[Primer/Deck] Burn

Update on that simulation I've written about a few times. All my code for it is written, all that's left really is to let it run with numerous deck permutations and generate data, then extract that data from the database, which is relatively simple though a bit tedious. Looking over the logs not all of the lines appear to be optimal, but they look to be good enough.

They're pretty long so I don't want to post any sample logs for people to scrutinize the lines, but if someones interested I can PM a couple games worth of output.

Quote:

---

Originally Posted by Dio

What do you guys think about replacing Barbarian Ring with Keldon Megaliths? I don't have much experience with burn and I know Barbarian Ring can deal the final damage, but I was wondering if it would be better to deal 1 damage each turn after emptying our hand.

---

Last year I think it was, or maybe the year prior someone managed to top 8 an SCG with Keldon Megaliths, if I remember correctly he made it all the way to the finals. In general though it's just not very good. We actually have a lot of uncounterable damage right now with Exquisite Firecraft and Volcanic Fallout, plus we can usually sneak some in as well. That reduces the value of just how many cards like Barbarian Ring you need. There's definitely not a high enough demand on the effect to use it.

Quote:

---

Originally Posted by Brael

Update on that simulation I've written about a few times. All my code for it is written, all that's left really is to let it run with numerous deck permutations and generate data, then extract that data from the database, which is relatively simple though a bit tedious. Looking over the logs not all of the lines appear to be optimal, but they look to be good enough.

They're pretty long so I don't want to post any sample logs for people to scrutinize the lines, but if someones interested I can PM a couple games worth of output.



Last year I think it was, or maybe the year prior someone managed to top 8 an SCG with Keldon Megaliths, if I remember correctly he made it all the way to the finals. In general though it's just not very good. We actually have a lot of uncounterable damage right now with Exquisite Firecraft and Volcanic Fallout, plus we can usually sneak some in as well. That reduces the value of just how many cards like Barbarian Ring you need. There's definitely not a high enough demand on the effect to use it.

---

Still looking forward to seeing what it generates on the random draw vs Magma Jet vs SDT (vs Brainstorm) question. I know simulations are hard work, thanks for working on it and volunteering to share results.

Quote:

---

Originally Posted by Brael

I would drop Lava Spike for creature space long before Eidolon. Lava Spike is the worst card in the deck that frequently sees play.

---

I'd never get rid of lava spike

Quote:

---

Originally Posted by Lord Darkview

Still looking forward to seeing what it generates on the random draw vs Magma Jet vs SDT (vs Brainstorm) question. I know simulations are hard work, thanks for working on it and volunteering to share results.

---

I enjoy working on it, plus it's good practice for me. It's just taking longer than expected because getting the motivation to work on it for long stretches of time during a summer where I'm trying to enjoy my time off is difficult since I have another project I'm working on too which competes for attention.

Quote:

---

Originally Posted by Sidneyious

I'd never get rid of lava spike

---

I fully cut Lava Spike and I'm at less than 4 Rift Bolt. I find that neither card is flexible enough these days. I'm typically trying to deal 18-24 damage, not simply play as a combo deck that resolves 7 spells to win and something that only sends 3 at their face isn't strong enough then.

http://uploads.tapatalk-cdn.com/2016...77b964c8c5.jpg
Thoughts? I'm rather intrigued.

@Sidneyious
I think you missed that it says target creature not target creature or player...

Edit:Even if it said player it would still be horrible because youd need 2 of it in the grave for it to be better than lightning bolt.

Quote:

---

Originally Posted by Sidneyious

http://uploads.tapatalk-cdn.com/2016...77b964c8c5.jpg
Thoughts? I'm rather intrigued.

---

No.

Quote:

---

Originally Posted by Sidneyious

Thoughts? I'm rather intrigued.

---

Absolutely terrible card. To begin with it's creatures only. Next, it's only 2 damage on the first one and then 4 damage on the second one. You need to draw two of them for them to be as good against creatures as Lightning Bolt is individually. Assuming it could hit players it wouldn't beat out Lava Spike until you cast the third copy.

Yay, at long last this simulator is finished. I'm taking a couple different approaches with it, one is running a large number of deck permutations at fewer decks each. This will take quite a while to finish (days, maybe even weeks). But I'm also running one using preset decks at a larger number of runs. I can get in about 1000 games in 20 minutes or so, but I can't really automate the process of setting up the decks so it's one of those situations where I can set it, forget it, and then come back to it later on when I'm near the computer. I have 1000 games complete with a deck list using Tops right now and I'll post those results here later, and then post a few other deck configurations with Magma Jet, Vexing Devil, and neither later, maybe some combinations of those cards too. I'm going to wait on any data though until there's multiple decks worth so that they can be looked at in relation to each other. The data with top for example is suggesting slow games, but the lines my simulator takes could be a bit slower in general.

Here is an example of the type of output I can generate
http://imgur.com/a/tKA8D

In this example you can see in the first image a deck with an ID of 2 and a deck list of 6 fetchlands, 6 lands, 4 eidolon, 4 goblin guide, 4 vexing devil, 4 flame rift, 4 fireblast, 4 atarka's command, 4 incinerate, 12 lava spikes, 4 magma jets, and 4 tops.

In the second image, we're looking at games instead of decks. It's filtered to only games played with deckID 2. Here is where all the mulligan information is stored. You can see in gameID 3 it mulliganed to one card, 3 of which were caused by no land hands (as you would expect from a 12 land deck). We're going to focus on gameID 8 though because it looks interesting.

In the third image we're looking at turns, and it's filtered to turns taken in game8, which itself is a subsection of deck 2. We can see that game 8 consisted of 6 turns and by looking at it's max mana the mana development went 1-1-2-2-3-3, so it was a 1 land keep off of a mulligan to 5 (the mulligan information having come from the games table). What we can also see, is that turn 5 was the big turn having dealt 11 damage. The prowess count tells us that two spells were cast and that there was 3 mana available.

Going a level deeper we can see the hand and board states in the next 2 images. On the board is 3 lands and an Eidolon and in the hand is 2 Bolts, Incinerate, and Flame Rift so we can tell that the 11 damage came from Flame Rift+Lightning Bolt (7), a swing from Eidolon (2), and a trigger from Eidolon (2)

If anyone knows their way around SQL and is interested I'll be happy to share the data and the program.

The dreaded triple post, but as promised, some data... scroll to the dotted line if you just want the results.
It turns out I have a bug in my SDT code (SDT is pretty complicated, so no surprises there... though oddly my deck ordering logic is sound, it's keeping it on the board that's messed up) so I'm leaving that out for now. Instead I'm giving data on three decks trying to answer the Magma Jet question, there's 4 cards that change between them
Deck 1: 4 Magma Jets
Deck 2: 4 Vexing Devils
Deck 3: 4 extra Bolts (16 total vs 12 in the others)

The Magma Jet deck took 5.600 turns on average to win the game
The Vexing Devil deck took 5.434 turns to win
The Bolt deck took 5.375 turns to win

Additionally, because it's somewhat relevant (higher curve on the Jet deck) here's some mulligan data.
The Jet deck mulliganed in 30.2% of games taking 850 total mulligans. This means that 70% of the time the deck was playing on 7 cards, but in the other 30% of games it was down an average of 2.81 cards meaning if you go to 6, you're probably also going to 5. This suggests that a mulligan for a proper hand is worth around 2 cards or 6 damage so you don't want to take one unless 2 cards are bad and the rest are below average. If you have one bad card, you're better off keeping it.

The Vexing Devil deck took mulligans in 29.2% of games with 854 total mulligans. While the chance of a mulligan was reduced here, it suggests that the effect of the mulligan is greater. This probably isn't a surprise because Magma Jet provides some manipulation to scry out of a bad hand.

Additionally, the Bolt deck also took a mulligan in exactly 29.2% mulligans (it's very odd that both were equal) but took the fewest total mulligans at 839.

-----------------------------------------
This data is very interesting to me because it suggests something much different from my spreadsheet and how I've built the deck in the past which claims creatures beat burn spells. Here a Lava Spike proved to be superior to Vexing Devil, and the library manipulation proved to be weaker than both.

The next question I want to answer is lifegain. A game of Legacy that we haven't lost can range anywhere from 17 life (3 fetches) to 26 life (Batterskull hit+DRS activation+no fetches) and I'm curious how the creature vs burn relationship looks at those extremes.

Quote:

---

Originally Posted by Brael

Absolutely terrible card. To begin with it's creatures only. Next, it's only 2 damage on the first one and then 4 damage on the second one. You need to draw two of them for them to be as good against creatures as Lightning Bolt is individually. Assuming it could hit players it wouldn't beat out Lava Spike until you cast the third copy.

---

The cumulative damage from 3 of these would be 2+3+4 = 9 (3bolts). Only casting the 4th would have any cumulative advantage of damage output by a whopping 2 points (5 damage from the last copy). So if you play 4 each game then yeah it's better.

Quote:

---

Originally Posted by snorlaxcom

The cumulative damage from 3 of these would be 2+3+4 = 9 (3bolts). Only casting the 4th would have any cumulative advantage of damage output by a whopping 2 points (5 damage from the last copy). So if you play 4 each game then yeah it's better.

---

It's still not better because it only hits creatures, not players.

Also, the damage from this burn spell does 2/4/6/8. Cumulative that's 2/6/12/20. Bolts go 3/6/9/12. It's on the third cast of this where you've done more total damage, assuming of course there's no overkill.

Quote:

---

Originally Posted by Brael

It's still not better because it only hits creatures, not players.

Also, the damage from this burn spell does 2/4/6/8. Cumulative that's 2/6/12/20. Bolts go 3/6/9/12. It's on the third cast of this where you've done more total damage, assuming of course there's no overkill.

---

Not quite. It's X+2 damage, not 2X damage.

It's still a jank card though, and should never have been brought up for discussion anyway.

Regarding your analysis on Magma Jet vs Vexing Devil vs 4th Bolt, it's quite interesting. What sort of conditions do you have for the deck, both in terms of land count and mulligan conditions? (I'm assuming a base of 20 lands (8-12 fetches, rest Mountains), and mulligan 0/5/6/7 land hands) It would make some sense that the deck with less 2-drops would in theory have to mulligan less, just based of the average CMC of the deck being reduced. SDT will be quite hard to program into damage calculations, especially as it's not something that is good for goldfishing with, so I would expect both SDT and Magma Jet to slow the clock down by about half a turn.

Lifegain will probably add an extra turn or two to the calculations, although if a Batterskull hits the field without an answer, we're probably dead anyway.

My LGS is finally doing Legacy once a month and I'm going to take Burn. We're playing for a FOW. Anyway, I have no idea what the meta will be since this is the first event they're having. I have a feeling I might face a few Burn mirrors. Any lists you would recommend? Thanks!

Quote:

---

Originally Posted by Speedbump

Not quite. It's X+2 damage, not 2X damage.

---

X+2 is 2/4/6/8 which if you add up total damage done between all copies of the card is 2/6/12/20. Bolt is a simple 3/3/3/3 or cumulative 3/6/9/12. So it takes 3 copies to overtake Bolt.

Quote:

---

Regarding your analysis on Magma Jet vs Vexing Devil vs 4th Bolt, it's quite interesting. What sort of conditions do you have for the deck, both in terms of land count and mulligan conditions? (I'm assuming a base of 20 lands (8-12 fetches, rest Mountains), and mulligan 0/5/6/7 land hands) It would make some sense that the deck with less 2-drops would in theory have to mulligan less, just based of the average CMC of the deck being reduced.

---

After posting, I started looking at things more indepth and saw some stuff that didn't make sense in the data with mulligans that I didn't catch in reading log output before. As a result I redid the tests tightening up the mulligan logic. After doing so, the results changed slightly, but because they were only over 1000 games variance can easily be at play. I'm not going to repost all of the mulligan stuff but more or less mulligans happened in 1/3 of hands. So 1/3 of the time you go from 7 to 6, and 1/3 of that you'll go from 6 to 5, and so on.

Wins this time were 5.423 turns for Bolts, 5.57 turns for Jets, and 5.258 turns for Vexing Devils.

Quote:

---

Lifegain will probably add an extra turn or two to the calculations, although if a Batterskull hits the field without an answer, we're probably dead anyway.

---

A couple more tests I'm going to run are an all out burn deck vs a very heavy creature list. Because of how the tests are set up though, it's easiest to add these to the list, and then test all the life changes at once after that since I added a way to automate a few specific deck lists at once, since that seemed more useful than broad categories of decks.

My land counts so far have been using 19 lands, 11 fetches 8 fetchables. After I test these creatures I can try some other land counts too. To give some idea on mulligan rates (not every mulligan reason is being recorded, but the land based ones are). Across the 4 decks chosen (so Top data is included here) there were 783 mulligans to 6. In that, 67 were for too much land (defined as 5 or more) and the other 715 were to too few lands (defined as 1 or fewer lands). 1 was for an undefined reason which could have been too many tops (2+) or too high curve of a hand (total hand costs above 10).

I suspect that creatures will heavily out perform burn spells though due to the fact that there's no opponent to block/kill them.

Eventually, in order to eliminate variance I'm going to have to do a few more rewrites and try to get some optimizations in, but for now it's good enough.

-------------------------------------------------
Edit:
It comes to mind that I have enough data now to compare the effectiveness of Goblin Guide vs Monastery Swiftspear and quantify how much damage one does over the other.

In this dataset Guide swung 8562 times dealing 17,124 damage or 2/turn on average. Swiftspear after Prowess did 18,316 damage over 9981 attacks for an average of just 1.835 damage/swing. This is very notable, because I was convinced enough in Swiftspears power that in the event a choice had to be made on casting Guide vs Swiftspear, Swiftspear came first, which is why it has 15% more attacks. Swiftspear looks to start a game strong, but peters out once your hand is empty and with these games going an average of 5.5 turns the hand empties fast. That's the best explanation I have. My turn tracking sadly didn't work (if you look at the screenshot I posted yesterday you can see in the turns table where currentTurn always reads 0) so I can't tell exactly what turn Swiftspear gets bad on but I can say that out of the 9981 attacks, 4707 of them involved prowess so the damage split involved 8335 prowess triggers or 1.77 triggers/turn. This information would suggest that Swiftspears power drops off radically after turn 4.

In fact, this is interesting enough that I'm adding another test case to the creature deck, one with Swiftspear and one without (going back to Hellspark Elemental instead).

All in all, these results are way off from what my spreadsheet has predicted.

-----------------------------------------------------------
Edit 2:
Figured I would just edit the data into here rather than triple post again. This post is new enough that no one has read it yet.

Finished running the dataset on three more decks, a creature deck without swiftspear, a creature deck, and an all burn deck. No Jet or Top in either. All with the same 11/8 manabase.

Creatures with Swiftspear
5.226 average turns

Creatures without Swiftspear
5.165 average turns

All burn
5.490 average turns

I'm going to run one more test before playing with different amounts of land and life, and that's using the closest analogue for the deck I usually play (and I'll just edit it into this post when it's done).
------------------------------------
Well, that was disappointing, my usual deck performed so bad it's not even worth listing, over 6.0 for the average turn.

Quote:

---

Originally Posted by RPS

My LGS is finally doing Legacy once a month and I'm going to take Burn. We're playing for a FOW. Anyway, I have no idea what the meta will be since this is the first event they're having. I have a feeling I might face a few Burn mirrors. Any lists you would recommend? Thanks!

---

Proxies allowed? Not allowed? Local ringers who already own good Legacy decks?

In my experience, new Legacy metas often revolve around cheap combo decks. That means you need good no mana interaction like Faerie Macabre. It also means that your opponents are probably going to play more fetches/duals than they should (especially if it's proxy) and not enough basics. Price of Progress preys on these scenarios.

If I knew for a fact that I would be facing multiple burn mirrors I would splash white off a Plateau for Lightning Helix. Also Searing effects and Lavamancer are very strong in these situations. Leaning towards more instants so you can try to get some free value in bouncing Chain Lightnings is also a strong play.

Quote:

---

Originally Posted by Brael

X+2 is 2/4/6/8 which if you add up total damage done between all copies of the card is 2/6/12/20. Bolt is a simple 3/3/3/3 or cumulative 3/6/9/12. So it takes 3 copies to overtake Bolt.

---

X+2 is 2/3/4/5, 2X is 2/4/6/8. If it were actually a Bolt, it only overtakes cumulative damage on the 4th card.

Quote:

---

Originally Posted by Speedbump

X+2 is 2/3/4/5, 2X is 2/4/6/8. If it were actually a Bolt, it only overtakes cumulative damage on the 4th card.

---

Oh, you're right. I was completely misreading the card. That makes it even worse.

Sorry, I know this is a bit spammy, but I'm a red mage so I like finding inventive ways to count to 20 and talking about them.

More results, I'll only post details on the top 3 decks. I ran another 1000 games with each of the 8 decks tested so far adding some lands going up to 22 from 19. This second batch used 13 fetches/9 lands while the original was 11 fetches/8 lands. The average kill turn went down pretty dramatically, almost a full half turn. The lists were pretty stock minus the specific cards changed.

The best performing deck was a creature based list without Taylor Swiftspear. Here's the list it used:
fetch = 13
land = 9
eidolon = 4
hellspark = 4
guide = 4
marauder = 2
vexing = 4
flamerift = 4 (the first 4 in this category are usually Price of Progress for 2 lands in a real list)
fireblast = 4
atarka = 4
bolt = 8

It had an average win turn of 5.044

One other list was close to this (within the margin of error given the sample size) which was creatures with swiftspear. it was the same thing just with Swiftspear over Hellspark at 5.131 turns. The other 6 configurations were all atleast .2 turns slower ranging from 5.28 turns to 5.96 turns, in order they were, based on my comments for their names (the general configuration should be apparent from the name): Devil/Top, Stock/Bolts, Burn, Stock/Jet, MyList, Stock/Top

When it comes to mulligans the 19 land set had the following rates:
19 lands
7 - 5593
6 - 1640
5 - 551
4 - 158
3 - 42
2 - 10
1 - 4

22 lands
7 - 6090
6 - 1446
5 - 361
4 - 77
3 - 25
2 - 1
1 - 0

I've known for awhile that more lands were better, but I think info might convince me to run a couple more Barbarian Rings, depending on how I feel about Wasteland. Sadly, I can't (yet) include Barbarian Ring into this simulator because I'm not tracking graveyard information and that's pretty far down my list of priorities so we'll have to make do with fetches and fetchables.

Next up for tomorrow is these same 8 lists with total damage needed of 17 and 24 rather than 20.

More data. Over the next day or two (depending on how long they take to run) I'm going to run a few more tests to determine optimal land counts, and then I'll take a break from it for a bit. There's a few bug fixes and features I would like to add for additional data mining (still think SDT is bugged rather than bad data), and a few optimizations I would like to try to speed things up.

If all goes well it will be several times faster with some more complete features like turn tracking which initially slipped through the cracks so that I/we can look at things like which cards in an opening hand most strongly correlate to the fastest wins, and which correlate to the slowest wins. And with increased speed comes more accurate reporting and reduced variance. So far I've been limited to 1000 game sets, which is rather small. I would like to increase them to 100,000 game sets at a minimum which I think is doable with some tinkering.

Anyways, for the life information that I ran last night and today.

Taking the decks as a whole (8000 games per set), the 24 life set used 48,559 turns for an average of 6.07 turns per game. The 17 life set was 39,531 turns for an average of 4.94 turns per game.

Against the 24 life decks the best performer was again a creature list without Swiftspear at 5.543 turns. Creatures with Swiftspear were right behind it at 5.724 turns, showing that across tests Swiftspear is consistently slowing the deck down by .1-.2 turns, and the creatures with Vexing Devil were 6.010 turns. The three burn focused lists were all around 6.2 turns with Magma Jet performing the worst, and the SDT lists came in behind that.

Against the 17 life decks everything killed faster, but the order for the fastest and slowest decks was the same but the gap closed slightly.

The take away from this is that the last 7 damage is worth about 1 turn, so in a life gain heavy meta a Skullcrack or Atarka's that stops some life is worth a lot for your clock.

At this point I'm pretty confident in the deck hierarchy so my next few tests which will involve land counts are going to use a smaller sample of decks. I'm going to use a Magma Jet deck on the theory that scry gets better with more lands, a creature based deck without Swiftspear, and an all Burn deck without Jet. I'll also go back to testing with 20 life. The reason for this change is that I can test 3000 games in an hour right now so a 12 hour test will let me test everything between 16 and 28 lands. Also, this upcoming land test will be without fetches. The optimal fetch count is the second step after narrowing things down to what land count appears to work best.