Mechanics, themes and archetypes that need support to work

[lujo]

Instead of breaking down and being irritating and unconstructive, or adding cards to a huge wishlist, let me try this: I'll make this topic specifically for mechanics. I'll list and write up every mechanic that appears to have insufficient support in the current pool or Manalink, and do my best to explain what the mechanic needs in terms of AI and individual cards. Also what it needs to make non-lethal or overwhelming decks with it.

The important thing is that if any of you guys would like to know more about a mechanic and deckbuilding with it, you can ask about it and I can write it up. If I'm familiar with it, ofc, I can't claim to know everything. Also, obviously, if anyone else has spotted something unobvious but missing that would help a mechanic out or have been trying to make a deck for the AI but it turned out the AI missplays it or stuff, feel free to add insight.

I'll cross out stuff as it gets done, there's no real rush - it's meant to be a list of beneficial concequences of deciding to pick this or that to work on, or conditions for making this or that truly worthwhile. In terms of AI deckbuilding.

---

GENERAL

Mana Fixing / Mana Bases:

EXPLANATION: Mana Screw and mana flood can happen to the AI. Adding too much lands cuts into how much room you have for other cards, trying to fix one or the other by upping land count inflates chances for weak hands or decks running out of steam. It also limits the amount of non-land cards in a deck, or consistency (if you trim the number of copies for individual cards).

UNOBVIOUS:

Mulligans | Open

- Many strategies, themes and flavors, and most famous decks which have been ported wholesale live and die by mulligans. Guides for playing most famous decks (or particular matchups) start with "mulligan every hand until you have card X or minimum hand size Y". A lot of what was ported doesn't play (or build) remotely close to how it would in real life because you can't mulligan.

- Decks are forced strongly into alpha duals + fetchlands (the absolute best mana base in the history of the game) to jack up the odds of having a playable hand. That mana base then makes decks too flexible in other respects.

- Whole archetypes get bent out of shape - weenie decks or just decks with a low mana curve are forced to play so many lands that it defeats a purpose of having a weenie deck in the first place. The specific benefit is that the deck could run more cards in place of lands making it more consistent, and another is that it runs out of steam more easily.

- Giving AI lands which produce only colorless mana will have the AI stuck with a land an unable to mulligan an effectively unplayable hand. It also makes giving the AI lands which don't produce mana a terminal liability.

- People trying to build decks for the AI who aren't fully aware of what not being able to mulligan means are likely to make non-functional decks simply because of this. It's not their fault, as you could be a MtG player for years and years, even a good one, and not have a clue how decks with old mulligan rules would look like - because you've never seen one and all decklists you have seen/made in your life have been made with mulligans taken for granted. It's very difficult to shift perspective, too, as cards have been designed with mulligans in place for longer than with just one mulligan.

- Insert_here. It's so huge that it could have any number of unforseen concequences.

WHAT HELPS:

- Mulligans. Enabling either previous or current mulligan rules would help this out greatly, along with sorting out numerous other issues. It's such a huge deal.

AI for cantrips | Open

"Cantrips" are cards which have "draw a card" tacked onto them in addition to a different effect. An example would be Defiant Strike. There are other variations of this, such as "slow cantrips" which draw a card at the beginning of the next upkeep, Scry cards which allow you to filter the top of your library, and cycling (and also possibly other effects). Playing cards like these helps decks out draw into lands, or draw more cards as they play as it lessens the effects of mana flood.

The current, NB 1, AI missplays these cards by valuing their Draw a Card or Scry aspect over their other effects. This is understadable, but what often happens is that the card gets played at wrong times thus distupting the way the deck was built - giving the AI a Defiant Strike is meant to give it a combat trick which happens to help it out, not a card draw spell which happens to occasionally be a combat trick. It takes a different slot in a deck and forces the builder to give it another combat trick which will be reliably used as a combat trick - but the deck then has 2 combat tricks and it probably doesn't want more than one. And one is getting played at a time when the AI should be playing other things, thus using up the mana and messing up the creature curve and expected gameplay.

So the AI can't effectively use these very useful and highly prized cards, and it's also liable to screw up whatever it was supposed to do otherwise because it has them in the deck. Which then removes an important tool for making AI decks more immune to mana screw and mana flood. It also forces a deckbuilder to make other cards in the deck more powerful, more mana efficient and often needlessly overwhelming (particularly low tier decks).

Another part of the problem is that cantrips are rather absent from either the mana-link pool and (sometimes consequentially) from the Shandalar pool. This also hurts any archetype that specifically benefits from them, and those can be easily identified by which block they are from - cards from blocks with a lot of cantrips (or colors which are suspiciously loaded with cantrips in a block) need these cards in large quantities and working to make any sense. Not just the ones which made it into tounament winning decklists, but all of them - those cards were in their blocks or a reason. Or multiple reasons. And plenty of mechanics from other blocks benefit from them, too.

Card Draw | Open

The main utility purpose of card draw isn't to draw into answers or key cards, it's to not get mana screwed / flooded, and help recovering from agressive mulligans. This may sound counter-intuitive, and there's plenty of card draw which obviously refills your hand, but it's true. Having a wide variety of card draw (basically every card ever with "draw a card" on it, the cheaper and more mundane the better) allows for a wide variety of decks to actually work. Yes, a lot of decks haven't had any card draw in them - but a lot of card draw those decks might have played wasn't available to them in their original environment. There isn't enough of that sort of thing (even if it looks like there's quite a bunch), and as far as making the huge pool useful getting it all in and playable sensibly by the AI would be a huge deal.

WHAT HELPS:

- Getting the AI to play cantrips right - they're not there for the card draw effect (unless it's a card draw card which also cantrips), they're there for the other effect AND card draw. Yes, playing it madly like the AI does now helps it draw into lands, but it does more harm than good because there's not enough slots in a deck and cantrips get played in removal/combat trick/whatever slots. The AI has to play these cards as if the other effect is the primary thing, and card drawing will happen on it's own and help it.

- Getting the AI to play spells more sensibly - meaning not madly prioritize them over developing the board. Because a lot of cantrips have their place in creature decks in colors which otherwise don't get card draw. And these decks prioritize developing the board more than wildly spamming out spells.

- Porting in every single easy-to-port-cantrip into Shandalar, Manalink be damned (or both, or whatever). They're the biggest boon to deckbuilding there ever was, and the power level of the actual effect is irrelevant, there are many tiers of decks in Shandalar, and they also help in scaling down the power of higher tier opponents.

The List

This is NOT "my whishlist", and I'm not asking this because any of these cards hold emotional value to me, these cards being in the pool and being properly usable by the AI means stuff works. If they arent - stuff doesn't or not nearly as well. Having them in and working right should shave hours off proper deckbuilding, allow for more things to work and be scaled properly, and give sense to existing and to-be-implemented cards:

In Manalink already but not Shandalar: | Open

Repeal
Rending Vines
Explore
Flux
Foresight
Hunter's Prowess
Omen
Ponder
Portent
Reprocess
Scarscale Ritual
Skulltap
Tormenting Voice
Wild Guess
Altar's Reap
Bone Harvest
Force Away
Gorilla War Cry
Perilous Research
Provoke
Snakeform
Sway of Illusion
Twisted Image
Hunter's Insight

(Note, I didn't go through Scry, mark Kicker, Cycling and similar things, although it stands to reason that those categories include cards with the same function as this type of card. I probably will at some point.)

Not yet in Manalink:

Worth porting under all costs: | Open

(Note - this might need a bit of sorting, but the red, green and white cards here are generally very important no matter how dumb they look in a vacuum.)

Zap
Withstand
Viridescent Wisps
Twitch
To Arms
Sudden Strength
Stun
Spiritualize
Solfatara / Turf Wound
Second Thoughts
Reviving Dose
Restrain
Repel the Darkness
Quickchange
Pressure Point
Needle Drop
Manamorphose
Leap
Lace With Moonglove
Jinx
Instill Infection
Inspiring Call
Inside Out
Ignorant Bliss
Hydrolash
Stream of Unconsciousness
Guided Strike
Gallantry
Afflict
Balduvian Rage
Accelerate
Urborg Upraising
Symbol of Unsummoning
Manipulate Fate
Nighthaze
Footbottom Feast
Flash Foliage
Flare
Festival of the Guildpact
Curtain of Light
Fevered Strength
Feral Instinct
Cease Fire
Carom
Boiling Blood
Cremate
Confound
Recover
Bewilder
Bandage
Aura Blast
Argent Mutation
Touch of Death
Irresistible Prey
Last Caress
Unhinge
Renewal
Due Respect
Painful Truths
Implode
Healing Hands
Life's Legacy
Magmatic Insight
Jolt

Wouldn't necessarily prioritize: | Open

(Some of these might just not be needed, or I haven't thought about them in the right context)

Infuse
Panic
Winnow
Vivify
Telim Tor's Edict
Ray of Erasure
Niveous Wisps
Manamorphose
Lamastide Weave
Heal
Gravepurge
Gravebinding
Grave Birthing
Force Void
Filigree Fracture
Equal Treatment
Ennervate
Due Respect
Dredge
Crimson Wisps
Clairvoyance
Chaotic Strike
Cerulean Wisps
Artificer's Ephiphany
Aphotic Whisps
Aleatory
Aggresive Urge
Touch of Invisibility
Prophecy
Cloak of Feathers
False Dawn
Forget
Magmatic Insight
Mind Spring
Sorcerous Sight
Formation
Foxfire

Includes Mechanics not yet in Shandalar: | Open

Although some of these, particularly the red ones are worth porting/hacking a simulation for the mechanic in entirely on their own, like the red card draw ones:

Morsel Theft
Refresh
Refocus
Reach Through Mists
Foreshadow
Desperate Ravings
Dazzling Beauty
Faithless Looting
Inaction Injunction
Research the Deep
Train of Thought
Aura Finesse
Dark Dabbling

---

That's the big thing. That would also help a bunch of smaller things, especially from blocks which featured cantrips, or blocks which were in Standard along with them.

[lujo]

Now for smaller things

(Writing up a bunch of smaller, concre things, reserving post)

[Korath]

Making the AI more reluctant to cast cantrips will tend to make it even more eager to cast removal prematurely, and vice versa. Fixing both is a whole lot less straightforward than you'd think.

Whatever you're using to figure out whether a card is in Manalink or not, it's not accurate. Sway of Illusion (not in), Manamorphose (in), Inside Out (in), Balduvian Rage (in), and Aleatory (in) are all wrong, and those are just the ones I can recognize at a glance.

It's not really helpful to try to predict which cards are easy or not to port to Shandalar. For about a third of the cards you list in the "Mechanics not yet in Shandalar" section, I can't even tell which mechanic you mean; and for about another third, the mechanic is simple enough that I wouldn't even have to stop to think to implement it separately. Even being in Manalink doesn't mean that a card works well in this engine - the AI deals with Manamorphose even more poorly than the cards you've been complaining about; Gorilla War Cry and anything else that has or grants Menace (or its 3-blocker analogue seen on things like Pathrazer of Ulamog) sort-of kind-of works when the human is blocking, but not at all when the AI is; and don't even get me started on things like Cavern of Souls where even the best possible approximation in this engine is so poor that the card should never have been enabled. Just let me do the triaging.

Except for a handful of cards in Portal Three Kingdoms, Unglued, and Unhinged with unique orthography, every card in the game should highlight if you spell it correctly. Goblin Hero and I went to a fair amount of trouble last year to exhaustively verify everything up to the end of Theros block, and I haven't seen anything printed since then that seems at all likely to confuse the card highlighter.

[lujo]

I'm sorry for the spelling, that's my fault.

I'm also sorry if I've mistakenly put some card into the wrong category, also probably my fault. I was using the deck builder (not the --deckbuilder), but I might have messed up.

As for easy or not easy - I wasn't trying to predict anything. Just separate things which are in manalink from things which aren't (to the best of my obviously limited abilities).

In case of Manamorphose it is more often used just as a card which doesn't cost mana to play and replaces itself. I've heard you say the AI handles mana fixing terribly, but having it just spammed madly in any mono-red deck whenever the AI has mana for it and it in hand would seriously reduce headaches with not overolading Red decks with impactful cards, and help it have less mana screw issues* (and obviously help anything that likes having cards in graveyards). *(Not color-screw, but, like, "I'm not drawing lands" kind of issues). It's the same reason I put in Seething Song - in theory it's something the AI can't possibly mess up in a mono deck. I wouldn't even attempt to use it for actual mana fixing. If the AI could always make 2 mana of the most common mana in it's deck (which I have not idea about), and it spammed it as madly as it spams cantrips in general, you could port it this instant and it would end up in at least 4 or 5 red decks immediately. *EDIT: Anyone complains about stupid AI not mana fixing right, you just tell them to go yell at me for that.

As for actually fixing stuff - I know that fixing either is less straightforward than I think, because I have no idea how complicated it is. What I do know is that the way the AI handles cantrips defeats the purpose of putting them in decks and causes a cascade of problems. It doesn't have to be fixed instantly, it doesn't have to be fixed gradually (if fixing one end would cause problems on the other end), it doesn't have to be fixed before the AI behavior with spells in general is fixed - it just really, really, really has to be fixed.

Both of these things just cause cards to not actually do what they are supposed to do. They stop being cards that you can put in a deck. Cards don't read "If the opponent has a creature on board you must play this card" or "If you have mana to pay for this card, and you have this card in hand, you must immediately play this card". If these things aren't fixed, you're not actually porting cards you think you're porting, you're porting stuff with very detrimental invisible rules text - as far as building decks for the AI is concerned. And if such cards are unusable, that then makes other cards, which rely on them being in decks, also not really be cards you'd think they are and decks not ending up what they should be.

I wish I could code to help out, do drugery, have money to hire someone who can to help, buy you sandwitches, do your housework. I really do. But this stuff just has to be fixed.

EDIT: Oh, and while we're there - the default AI thinking time in the .ini (the three digit one) is beyond too low. The effects are clearly visible if it's increased, took me about 4 turns of gameplay after quadroupling it to clearly see the AI go from "I can't use Goblin Tunneler at all, and have also been shot in the head" to "I can use the Goblin Tunneler, keystone and whole point of my deck, exactly as it was supposed to be used". Please bump it back up as the default setting, I would not have lost my mind (and possibly scraped dozens of workable deck ideas in the last month, and made much better balanced decks) if I knew about this on time. I am trying hard not to thing too hard about the idea of people playing at the default number and not noticing the difference.

[lujo]

I just remembered something that could help with getting the AI to play cantrips sensibly. The current problem is that the AI is highly likely to attempt to cast a cantrip at any time for no reason but to draw a card.

If you could make it "unaware" of the fact that the card draws a card, and only aware of it's other effect, it would still serve it's purpose. When they first appeared people didn't know how strong the effect was, they just played them (and occasionally grumbled about them being "strictly worse" than regular cards because they tended to cost more). It took a while for people to realize just how much the cards helped them. They put them in decks for the effect, just like a deckbuilder would/should put them in AI decks, and when the AI plays the card it will draw them a card and that helps it on it's own.

Like with Defiant Strike - I put it in there because having a pump in hand encourages the AI to attack. If it just madly attacked with it's dudes (like it always does) and cast it to kill one of your dudes (like it always does) or just cast it for no reason but to deal more damage (like it always does) it would've tricked itself into card advantage, the cantrip part would've spontaneously helped it's silly attacking limitatons/patterns. That's the beauty of cantrips - they help even complete noobs who aren't aware of it.

When someone's building a deck, he can just toss a cantrip in a slot and know that the AI will be secretly helped whenever they do X, but it's not worth making the AI not do X and just try to draw a card by any means necessary rather than X that causes insane plays and general deck idea/strategy meltdown.

So in short - don't nudge decks to play cantrips because they're cantrips. If that causes problems, then the problems are elsewhere, and if they're big problems - they're even bigger than it looks because it has nothing to do with cantrips and the fix can't and shouldn't involve them.

---

Now, as for the AI madly spamming removal and messing up stuff fundamentally, yep, that's a huge one. Why and how it happens mechanically is beyond me, but what should it be doing instead and how complicated/feasible that is to achieve - I can't say right now. I could probably help with "what it should be doing" if I knew what exactly is causing it to spam removal madly and why. Have you ever mapped out the AI in a way a layman could understand, like the point awarding system it uses for determining priorities and such?

For example, a layman like me would ask "If the AI doesn't have anything on board, why did it choose to bounce my Llanowar Elf instead of playing a 3/3 creature it has in hand." How does the answer to that translate into code? Or half way into code?

[stassy]

A partial answer to AI management in Shandalar by Korath can be read here.

If you take a look at the source code, you can see that it is possible to affect the AI behavior by using some conditionnal bruteforce orders in the card code itself, but that would require to review all the cases for a single card...

[lujo]

Hmmm, I'll take a look, but that might not result in anything because I can't tell an elbow from a backside when it comes to code. If were talking cantrips (that is, card which have card draw or filtering tacked onto what is technically a card with an unrelated effect, as opposed to card draw which is just card draw), they will all cause gross missplays one by one if the AI pays attention to the part of them that says "draw a card". They'll work very well for it if they just get played normaly completely ignoring the card draw part and a deckbuilder can easily make the AI trick itself into drawing cards and having a smoother game that way, or fill up all non-creature slots with them to help enable threshold in a card-efficient way, or whatever.

Since they'll all almost certainly cause trouble, and thus eventually need tweaking, why not just tweak them as a category somehow? They're treated as a category by the AI (that is they're treated as card draw), and they're very disparate when it comes to effect and can't be treated as a catgory in-match. The decbuilder treats them as a category when he chooses to put them into the deck, the player treats them as whatever actual category (pump, removal, fog, whatever) they belong to. Very rarely is playing them for card draw only something you have to go out of your way to do. If we could get it to not treat them as card draw, we'll still have the "AI plays spells stupidly" problem, but we won't have "AI plays some spells extra stupidly, and they all do different things, wth is going on" on top of that, one problem is better / easier to solve than two at the same time, right? I mean, you can, to a degree, with some spell categories like pump, help its stupidity by using cantrips. If it just plays them like pump or cheap removal, at least it doesn't lose a card in hand, this makes the situation quite a bit more bearable. If it's prone to go extra haywire with cantrips in a way which makes it use pump/removal as card draw instead of pump/removal (like it does now) then that magnifies the problem.

So, how to get it to treat cantrips as "not-card draw"? To fix this in a way which doesn't end up with a million reports and card-by-card decisions, etc.

*EDIT: I can, though, sort out the list of cantrips into "not-card-draw" and "really, this is here only as card draw / filler", with, say, Manamorphose or Sungrass Egg and such in the second category (meaning that the purpose of it really is just to be madly spammed whenever just to draw a card to smooth draws/fill grave/trigger prowess or whatever).

[Korath]

They're treated as a category by the AI (that is they're treated as card draw)

No, they're not. The AI has no concept of card draw. What it sees is that it has a common instant with a cmc of 1 in its hand (and a bunch of other cards) before casting a Defiant Strike, and a white sorcery with a cmc of 4 (and those same other cards) afterwards. That way outweighs the lowered value of one Plains being tapped that wasn't before.

[lujo]

Ok, that was very helpful. Now I know what I didn't before. That's also quite messed up. And technicallyeffectively they ARE treated as a category, it has no label for it, it just sees the effects of it, and in order to use them properly it shouldn't. Can card draw be worth something to it without it being able to see the actual cards? That's how players rate it, and also part of the reason why players attack into trades (as long as they're trading evenly or in their favor concerning the raw number of cards, they will generally attack, the AI won't). Also - can it be prevented from being able to see the results of playing the card that relate to card drawing?

And a very important question - why is the AI casting these things in it's post combat phase and not at the end of opponents turn? I've just seen several misplays with Shelter which probably wouldn't have happened if it held onto it until then. I understand why - it sees them as card draw (sees playing them as resulting in benefits of drawing cards), but how does limiting cards to phases work?

[lujo]

These are cantrip and carddraw-related aura's, as I've only sorted out instants. A lot of these are also sorely missed in general and many support existing archetypes. These aren't present in Shandalar (according to the cardlist), haven't tried to check with Manalink as I've messed that up last time. Most are well worth porting straight to Shandalar and would be a great boon to deckbuilders:

Card Draw/Cantrip Auras | Open

Special priority to black and red ones.

Abundant Growth
Angelic Gift
Bestial Fury
Betrayal
Capashen Standard
Casting of Bones
Dragon Mantle
Epiphany Storm
Fate Foretold
Frog Tongue
Fruit of the First Tree
Helm of the Ghastlord
Hobble
Illuminated Wings
Infernal Scarring
Karametra's Favor
Krovikan Fetish
Krovikan Plague
Lingering Mirage (cycling, but V handy)
Nylea's Presence
Ophidian Eye
Ordeal of Thassa
Pentarch Ward
Power Taint (cycling, but supports Psychic Venom decks)
Private Research
Quicksilver Dagger
Ritual of Steel
Savage Hunger (cycling, but nicely minor buff)
Scavenged Weaponry
Sisay's Ingenuity
Snake Umbra
Stratus Walk
Stupefying Touch
Tainted Well
Traveler's Cloak
Underworld Connections

---

For supporting Psychic Venom / Contaminated Ground decks:

Gulf Squid , pretty core
Mistbind Clique (if it can be done, especially since Pestermite is core in those decks)

---

Very quick rundown of some (not all) missing essentials for threshold without writeup:

White land sacrifice, supports a lot other white and white/X stuff too:

Trenching Steed
Aura Fracture - this is actually in
Troubled Healer
Jeweled Spirit
Exalted Dragon
Reaping the Rewards
Pegasus Stampede

And it all combines with Land Tax, Tithe, Knight of the White Orchid, Knight of the Reliquary, Planar Birth, Terravore and and so on and so forth. Even if the AI can't use them, they are essential for the player if they want to make use of all the white threshold stuff in the pool on a playthrough. Also well worth porting straight in.

These two are quite necessary:

Millikin - Can't stress the importance of this guy enough
Sandstone Deadfall - Same, 3 cards in the grave, kill whatever

Red land sacrifice:

Tremble
Raze
Need for Speed

---

Dwarf Support:

They need Dwarven Soldier to really work at all, but can also make use of Heart Wolf.

---

Mercenaries moved from the whishlist thread - it's not so much as whishlist as a "there's this half-supported but really useful mechanic for black which has essentials missing I wan't to cry whenever I have to scroll through them" and also gives the player an option during a playthrough (and also ties into black threshold with the land sacrifice and generally sacrifice fodder themes obv.):

Agent of Shauku

Bog Glider
Misshapen Fiend
Rampart Crawler
Goblin Turncoat
Silent Assassin
Strongarm Thug

Spiteful Bully
Death Charmer
Phyrexian Driver

That's not actually ALL mercenaries ever made - but if half the ones which are in the pool weren't in the pool they wouldn't be on this list, but these would. That's the problem - the actual essential ones aren't there!

---

For Sea Dragon/Spreading Seas type decks

Manta Ray - It's very important for that type of deck, boosts it's performance dramatically, doesn't hose blue decks which would fight it because they're blue and provides a much needed cheaper alternative to Segovian Leviathan
Dreamwinder / Kukemssa Serpent - also double as threshold support obviously, and make it possible to build that type of deck with a much lower curve entirely.

---

Added whishlist stuff to the actual whishlist.

[jiansonz]

Power Taint (cycling, but supports Psychic Venom decks)

How does it do that? Looks to me like the controller of the target enchantment is the one that takes damage.

[lujo]

Remembered what I meant with it

It's like Power Leak but you can cycle it if there's no targets, so you can actually put it in decks like that, as opposed to power leak. With Psychic Venom decks you want to be able to keep the opponents mana tapped, and if you're gonna have an easy thing to cycle for lands it might as well be that.

[lujo]

I've been avoiding talking about this for as long as possible, but let's have it.

Fires of Yavimaya IS Red/Green. In the sense that no other card, not even Kird Ape, defined Red/Green historically as that one did, and pretty much the only reason it wasn't in more RG decks was because it wasn't legal in the format, otherwise it would have been. It's severely misplayed in Shandalar, and here's how:

1) If there's Fires of Yavimaya in the deck, and the deck doesn't have in on the board, it should get priority. The AI plays creatures instead of it and that's the wrong way around, but it should only madly rush to play the first copy. No buts no ifs no you don't understands, if it's not doing it, the cards should no be in the deckbuilder.

2) It should be played before combat, that part might even be working.

3) It should be very hesitant to sacrifice it unless there's several of them on board or for the win.

4) !!! It should bloody well be able to realize that all the stuff in it's hand has haste, and that playing creatures post combat is not what it wants to do.

5) !!! It should bloody well be able to realize that all the stuff in it's hand has haste, and that it should be playing creatures before combat.

6) !!! It should bloody well be able to understand that if it plays creatures before combat it's doing the only thing that's required of it, and play creatures before combat, otherwise it's missing the entire point of even being in it's colors in a format where you can play Fires of Yavimaya.

Right, sorry about the tone. I understand that it's probably impossible to achieve this with all cards being globally scripted but if there's one clear example of why that's unworkable bullshit - that's the one. It's been popping all over the place in various incarnations, I know, but Fires of Yavimaya decks have been "scripted" IRL, turn 1 - one of the 7 mana producers, turn 2 - Fires of Yavimaya - turn 3 4cc fatty (Usually Blastoderm or Iwamori of the Open Fist or whatever) - turn 4 5cc fatty (whatever), with a counter-fires variant of turn 1 elf turn 2 3cc elf, turn 3 fires, turn 4 Thorn Elemental.

If an engine would make it as impossible as it is to make this deck work, that engine is bullshit as it seems that it would be easier to just script decks individually. If my life depended on it I could call my demented grandmother over the phone and guide her through a MtG tournament (over the phone) with a Fires of Yavimaya deck without explaining the rules of the game to her or having her understand english or even what's going on at all. She might not win this tourney, she might, but she would not once fail to perform the script in a manner equal to that of a long time mtg player.

1) Why is the AI unable to execute Fires of Yavimaya properly? It's very important because the it won't be able to do a lot of things which are fundamentally the same only more complicated (and from what I've seen it isn't which is why I'm so worked up right now).
2) Is there anything that can be done about this, not to make it play it "better", because there's no better and worse, you're either doing it right every single game or you're not supposed to be playing MtG at all. You just follow the script and that's it, it's always the same. If nothing can be done - the system is utter crap, and needs work until it can. Otherwise asking someone to make sensible decks in this engine is even more insane than me asking someone to code up a proper engine, except I know how insane it is of me to ask this, but the other person probably doesn't know what they would be asking for from any deckbuilder. Because if they did, noone would have to be asking anything. It's a crazy wonder so many things work out (somewhat) at all.
3) Is is possible to allow individual scripting for decks? A way for deck X to override global scripting for individual cards or card types while it's being played?

Honest apologies for the tone. I'm known to advocate porting cards that aren't necessarily fully functional, but this is ridiculous.

---

Oh, and a sidenote - how come the AI sees what cards it will draw anyway?

That can't stay that way because it causes the AI to evaluate things using values of other things in a way that makes it impossible to predict what it's going to do with almost anything that draws cards. If it was like that originally, well, that was bullshit, needs to be different. If it wasn't but was changed so it would improve something else - that something else needs a different kind of fix.

My 12 year old brother could play MtG without peeking at cards from the top of the library to see when to cast a card draw spell, and could barely read english. He won tourneys (with Fires of Yavimaya). There's no way the AI absolutely needs to be able to see cards it would draw.

[lujo]

Here's something less angry and very important on this topic:

That card, that combination of colors, that deck and many other cards and decks don't in fact need an AI capable of playing MtG in general - they need a bot. That's the big problem - most interesting decks and stuff don't in fact have all that much to do with MtG in general, they're just scripts to be performed in a certain way. For Fires of Yavimaya it's - play Fires of Yavimaya - play creatures - turn them sideways. There's nothing else to it, no logic or exception - your creatures will be green and therefore above the curve, you'll always trade favorably and you don't mind trading, you don't have to think or know most the rules, or anything, you have to perform the script mindlessly and you either win or you couldn't have possibly won. You can't have outside, common logic interfering with the execution (like removal being played at the wrong time, attack being dictated by having pump etc). What the AI is supposed to do is determined down to a card by the deckbuilder, all it has to do is go through the motions and not attempt to think for itself too much.

A hundred things I've tried to make work can't possibly work if general wisdom is applied - that's why they're interesting, or well known, or work, they got their own script and the perform it and that's that. They're meant to be played with a plan already almost set in stone, the player can't be trying to figure out the deck at the time they're playing it or have the cards scripted according to their general usage. It has "that" card in that place in that deck and means to use it in that exact way and that exact sequence, no ifs buts or any outside context. That's why in RL magic "nobody blocks" - blocking is interacting with your opponent instead of doing your own thing, and you have cards in your deck for that (removal), if there are creatures in there they're supposed to be attacking (apart from ones which are there just to block, but then it can't happen that you don't block with them, you know what they're in there for).

This is why the AI sucks so hard as it does - you guys are trying to go along with it's hugely flawed plan of scripting cards outside of very particular contexts cards in actual decks are used for. There's a limited number of cards which can actually be scripted in ways which are applicable in more than one deck. And even then every deck will prioritize creatures and removal differently - but not according to the board most of the time, but according to a pre-made plan the guy who made the deck made before they started playing. One guy will throw out Arc Lightning as soon as there are 3 mana on board because he's playing a burn deck. Another guy will see if he has Fires of Yavimaya or a creature first. A third guy will only ever play it on an evasive creature because that's his only way of dealing with them.

That's how magic works, but this AI can't even theoretically play that game, because it has no idea about which deck it's playing and what that deck needs done in what sequence, and you can't set that globally. It takes hours and hours and hours of fiddling with a rather lousy deckbuilder and praying that you manage to put a deck together in a way which makes it impossible (to a degree) for the AI to miss the point of what card is supposed to be doing what in it's deck. Or you can give it cards which do everything at once, like Flametongue Kavu, but that just shuts down the players options and doesn't solve anything - decks made with those kind of cards that I have in Shandalar right now are completely unplayable against. Or you can do your best, like I did, to figure out what the global scripting boils down to and only make decks for it which fit that particular mold, but that's a very limited combination of samey decks. I've made about 50 semi-finished decks and most of them are the same deck and most of them are bent out of shape because what the AI is doing only occasionally intersects with MtG.

I've scraped about three times as many because they weren't that same deck. And those would work if I could just script the AI, when playing those decks, to do it's thing - this is for this, this is for this, turn 1 you do this, turn 2, you do this, turn 3 you do this, you cast this if there's this on board otherwise you don't, etc. Those wouldn't and can't possibly be global rules, and the AI wouldn't need to know, for example, what's on top of it's library - it would know when it's supposed to cast card draw regardless of what it would draw. These creatures block, these creatures don't block unless you'd die, etc etc. That's the only way it can work.

Heck, a rare way to make an interesting deck is to take a card like Battle-Mad Ronin, which is forced to be played similarly to how creatures in MtG are usually played and then you have a deck which you can sort of mold to play out a script like a MtG deck usually does. I have a feeling that if you just tacked on "must attack every turn" on every creature it would be possible to make more functional decks than it is now, simply because the AI then can't avoid doing what a deckbuilder can actually work with.

How difficult would it be to make it possible to script individual decks? Have them come with overrides for individual card AI? Nothing you could do to the game at the level of match playing, except adding proper mulligans and enabling players to [disable ante cards (the ones which mess with ante during a match) and not remove cards determined as actual ante from decks until the end of the match], is anywhere close to that.

[BlueTemplar]

Ah, the choices in "AI" between scripting and more generalist...

I'm willing to bet this choice was picked because it was Sid Meier that worked on Microprose's MTG, who made Civilization before that, and I'd be surprised if it was a scripted game (unlike, say, Starcraft 1, which had scripted build orders for its skirmish AI's...).
Or maybe just because unlike the "generalist weighting" (or whatever the AI is using now), scripting doesn't work at all as soon as you change game's "rules" by giving the AI another deck (remember it was supposed to play any deck you gave it in Duel mode).

Then also at that time hardly anyone knew about the various Magic deck "archetypes" and "metagame" as they are understood now.