards that provide aggression, which pushes you back to Bant.
The more I use it, the more I'm liking Blighted Gorge. I won a round against Affinity today because I was able to Gorge an Etched Champion. I had no cards in hand so my opponent moved in on a Ravager to pump him up and end the game, and in response I tutored up Gorge and killed it.
I have several updates planned to this approach, each deck requires it's own simulator to run because the strategy (basically just the algorithm for what casting priority cards have) changes by cards included and goal. Right now I'm writing one in a different style for Legacy Burn (I'm trying to empirically answer the question of Sensei's Divining Tops or not, and if so how many) which lets me make several enhancements, most notably results going to a database rather than a spreadsheet (and ultimately a web form for anyone to easily query results), the ability to have some minor interaction from opponents (blockers, removal spells, maybe counterspells... think a generic midrange deck or a control deck, aggro vs aggro is oddly enough the hardest to do), alongside changes to manabase construction for fetches vs non fetches. Those enhancements will all be moved over to the program for this one as well but it takes time.
My Burn one will probably take another 1.5 weeks depending on how much time I spend on it, then time to run games (though I can do that part while writing the other one), so we're talking probably 1-2 months to make the updates to the Knightfall program that I wish to make.
Anyways on to the deck, I just played a tournament Saturday, I don't get many opportunities to play Modern lately. I did absolutely horrible in it but that was due to making bad plays rather than most deck choices themselves being bad. Here's the list I went with, I'll save the tournament report because it was extremely bad (we're talking potentially the worst I've ever done at an event in several years), but I will comment on some card choices and what I learned. Note that I played 61, the 61st is probably Pillar of Flame.
4 Misty Rainforest
4 Windswept Heath
2 Arid Mesa
1 Flooded Strand
1 Hallowed Fountain
2 Temple Garden
1 Sacred Foundry
1 Stomping Ground
1 Breeding Pool
1 Horizon Canopy
1 Blighted Gorge
1 Cathedral of War
1 Stirring Wildwood
1 Forest
1 Plains
4 Noble Hierarch
4 Wild Nacatl
2 Voice of Resurgence
1 Meddling Mage
4 Tarmogoyf
3 Geist of Saint Traft
4 Knight of the Reliquary
Enchantment 2
2 Retreat to Coralhelm
Spells 14
3 Collected Company
4 Lightning Bolt
1 Pillar of Flame
4 Path to Exile
2 Celestial Flare
2 Kitchen Finks
1 Retreat to Coralhelm
2 Stony Silence
1 Crumble to Dust
1 Thalia, Guardian of Thraben
2 Spellskite
1 Deflecting Palm
1 Worship
1 Burrenton Forge-Tender
1 Gut Shot
1 Kataki, War's Wage
1 Destructive Revelry
So here's what I learned on the various card choices. I went with 23 lands+4 dorks because it lead to a higher number of fast starts, going purely by the numbers I posted above the fastest decks went all the way up to 28 lands with 0 dorks, but 4 dorks over 4 lands was only about 0.1 turns slower while having a healthier mix of combo and non combo wins, and I thought the versatility was worth the speed. Additionally, I couldn't think of another 4 utility lands I wanted. Hierarch makes colored mana which is important but the utility lands were colorless... I think the higher land count with fewer dorks is more defensible in 3 color land. With 4 colors you just can't afford for any of your first 3 land drops to be colorless. The low number of bears is another reason I used 4 dorks, I was already skipping 2 drops so I needed 3's to cast. Anyways onto specific cards:
1. As the simulator suggested, Blighted Gorge was fantastic. Tutoring removal is great and mana acceleration or Knight combo mana is just enough to get you to activations. Gorge won me 2 games (once tutored, once drawn) and could have won me another had I sequenced my turn better. I think this is a 4 color only option, Naya tries to close the game out too fast and Bant can't pay the red. Similar logic should hold true for a version using black with Blighted Fen.
2. Cathedral of War was horrible, I was experimenting with various land toolbox cards and unlike Gorge this one didn't work out at all. The manabase can handle the second colorless land but this isn't the one you want it to be. Kessig is just so much better, and any option needs to not come into play tapped.
3. Stirring Wildwood and Horizon Canopy rounded out the land toolbox and were fantastic.
4. The colored mana sources didn't work. I took way too much damage from my lands (especially mid combo), and the colors didn't come together. I needed more fetches because twice in 6 rounds my land drops demanded I get a Steam Vents which isn't in there, and not being a Forest/Plains is a bad include anyways. I also felt like I needed more basics, and more fetches. I'm not sure if this is an excuse to goto 3 colors, to add more lands up to the suggested 28ish, or both. More fetches or Ghost Quarter seem important though, Knight just wasn't getting fed well enough.
5. Meddling Mage was better than Voice in the field I faced, I wished the numbers were reversed. A split is definitely right though so that Collected Company has the potential to hit what you need.
6. I'm rethinking the whole Retreat plan, I only comboed once and it honestly wasn't even all that good when I did. Every other time my life total was too pressured to combo.
7. Celestial Flare is something of a pet card, when I played against Bogles round 2 and got a free game 1 win off of it I was happy it was there. My opponent actually got a judge to deck check me on it he was so surprised. That's why I like the card though, no one ever expects it and it is by far the best edict in the format. Plus, once you show it a single time it's a very easy card to bluff in future games which makes it the perfect game 1 card.
Now for the sideboard
8. Retreat to Coralhelm was a waste of a slot, I never once brought in the third copy. In hindsight I'm not sure where I ever would. Wanting the extra copy would imply my opponent isn't pressuring my life total and doesn't have some way to interact with a Knight. If that's what's happening, why not just have another threat instead?
9. Worship was... interesting. It has a great interaction with Geist of Saint Traft and while it prevented me from losing several games it didn't actually provide me with an opening to win the game. I drew it in every round I brought it in, and every time I played it (which was most games as I drew it a lot) the game went to a draw.
10. Destructive Revelry should have been a Reclamation Sage, I just didn't remember that card existed until the middle of the event.
Draw
https://www.dropbox.com/s/tflyunjsslh2zpf/KnightfallWide10000gamesDraw.txt?dl=0
Play
https://www.dropbox.com/s/02o0xyn0xz77n6h/KnightfallWide10000gamesPlay.txt?dl=0
I did something different with this data and included in each deck between 5 and 6 extra removal spells. The simulator never actually cast these since there wasn't a target to do so, but they do still count as cards in deck construction. They're labeled in my data as Path but it could just as easily be Condemn, Oust, Celestial Flare, or anything else, so between these and burn spells you're looking at 9-12 removal spells per deck, which should be somewhat accurate.
The first category I want to show is the fastest win on the play. For those who didn't read my previous posts on the subject, the reason I'm posing the top several decks rather than the top deck is because variance is a thing in Magic and it's important to look at the similarities/differences of several top lists.
http://imgur.com/DARTkYv
Next is the top deck on the draw
http://imgur.com/ZjSh862
Because of my sort parameters on these, the 5 path decks will naturally be faster than the 6 path decks which should only make sense. I'll save the spam and not post the top 25 with 6 paths but instead I'll just write down the results. On the play the average win turn goes up by 0.035 turns and the cut is a bear. On the draw the average win turn goes up by 0.04 turns and the cut is a Retreat to Coralhelm.
Next I want to talk about what deck combos early the most often. Nothing is able to turn 3 a consistent amount, with the highest on the play combo being 509 times out of 10,000 games but there are decks that win early and often. For this I'm going to use the sum of the decks turn 3 and turn 4 wins. These decks are slightly slower than the fastest decks but not by much. They have a high rate of comboing to win the game (around 29%) but the turn 3's don't show up too often, it's mainly a list of what does the best on turn 4.
Play
http://imgur.com/4qoIcjM
Draw
http://imgur.com/VDNBfCN
It's interesting to me that Geist doesn't show up in these decks.
Overall, I think it's most useful to look at the fastest decks and then to look at the decks most likely to combo if it's a bad matchup. A good game 1 configuration probably involves maximizing for the play and games 2/3 involve whatever you'll be on in that game.
Last, I want to talk about land counts. If you look at these results, alongside my previous results the first thing that will jump out at you is that the counts are very high. I've checked and double checked why they're so much higher than the traditional amounts and I'm pretty sure at this point that it's not a flaw in my logic, it's just something revealed through testing. I'm not really sure how I lean on that right now. In the play data, you can drop 4 lands for 4 mana dorks and see about a 0.1 turn slowdown in your games, but the starts are much faster and they appear to be less reliant on the combo.
Here for example are the top 25 fastest decks on the play with 4 mana dorks included.
http://imgur.com/L9Mey6K
And on the draw
http://imgur.com/92CuXAu
What this makes me think more than anything is that utility lands are better than we usually give them credit for being. Horizon Canopy seems way too good to pass up now, and if you have red even Blighted Gorge looks pretty solid given how much this data suggests burn spells are what you want.
My main thought is that a deck like this is poise
I had similar thoughts which is why I'm now playing the deck. Where you want to be in the format (and this was true pre ban as well) is aggressive with a combo win.
Anyone should be able to download that (unless Dropbox gets mad at me), open it up with their spreadsheet program of choice, delineate the columns with tabs, and have all of the results. Then you can sort by whatever metrics you wish. I will note that you'll have to make one extra column, the WinTurn column gives a total of the wins so in order to get the average you need to divide it by the total number of games, 10000 in this case. It's a small annoyance but one that had to be done since my program wasn't printing the results correctly otherwise (and it also saves a small amount of CPU power while running the simulation). There's also a handful of decks that are repeated twice in there (bonus points if you can find them... they're somewhere in the 26 land decks), this is because my computer froze halfway through the simulation and then again a few minutes later once it started running and I had to restart it at that point.
Anyways, this one took approximately 48 hours to run and is on the play. I'm doing another now as I type this which will be on the draw which I suspect will give very different results. A few days ago I did a smaller run to compare play/draw and the results were interesting (those were posted above). This current run is with Path to Exile included along with a couple other removal spells. My goal was to try and figure out what the correct cuts are for more removal without sacrificing speed. I went very wide in looking for results here which is why it's taking so long to generate answers.
The possible deck pool was:
Land - 20 to 30
Manadork - 0 to 4
Nacatl - 4 to 4
Bears - 0 to 6
Goyfs - 4 to 4
Geist - 0 to 4
Retreat - 2 to 4
CoCo - 2 to 4
Bolt - 4 to 4
Shock - 0 to 2
Path - 5 to 6
You'll notice a few cards which were set to 4 as a constant and some others that I made 2-4 of's rather than testing the full range. That was a time saver at the possible cost of some accuracy, if I hadn't done that the run time would have been 6750 days (18.5 years) compared to 2 days.
In the end 41,149 decks were tested which means 411.5 million games of Magic systematically analyzed. As expected the 5 path decks were faster than the 6 path decks (which is just logically correct so helps prove the simulator is running properly). I won't post screenshots this time, I think it's a bit redundant with me simply posting the full data (if you don't know how to work a spreadsheet but want to look at it send me a PM and I'll explain, it's pretty easy to do) but I will post the fastest deck of each for some comparison, plus with such a large dataset I don't think just the top 25 lists do it justice. These tests are on the play, previous results suggest that on the draw you want to combo more, cut lands, and bring in burn. Essentially the strategy on the play is to win through board presence while the strategy on the draw is to use the combat step less.
5 Path:
28 Land
4 Nacatl
3 Bears
4 Goyfs
4 Geist
4 Knights
2 Retreat
2 CoCo
4 Bolt
5 Path
6 Path:
28 Land
4 Nacatl
4 Goyf
4 Geist
4 Knight
3 Retreat
3 CoCo
4 Bolt
6 Path
I'm very surprised to see the land counts being what they are. I would have thought something along the lines of 26 lands with 2 manadorks would work better than 28 lands but those configurations are very far down the list. Here's the highest performing 26 land/2 dork list.
26 Land
2 Manadork
4 Nacatl
3 Bears
4 Goyf
3 Geist
4 Knight
3 Retreat
2 CoCo
4 Bolt
5 Path
This configuration did have a 1.5% turn 3 win rate, but the average game was 5.2002 which is a good deal slower than the other options at 5.1201 and 5.1564.
One more interesting list is the best performing turn 3 list. I think this is a metric worth looking into further because in bad matchups where the game plan may be to combo as quickly as possible as our method of racing, this is a configuration we should keep in mind. The fastest turn 3 configuration was able to turn 3 in 509/10000 games or 5%. It is
25 Land
4 Manadork
4 Nacatl
2 Bears
4 Goyfs
0 Geist
4 Knight
4 Retreat
4 CoCo
4 Bolt
5 Path
Obviously going beyond 4 mana dorks would help here, but my constraints set 4 as the maximum. This is a configuration I'll look into optimizing more later on.
I said that would be the last but I have one more I just thought of which is closely related to the turn 3 metric, and that's which deck put together the combo most often (remember that my simulator is capable of playing scry 1 so that aspect of Retreat to put things together is being utilized, and it can play multiple scry 1's in a row... though it has no idea how to play a scry 2+). So here's the deck that comboed the most (keep in mind that it may have comboed the most because it couldn't quickly win on the board).
30 Land
4 Nacatl
1 Bear
4 Goyf
4 Knight
4 Retreat
4 CoCo
4 Bolt
5 Path
The average win turn on this setup was 5.2291 with a combo rate of 3301/10000 games or just over 33%. I suspect this one is variance however because of the 0 manadorks. The next 3 highest combo decks all played a few.
I have wondered the same thing, I haven't posted most of my results for that very reason. Assuming my model is correct (and I don't see why it wouldn't be... for a goldfish, real games may require more interactivity though) I think that what is really being illustrated here is the power of utility lands, it's very possible that the community as a whole has underrated just how many lands we should run, and just how good utility lands are in order to get spell like effects while hitting land drops. Based on my results I actually just picked up some Horizon Canopys for myself, and I'm even considering grabbing a couple Blighted Gorge (I already have the other Blighted) on the theory that having more burn post board is actually a good plan (pure speculation on my part, Blighted Gorge is going deep but I do love a land toolbox). Also keep in mind that I haven't calculated the correct total with fetchlands which should in theory change it by 1 land or so. Calculating fetchlands is on my list of things to do but I need a narrow range of other cards to test first.
Also, keep in mind that what's optimal in a goldfish isn't necessarily optimal elsewhere. That's why I added removal even though it makes a pure goldfish worse. I want to see just how much things slow down by adding those cards. Anything that reduces speed by less than the value the extra card gives should be worth it, but that won't necessarily show up in a speed test.
Here's the top 25 configurations I came up with (keep in mind that variance is still a thing)
First is on the play
http://i.imgur.com/xTpcCqZ.png
Second is on the draw
http://i.imgur.com/S99hIPa.png
What I find interesting here is that on the draw there are considerably more turn 3 wins though the number of combo wins remains fairly steady. The general strategy for an optimal kill speed looks to be cutting a couple of lands and increasing threats. If nothing else, there's probably a valid strategy in cutting a land for an additional CoCo or good 2 drop in sideboard games when you're on the draw.
As a result I fixed the bug and ran a shorter 1500 game test on the play, and will have another one on the draw later tonight. And after those results I'm going to add in some extra removal that slows the deck but gives more interaction.
Screenshots of the best decks will have to wait until later tonight where I have access to a computer with a high enough resolution to take them but for now I can report. The results are starting to look rather in line with what has been seen in well performing lists.
4804 decklists were analyzed, the fastest deck has a win turn of 4.948 (that previous bug really sped things up) and the most common numbers of each card among the top lists are
Lands - All over the place, 25-29, with 27 being most common
SmallOne (Manadorks) - 2 in the top lists 2-3 in the top 25 lists
LargeOne (Wild Nacatls) - 4, this was set as a constant so 4 is the only possible value
SmallTwo (Bears) - 3-4, 4 is most common
LargeTwo (Goyf) - 4, this was set as a constant so 4 is the only possible value
Geist (Geist of Saint Traft) - 3-4, 4 is most common
Knight (Knight of the Reliquary) - 4, this was set as a constant so 4 is the only possible value
Retreat (Retreat to Coralhelm) - 2-3, 3 is most common
CoCo (Collected Company) - 3-4, 3 is more common in the top 10, 4 is more common overall in the top 25 or even top 100.
Bolt (Lightning Bolt) - 4, this was set as a constant so 4 is the only possible value
Shock (Shock) - 0-2, it's split pretty much 50/50 but decks are either all or none, very few 1 of's.
So, putting that all together that in my opinion locks the ideal skeleton to something like this (I should note that several of my top performing lists came to 61 rather than 60 cards, though not all of them)
By CMC
Land - 26
1 CMC - 12
2 CMC - 8
3 CMC - 11
4 CMC - 3
By Type
Land - 27
Creatures - 22
Spells - 11
By Role
Mana - 29
Beaters - 19
Spells - 11
Flex - 1
Looking at it this way, the most interesting thing I found is that Collected Company seems to be working optimally with 22 good creatures, which makes me think that shooting for 27 creatures at the cost of removal would be a disadvantage because the fastest lists don't need that many hits, though something like Fiend Hunter/Banisher Priest to be both a creature and removal is probably pretty excellent.
This is the thread that I was referring to, there's not much discussion on it though sadly. Here is the script as it currently sits in my Dropbox, though the exact script that I have running on my spare computer right now is slightly different because the parameters were changed. It's easy enough, just have python 3.5 installed and run it via command line or use an IDE like PyCharm or PyScripter and run it there. For reference my runtime for the above with a 2.9 ghz quad core cpu and 32 gigs of ram (not that it was ever using more than 4... it was probably running on just one cpu core too) was 15 hours. My goal is to keep the runtime around 10 hours per iteration on my computer. With the above settings 5220 decks were tested at 2700 games per deck.
Geist could be dropped and replaced with Smiter easily enough, it's just a matter of changing it's power from 6 to 4 and removing the legendary tag. Bolts can become Paths but the optimal deck will run 0. This is because of the current approach which is just a goldfish. Bolt can be used as an interactive spell because in this case it's always being aimed at a persons face, but you could change it to hit a creature at the cost of your clock easily enough.
The SmallTwo is technically a 2/x for 2 mana. It could equally be a Meddling Mage, Grizzly Bear, Pridemage (without Exalted), or Voice of Resurgence, it could even be an Aegis of the Gods, as a goldfish the only thing that really matters is the power and mana cost. Making something more interactive than that is extremely difficult because it gets computationally complex though it is on my to do list. Fortunately, I have a pretty light semester right now so I have considerable time to work on projects like this.
The more complex something is the longer the runtime gets. Adding the logic from my previous version of the simulator to the Knightfall specific version for example dropped my runtime from ~0.2 seconds for 10,000 games to ~2.0 seconds for 1000 games (a 100 fold slowdown). This means that the more detail you add to your card types the higher your runtime goes, which means the fewer results you can obtain in any given time frame so you gain uncertainty due to the very high variance in Magic. On the other hand if you use very generic cards with a large number of games things become overly simplified and while you eliminate variance you add uncertainty in the cards acting in the ways you imagine they will. Ultimately you have to take one form of uncertainty or the other, I'm aiming for something of a middle ground leaning more towards simplicity as that's far easier to implement.
Adding a new card increases runtime between 100% and 500% depending on how many you're running. Which is just simple math to prove/explain: The number of possibilities (not all will run because they don't meet the total card criteria) would be for example 14-30 lands (17 possible land counts), 5 possible geist counts (0, 1, 2, 3, 4, 5), 5 possible bolt counts, and 3 possible retreat counts (0, 1, 2) so that's 17*5*5*3 or 1275 possibilities. Adding a 1 of path is 0-1 possibilities, which makes that 17*5*5*3*2 or 2550. As you can see, it skyrockets very quickly, the possibility of a second 1 of would double it again, or the possibility of a 2 of Path would be a 50% increase. This is why it's stepping in values of 2 in order to find ranges. That constrains a 0-4 to 0, 2, 4 or 3 possibilities. If you do that to the above that gets you 9*3*3*2*2 or 324 possibilities, and then running it a second time for the inbetween values is another 324 for 648 possibilities total rather than 2550. This reduces the base runtime but the exponential increases are still there so it's a very real constraint that always has to be kept in mind, otherwise it spirals out of control from the range of what I can do on a spare desktop PC to something that would need a supercomputer.
Currently I have 11 card types which is very high, reducing that by 2 cards from 11 to 9 would cut the run speed to nearly 1/10 what it is right now (1/9 actually). This is just an unavoidable fact of the computational complexity of Magic which I explained above, it is an extremely complex game. Because of this, making an exact optimal deck isn't actually possible (and the optimal deck would vary based on what you're playing against anyways), instead the goal is to break cards down by role with groups of similar cards and to make a skeleton such as wanting a mana curve of x 1/2/3/4 drops, y ramp, and further subdivided into mana dorks, combo pieces, CoCo, hits for CoCo, and so on. Once that is done, you have to go back to more traditional deck building methods.
My simulator knows the following 11 cards, and was given the following constraints (you'll note there are some auto 4 of's, this is due to previous results of more simplistic decks indicating this is right, and I could save A LOT of processing time by doing so). Cards are only available in groups of 2, so 0-4 cards means only 0, 2, or 4 cards can be tried. The followup which is running now will then use for example 2, 3, and 4.
Land - 5 color land, 16-30
SmallOne - Birds of Paradise, 0-8
LargeOne - Wild Nacatl, 2-4
SmallTwo - Grizzly Bear, 0-8
LargeTwo - Tarmogoyf, 4
Geist - Geist of Saint Traft, 0-4
Knight - Knight of the Reliquary, 0-4
Retreat - Retreat to Coralhelm, 0-4
CoCo - Collected Company, 0-4
Bolt - Lightning Bolt, 4
Shock - Shock, 0-2
The simulator is programmed to recognize the combo and to prioritize the combo over other lines if it comes up. It also recognizes the legendary rule for Geist, and CoCo has a priority on cards (Knight if you can combo, Geist if you lack one, Knight, Goyf, Nacatl, Bear, Birds). For complexity reasons there are only 2700 games being played with each deck though the next iteration should be able to manage 10,000 of each (more possible decks means fewer games of each deck can be run) since these results narrow things down a bit.
Linked is an image of the top 25 deck lists given these parameters (I would have attached but the forum is claiming my 260k image is over 1 GB and thus is too large to attach). Of particular note is the #1 build which comes in at an average win turn of 4.0148. I am optimistic that with some fine tuning of the card pool and a more tuned mulligan strategy it will be possible to break the turn 4 barrier on average. All games are on the play, once I've discovered an optimal play deck I'm going to attempt to build an optimal draw deck.
I'm going to try to get some more indepth results as well, but my hope is that this gives people a better starting point to building the deck as it's currently all over the place.
http://imgur.com/72kVURY
In the videos I've watched I like the way the Zoo variant plays, it's aggressive and gets a lot of extra percentage points from unknown information. For example, if your opponent kills the Goyf they have to risk your KotR killing them, but if they don't kill the Goyf it's going to kill them eventually. I like having diverse threats like that.
When I've tried some sample lists online though before making my own I keep running into a problem that I didn't see in the decks video primers. The manabase is simply costing me games. It feels way too fragile as a 4c manabase but I don't see an alternative. If you want to pressure the opponent Nacatl puts you in red which also then lets you pick up Bolt for free, while Geist and the combo put you in blue. I haven't read this entire thread but I've read about the first 10 pages along with the last 30 and I haven't seen any manabase discussion at all other than on Sejiri Steppe and Kessig Wolf Run. It seems like people are just throwing some lands together and calling it done.
So, on that subject what have people thought about the various land counts? Collected Company really wants around 27 creatures and 3 CoCo (more on that below) brings you to 30 cards. 22 lands makes it 52 which leaves 8 other spells. That setup seems to interfere with the combo though. If there's only 8 other spell slots you have to sacrifice either removal or combo potential. Has anyone been able to work in Ghost Quarters or Horizon Canopy? What fetchland balance is optimal? Obviously some green fetches are good because all of the real T1 plays involve green mana but beyond that it becomes more of a question. Is Wooded Foothills really offering anything over Flooded Strand for example? Arid Mesa and Windswept Heath are clearly the best ones because they gets all 4 colors without having to run a Steam Vents, but is it right to be running say 4 Wooded Foothills over Misty Rainforest? Or even over Scalding Tarn?
Outside of the manabase when I build or tune decks I like figuring out a deck skeleton based on what breakdown for the various slots is optimal. That will take me awhile to decide on for this deck, but in the meantime I have some thoughts on the three drops that might be worth discussing.
With Collected Company, I've seen a lot of lists running 4 but I'm not so sure that's correct. I was playing with a bunch of percentages today in a spreadsheet and basically what I found was that the sweet spot between deck slots invested and consistency was at 3 copies of CoCo attempting to find 2 of's. This invested the minimum number of resources to have a reasonable change of finding the card. It seems to me that a lot of these lists are running more 3 drops than the format really allows you to run, so I think making a few trims here to get a better curve could improve lists as the ~13 3 drops definitely feels like too many.
For example the three/four drops could look like
2 Geist of Saint Traft
3 Loxodon Smiter
3 Knight of the Reliquary
3 Retreat to Coralhelm
3 Collected Company
That cuts 2 slots without appreciably hurting the chance at a combo, which then frees up two slots further down the curve for either more aggression or protection.
To put this in perspective a 2 of with 3 CoCo's is as consistent as a 3 of without any CoCo and a 3 of with 3 CoCo's is as consistent as a 4 of without any. The pattern holds for 1 of's as well. Basically you can think of 3 CoCo as being worth +1 copy of every other creature in the deck. The 4th CoCo can push that further but because of how math works, the second CoCo has less impact than the first, and the third has less than the second, so really that final copy isn't very good. If people wish, I can post the spreadsheet I built on this stuff later.