Measuring with paperstrips

Since the goal is to beat modern, I think they will become solid enough to just invade pauper. With 30 different halves
With their own specialties they should be pretty versatile, and there's more than 300 ways of putting them together, so pauper could see a sudden overnight expansion of 300+ decks if my project ever reaches out to paupers.

It will take a number of years before I reach the 50/50 win% against my own testdecks, but when I'm at that stage it might be fun for me to participate in a couple of pauper tournaments, and from there I can hand out links to the full scale project.

Imagine the shock effect of such a happening :)

My best guess is that the format would be killed, and everyone would be stuck with "wicked decks" which would take on a fast pace life of their own, which I might then use to upscale my project against modern.

Ofcourse paupers could end up taking a horrible revenge without even knowing it, as my own project only uses modern legal cards.

Let me counter that question with another.

If you have room in a deck for seven cards, and you want to know if you should play 4 raging goblins and 3 lightning bolts, or if 3 raging goblins and 4 lightning bolts is statistically better, how would YOU
Find the answer to that ?

If I really wanted to find the answer experimentally, I guess one way would be to playtest the deck 3x bolt and 3x goblin and one "wildcard" (for example a basic land outside the deck's color) that I can decide to be either a goblin or a bolt as I draw it. Then I could just track and compare the amount times I needed it to be a goblin and the times I needed it to be a bolt. No paperstrip in sleeve needed.
Of course, in this particular example, Bolt would be the better option since it is a far superior card and the scenarios where I would want to replace a Bolt in my hand with a Raging Goblin are very rare outside of some very specific deck concepts, so the tedious experimenting wouldn't be worth it.

How about my question, why track individual cards with paperstrips?

Multiple data points :)
You may be able to track that one card on paper, but on a large scale it's easier to track directly in a sleeve than on a paper.

But what is the point of knowing that Healer's Hawk #1 was cast 23 times, Healer's Hawk #2 was cast 21 times and #3 was cast 20 times? It's just going to average towards the same number as the sample size increases (though the lower numbered cards will arbitrarily get cast a few more times due to your rule)

The point is that I will know more than the efficiency of hawks related to each other, I will know which cards got cast the most during all of the games, which in the end shows me which cards to cut :)

More important, the points reflect my playstyle, though it may be hard to read it that way.
There's a reason why porcelain Legionnaire #1 is at the top, while vault skirge #1 is placed a lot further down the list.

Theres a reason why healers hawk all seem to be better than the vault skirges, and I know why in that case, I prefer to cast the colored spells first in case that the mana is somehow taken out early.

All the small tricks I perform during games impact how often things get cast, and the system clearly registers this.
The points are an image of my entire series of battles. Interpreting that image can be hard or easy. The easy read is to just say higher numbers are better and lower numbers are bad. The hard way to read it is to ask, why a card got played less, because after all there is an equal chance of drawing them (not counting fetches, because that does affect the rate)

Also I wrote that the cards get points when cast, AND gets point detracted from them when they don't get cast, the number of times they weren't cast is subtracted from the number of times that they were cast.

So, if during the matches I face a chalice of the void set to one, at the early game, a lot of 1cc cards lost a point each.

But the scores highlight the best cards.

It also highlights the fate of the second, third and fourth copy in relation to the first. If they cluster together in the top, they all have an equal effect on the game, but if there's a lot of space between them they were becoming less important during gameplay.

You might also want to take a closer look at the singletons, as there are only one of each of those, they naturally see play less.

Youthful knight scored 9 less points than porcelain Legionnaire, so it's performance has been good.
Since it scored 5 more points than the second porcelain Legionnaire it might indicate that I was already low enough on life to prefer casting it more than a second porcelain.

Theres plenty of data to extract from these lists for me, and sometimes people spot something interesting in the points that allows me to dig deeper.

The way that the first two chromatic star are high scoring and close together indicates the use of them during early games with few mana where I desperatedly needed to color the mana of a land, and then later, where both of the last chromatic star are less important, but I found time to play them.

I'm sorry, but this still doesn't make any sense to me, and to me it seems you are misunderstanding statistics.

"It also highlights the fate of the second, third and fourth copy in relation to the first. If they cluster together in the top, they all have an equal effect on the game, but if there's a lot of space between them they were becoming less important during gameplay."
As far as you've described this process, no it really doesn't. All the battlefield raptors got clustered together because it's statistically the most likely outcome given a large enough sample size. In any situation in which you had Battlefield Raptor #1 in your hand without any other raptors and played it, you also would have played Battlefield Raptor #2 if that one was the one that ended up in your hand instead. The only difference is which paperstrip gets a point. The fact that you drew Battlefield Raptor #1 instead of Battlefield Raptor #2 tells you absolutely nothing, the only meaningful data is that you drew one raptor and played it. They will most likely get ordered from #1 to #3 though because of your rule of playing #1 first (there will be some situations in which you draw both #1 and #2).

Similarly, a large disparity between the most played copy and the second most played copy doesn't tell you much at all. The reason for the disparity between your copies of Chromatic Orb is mostly due to your system of playing #1 before higher numbers, and the more copies you have, the more likely it is that there are multiple copies in your hand so this rule takes place.

Now, to be fair, this could SOMETIMES give you a bit of information: if there is a card that you usually only need to cast only one copy of, there would be more situations where #1 overpriorizes the others and the higher numbers don't get played, thus widening the gap between the copies a little.

However, this information could be gotten much more easily by just noting ("2 copies drawn, 1 played") on a separate sheet of paper or Excel. The individual cards are a pointless confound, the only relevant data is how many copies were drawn and how many were cast. Your system is flawed, but in a way, your number rule happens to slightly move the results towards the actual probabilities.

Let's imagine a silly scenario. You've got four Delver of Secrets and four Fugitive Wizards in your deck. You need to cut one card from the deck. In the games, you cast every single Delver you draw. However, you happen to get a bit unlucky in your sample games and very rarely draw the Delver #4.

You might the following data:
7 Delver #1
6 Delver #2
4 Delver #3
2 Fugitive #1
2 Fugitive #2
2 Fugitive #3
2 Fugitive #4
1 Delver #4

Your system would cut the fourth delver because you got unlucky with that copy, which makes no sense.

Without tracking specific copies it would be:

Delver: 18 drawn, 18 cast
Fugitive: 18 drawn, 13 cast (which would mean a total of +13-5=8 points in your system as shown above).

My system ignores the unlucky Delver and correctly removes the clearly weaker Fugitive Wizard. In my data, I didn't specify what happened in which game but that could be done easily as well.

You could argue that this scenario was just unlucky. Yes, it was unlucky, but why use a system that takes more effor but provides less reliable results?

EDIT: oops, accidentally posted the comment multiple times due to lag :D

Agnomakhos has deleted this comment.

Agnomakhos has deleted this comment.

Agnomakhos has deleted this comment.

There's a lot of details involved in the whole process.
One failsafe against being unlucky is the reintroduction of lost cards to the pool of mutations.

I can read from your questions that you don't grasp what I've outlined above.

All of this has been in progress for years, and I've outlined larger details in many older articles.

I'm putting up a new post in a few seconds covering a bit of the details, and there's a link to the deck currently being worked on, with a list of mutations.

What I'm actually doing is that I'm taking a natural process
(Players all over the world make minuscule changes to decks)
And setting it on turbo.

I've been doing this for so many years, and for me it's been more of a matter of improving the process than actually winning in tournaments.

I've reached the 50/50 winrate in the past on many other evolved decks, so I know the process works.

The key for me is to improve the speed at which I can evolve designs, and right now I'm aiming at evolving 30 halfdecks at the same time, to see how long time it will take me to reach a 50/50 winrate.

All testdecks are old proxies from 2018, but I think magic as a game will contain enough key elements that it won't matter too much.

I can always finetune if I fail to hard when I take it all to reallife tournaments.

Why produce a system that sometimes produces inaccurate flawed results ?
Ask your dna :)

Evolution produced us, so why not use it...

I'm not arguing against evolution as a deckbuilding tool. I think the survival of the fittest cards is a pretty interesting concept! Sorry if I come across as too aggressive. :D
All I'm saying is that tracking duplicates individually instead of tracking all copies of the card as a whole doesn't give you any useful data, it just produces errors. And no, those errors can't be compared to mutations in evolution, since these are errors in your measurement method, not a "error/mutation" in the deck.

It's a really strange thing that I can read data from these points then.
I don't remember how I developed the system, but I know that at some point I reached out to a lot of people trying to understand exactly what it was that I had, because classic math says I shouldn't be able to gain data from this, when we speak of normal statistical structures. However, magic is a game meant to break statistical rules, which might be exactly why I can get a read from it.

I've only ever come close to a function describing my system when comming across an article about distribution of riches described mathematically, it was an article where the authors showed how rich people got richer described as a mathematical phenomenon.

There is also data points in the way that points are dealt to cards and subtracted from them, though it's generally "invisible" because I avoid writing all of the rules involved.

The real hard thing is to understand that when you suddenly track cards, they gain a behavioural dimension, every time I've done this, if there are 4 cards, one card will become a low score, one will become a top scorer and the two remaining will settle for middle values.

Apart from that article of distribution of riches there is nothing in math that describes why that card gets less played, but I use it nonetheless to choose which cards to cut, and for some reason, removing what my system measures as bad, and putting in what my system measures as good improve the design and winrate. Theres a small difference in real winrate and during the process winrate that differs 10% because during the process I get to choose exactly the card I need to play which means I get a top draw out of nowhere, then later the winrate drops as more and more cards are assigned permanently. In tests where I measured the final results all over again I discovered that the winrate drops 10% when everything is settled. Over the years I've gotten used to just measure the results, and then cutting 10% of the winrate.

Part of the "hidden processes" there are some stuff that sticks out.
For example some cards need to be measured differently, to give an example, chromatic star was a card that almost always got cast, but I could have been measuring it's impact in much more detail, which is why I during the mutation phase will be performing a second read of it to learn exactly what it got it's points from. One part of the process was that I cast the card, but I still didn't give it a point before I cracked it. And I also need to learn "why" I cracked it. Did I do it for a draw or did I do it to color a mana, because overall the card costs two mana to perform it's service, and I want to measure which of the two services it performs the most. It also becomes another data point, as I get to read from the individual cards what the destiny will be of those cards played first and that of those cast later.

I guess that might be the real data point that I read.
During early games there will be cards that take precedence like the cheap lifegain creatures, then they are followed by the slower cards, and sometimes when needed I cast removal. Like I've described, the first porcelain Legionnaire is the one with an extreme amount of points, which is because I usually spend my turn 2 on casting one of those, paying life in the process, and in games where I'm manascrewed and life becomes important I play the rest of the legionnaires less. That same thing applies to vault skirge which lategame can be too risky to play compared to healers hawk.

Back to the hidden processes.
I have tried to figure out why the process favors coincidental combos and scorns planned combo.
When I was trying to use my tool to hybridize original decks by measuring them and cutting the "bad" cards key combo cards often yielded to draw based tech, because you cast the combo as the last card, so all other cards were in essence measured first, sort of how chromatic star has been cast at a higher rate because it draws.

It's a weakness in the system that I'm trying hard to crack.

Just before I started on the halfdeck project I was trying to measure my 2018 cheerios. And the combo was skewered. A deck full of 0costs that does nothing isn't strong, so the way to measure that decktype needs an overhaul.

I've been considering to give an added bonus to the cards that kill and their enablers, perhaps a fraction of a point, but it will take many experiments to generate a universally fair measuring system

If chromatic star turns out to be used most for the carddraw, some could be changed into relic of progenitus, while others could become better colorraters.

"It's a really strange thing that I can read data from these points then."
You're not actually reading anything but statistical "noise" when looking at the differences between duplicates.
The human brain tends to try to notice patterns even when there actually aren't any. Sometimes, of course, you might get lucky and improve the deck based on the noise.

"The real hard thing is to understand that when you suddenly track cards, they gain a behavioural dimension, every time I've done this, if there are 4 cards, one card will become a low score, one will become a top scorer and the two remaining will settle for middle values."
Here's another example of seeing patterns when there aren't any. The cards do not gain a behavioural dimension, that's superstition. Let's say you have a 60-card-deck with 4 Lightning Bolts. If you draw 15 cards during the game, the expected amount of Lightning Bolts drawn is 1. Sometimes you will draw 0, sometimes more, but the average will be exactly 1 Bolt in a game where you draw 15 cards.

So, if you are tracking individual Bolts across multiple games, there will be a specific expected amount times a copy of Bolt will be drawn. The expected value will be equal across all four Bolts. Most of the time, the actual number of times each Bolt gets drawn will be close to the expected value, however, with a smallish sample size there will also be some deviation from the expected value due to random chance. This is what you are seeing: two copies were close to the expected value, while one got drawn a bit more than expected and another got drawn a bit less than expected.

Your system overall, as far as I understand it, will always favour cards that are easier to cast as they have more chances to get cast and score points during the game, regardless of their impact. I personally would only do comparisons between cards with a same role and CMC. Think of it like two alleles of a gene competing or two animals competing for a specific niche in the ecosystem.

It's never "a bit more"
Theres never a deviation from that pattern, which is why I've written about it so many times.
I remember having this type of conversation a lot in the past.

If something is consistent, and never deviates, then it's a solid pattern.

And it's something that has never deviated.

Those numbers I've put up here, follows that pattern, and I can link to plenty of other times where I put up similar list with the same pattern.

When you say there's no pattern there are four possible reasons, you are lying, I'm lying, or ordinary math is not covering this scenario, or someone has been messing up the math that you are taught, either by lying or by mistake.

Given the amount of trolls in here who tries to mess with me, that leaves two scenarios

And these samples, they follow your rule of playing copy #1 first if you have multiples in your hand, right? When using that rule, it will also mess with the expected values and is very likely the result for the pattern you noticed.

If you simply picked the copy to play randomly instead of following the number rule, as your sample size grows, the differences between the duplicates would become percentually smaller and smaller towards 0.

EDIT: if we are talking about only cards drawn, the second point still stands: the pattern starts to fade away as sample size grows towards a sufficient size.

Given that I started out asking you how you would perform a measurement, and you answered with the method I use myself it seems like we've had this conversation before.

The question is why you would use my technique if it's flawed.

Second, you still deny that there is any pattern to the stuff, without showing interest in the article I mention, probably because you are the same guy I told this to last time we went through with this farce, so you are either distracting me from something or putting a smear on my work, which raises the question, who is it that you hope to convince my work doesn't work.

It's either gas lighting again, which we've been through a couple of times, or someone new is on mtgvault, which might be why I was offered a bribe to stop writing about thieves.

I guess your trying to open up for business again.

Well, I've stated a price.

"Given that I started out asking you how you would perform a measurement, and you answered with the method I use myself it seems like we've had this conversation before.

The question is why you would use my technique if it's flawed."

You're talking about Bolts and Goblins right? That part of the technique is not flawed IMO, since it doesn't contain keeping track of duplicates which is the main thing I'm criticizing. I haven't discussed with you before, so you're thinking of someone else.

"Second, you still deny that there is any pattern to the stuff, without showing interest in the article I mention, probably because you are the same guy I told this to last time we went through with this farce, so you are either distracting me from something or putting a smear on my work, which raises the question, who is it that you hope to convince my work doesn't work."

I'd be glad to look at the article if you want to link it! Though I'm skeptical of whether accumulation of money and drawing cards from a deck can be compared to each other. I'm not trying to convince anyone that your work doesn't work or is worthless, just pointing out a methodological flaw that could be easily fixed.

By the way, the numbers for Chromatic Orb in this deck seem to debunk your pattern, don't they? Chromatic Orbs #1 and #2 grouped at the top, while numbers #3 and #4 grouped together lower. This isn't what your pattern was supposed to be. Of course, these are only the points that they scored, but if you want to show me the data of the draw counts then I'm happy to take a look.

???
You are right, chromatic doesn't follow the usual pattern, and neither does the rest of the stuff...
That's pretty wild.

Something is different, and it's pretty obvious to me what the difference is.

In the past I've only used this technique with normal decks.

This time I've used halfdecks where two different decks sort of battle for space in the deck.

That might be the cause of this new behaviour and it's an interesting phenomenon I'll have to look out for.

Regarding the trollery, there's a standard pattern where a new account is opened up, and out of all the active accounts in here chose to open up conversation with me, and after a few openers of conversation starts to say I'm wrong.

Theres a second less used pattern, that of an old account having been dormant for several years starts pumping out decks following another certain pattern of adding new types of decks.

When discussing it with friends I've called it a deck-bouquet because it has an ornamental look to it, as if the user cares more about how their deck collection looks, rather than caring for a continual building on a single favourite decktype.

Come to think of it, there might be another reason as to why the numbers been different
At the time I listed the numbers above, I still had 10-12 decks left to play, and I've described how everything seems to fall in line with the very last deck.

I'll list the complete points and list the old values besides each card, maybe theres something interesting in there.

It's been a while with the "why the rich gets richer" article but I'll see if I can find it.
It has relevance because of the way that they represented random transactions which seemed to mimic the way I put points on my paperstrips.

Someone's also spamming while we've been having this talk.
Multiple necroing of tylowrath's older posts, and a1b2c3 writing good in many posts.

You take this odd situation rather calmly for a newcomer :)

Whose your mighty rider by the way ???

The "pattern" was different this time is simply because there never was a pattern here to begin with. It is statistical noise resulting from a sample size that is too small. As you approach sufficient sample sizes (for example, 1000 starting hands might be enough), you will notice that the differences between duplicates become less and less significant until they are basically nonexistent. The "behavioural dimension" is simply an illusion, and as we noticed the "solid, never-deviating pattern" you proposed was debunked easily by this very deck. Clearly the playsets do not always get grouped that way. But yeah, I don't think I should debate this further, I'm just wasting both of our time.

Keep doing your thing, but please consider that maybe, just MAYBE you aren't making findings that would completely break all laws of probability and statistics and instead are just drawing way too grandiose conclusions from tiny sample sizes.
Occam's razor.

I've been browsing Vault for some time but haven't had an account before. I'm not trolling, I just happen to be bit of a besserwisser :D But yeah, I noticed the spam by a1b2c3 as well. The reason I made this account is that I was thinking of posting some "encounter decks" that serve two purposes: they are playable in MtG but also function as random encounter tables for DnD in Theros or Ravnica settings. They aren't finished yet though.

I don't understand the mighty rider question, are you referring to my username? (Agnomakhos was a mighty archon on Theros who rode a flying bull)

Let's assume that it's random noise, then what exactly keep the decks themselves from becoming utterly random.
When looking at what got cut in the end, there seems to be some sort of pattern to that as well, how is that unexplainable?

As for the anomalies being debunked, I believe I mentioned a difference in measurements this time around, and have to wonder why you three times neglect possible explanations.

Sure, discussing this with me, won't make me stop believing in it, unless you can give some very concrete examples of why it doesn't work as I see it.

I've been using the method to evolve decks forever, and they do improve, which is very much why I still use them. If it didn't work I'd abandon it within days.

I don't know if you've ever noticed some of my own encounter tables when I've discussed them in here, as they are just as advanced as my deck building.

I googled your name, and I think agnomakhos was actually the beast.

I'll try one more time to explain my point :D

"Let's assume that it's random noise, then what exactly keep the decks themselves from becoming utterly random.
When looking at what got cut in the end, there seems to be some sort of pattern to that as well, how is that unexplainable?"

You misunderstood me here. I'm not criticizing your method as a whole. Your evolution principle could be really handy when comparing, for example, the usefulness of two similar removal spells. I could put a playset of Ultimate Price and a playset of Doom Blade in a deck. Then I could count the times I drew a Doom Blade and the times I drew an Ultimate Price (as there are 4 of both, I will on average draw an equal amount of each card, but due to my small sample size there will be differences and thus that needs to be accounted for). Then I could also count the times I cast a Doom Blade and the times I cast an Ultimate Price to see which one gets more use. This is all useful data, NOT noise! When applying this principle, and removing one copy of the card that more often waits in the hand unused, the decks are not going to become utterly random.

HOWEVER, if I were to start comparing the times I cast Ultimate Price copy #1 compared to Ultimate Price copy #2, that would be looking into completely meaningless randomness. In one game I might draw Ultimate Price #4 first and cast it, then draw Ultimate Price #2 and cast that one. Everything would be exactly the same if I were to draw and cast Ultimate Price #3 and then Ultimate Price #1 instead. That would be just as likely, and that is a FACT.

"As for the anomalies being debunked, I believe I mentioned a difference in measurements this time around, and have to wonder why you three times neglect possible explanations."

I neglect your alternate explanations because I have studied statistics and these are very, very basic principles of how probabilities work. A coin (that is not weighted) is going to land on heads 50% of the time. Sometimes, in a tiny sample of 10 flips you might get 8 heads and 2 tails (80% heads), but that doesn't mean that the probability not 50%. In fact, a distribution like this would happen 4.395% of the time. The 30% difference between the expected value and our result is due to random chance.

Were you to continue flipping and made 1000 flips, the result could be 507 heads and 493 tails. Now it's 50,7% heads, so we are approaching the true probability as our sample size grew. The 0.7% is the random noise that is still not eliminated completely. Getting 800 and 200 with a sample size this large would be astronomically unlikely, basically 0% chance.

The same goes for MtG cards. In my sample of 10 games I might draw Ultimate Price #1 4 times, Ultimate Price #2 6 times, Ultimate Price #3 3 times and Ultimate Price #4 only 2 times. Similarly to the coinflips, a sample of 10 is going to have a lot of randomness, but a sample of 1000 a will have much less.
There is no point in experimentally tracking the differences between these duplicates.
There is a point in tracking the difference between how often an Ultimate Price gets cast compared to a Doom Blade.

I'm not saying your evolving method doesn't work! All I am saying is that there is a component in your process that gives you extra work but doesn't do anything helpful, that is, tracking the duplicates of a card separately instead of as a whole. You should apply the principle of evolution here as well: remove the part of the process that is redundant. Your process will still produce useful results despite the flaw, but it could be better.

And nope, Agnomakhos is the tyrant, his flying bull is unnamed. Kynaios and Tiro defeated him in the past. But cool that you researched my username :D I'm gonna take a look at the random tables you made!

In reverse order:
I research all names because I like to see the symbolism at work in mtg because trolls, psychopaths and aspergers syndrome have a common fascination for mythology, and I'm researching on all of them.

In my oppinion true statistics is basically an illusion, because by default, all real coins are "weighted" through wear and tear down to an atomic level, which is why lottery balls are extremely controlled and is only used for a certain amount of time before they are trashed, because they start falling in the lottery machines at a predictable pattern.

I do believe some of the behaviour in my measures are down to wear and tear of the cards, the smallest grease stain makes two cards cling together and I have a bunch of routines that includes as much randomness into my shuffling.

I get that statistics have it's own views of how magic behaves, but having spent the first half of my magic life on simulations I started to deviate from other similar minds at my time and as a result I managed to produce better results than my competitors, which is why I'm very wary at people who see statistics as the ultimate viewpoint. Sometimes a bit of intuition can crack through many barriers that others try hard to work around. At a time in my life I
Had some actual talks on the web with frank karsten about statistics and simulations and how I viewed it.

I believe that the exact way that I generate points on the paperstrips by making them individual is the only possible way of getting the paperstrips to cut an appropriate number of cards from the deck, because if all individual cards scored the same as the other copies, you would have to remove the worst performing copies rather than actually individual cards. It may be a function of adding increased richness to some cards, but there are singletons, that despite being singletons, (and therefore gets drawn a lot less) create a border between copies that would otherwise always be fully above or fully below that card.

It's an artificial structure a tool, that creates a specific effect in the way I get evolution to build.

How would you get around the problem of having to measure cards that would produce the same values ?
How would you know what to cut ?