Added creatures with low power that have
-abilities that tap for something really valuable like drawing a card
-abilities that usually need to be saved until enemy turn (counter a spell, tap a creature, prevent damage)
-abilities that usually deal equal or more damage if not attacking (prodigal sorcerer, grim lavamancer)
Generally avoided abilities where activating before combat is an acceptable choice and thus the attack decision doesn't come up anyway as the AI already used the ability.
However, this behavior is unacceptable on certain abilities (mostly those with higher mana costs) which should not be used before the end of main 2 as it prevents the AI from actually playing their cards. (like Hanna Ship Navigator)
I suspect this will need quite a bit of improvement in the future but we have to start somewhere.
The AI will now not play a second copy of these cards if and only if, there is no copy of the card owned by that AI in play, except copies that are enchanted by Auras owned by other players.
Later, a better logic that actually determines whether the enchanting aura removes the relevant ability from the card or not should be added - Warp Artifact will not stop a disk from activating, but Arrest will (if it's a creature).
Note : Only add NonStackingEffect to permanents the AI shouldn't play more than one of at a time if and only if, removal of enchantment is not pontentially leathal to the AI.
For example, do not add to Ensnaring Bridge or Moat, as loss of those allows the enemy to swing with possibly accumulated large armies for likely too much damage.
Those cards will need a different logic, that avoids playing a second copy if and only if the threat the card protects from is already too dangerous.
Most decks were tested in the old version (roughly 2017 October-November?).
They are AI playable albeit the AI doesn't play them perfectly - still it's good enough to include the world in a release. AI can always be improved later, as long as it doesn't self-defeat all the time with all the deck. (which was a real issue in the Tempest block where ~20 out of the 40 decks were unplayable for the AI, thus that world is still unfinished - AI has to be improved first.)
This is a list of issues I observed with the AI while testing the Invasion decks :
AI minor :
2. AI should prioritize returning Flametongue Kavu or Kavu Climer or other beneficial trigger creatures for gating creatures (or to any other bonuce effects)
3. Pernicious Deed should have that flag that prevents the AI from playing a second copy (same for Obvlivion Stone, Nevynnirral's Disk if they don't have it yet)
4. The AI seems to prefer playing normal lands first, "enters play tapped" lands later. That's usually the wrong play for the first few turns, where the AI wouldn't have a card to play from the mana anyway.
5. The AI seems to randomly send Captain Sisay to attack or search a card, this should be done smarter, always search if hand is not full and enemy isn't below 7 life?
6. Would be nice if the AI could recognize Rith's ability as one that produces potential blockers for the next turn.
9. When the AI plays a second copy of Teferi's Moat, it shouldn't pick the same color (and shouldn't play the second copy if the opponent plays only one color)
11. Not related to the invasion block but a generic issue : The AI prefers to kill the creatures I steal with Preacher instead of Preacher itself. I'm winning 99% of the games where I draw that card due to this, which likely messes up the difficulty testing results.
13. AI probably shouldn't play Skizzik without kicker as it sometimes does so then doesn't attack anyway or should learn to always attack with it if it wasn't kicked.
15. The AI had enough mana to play Urza's Rage with kicked and could have won the game by doing so, yet it played it without kicker and dealt only 3 damage. This happened more than once, it probably can't pay the kicker cost at all.
17. The AI uses Order // Chaos to kill creatures that are already having leathal damage on the stack from Powerstone Minefield
18. The AI sometimes doesn't seem to be able to use Tahngarth's ability, other times uses it during upkeep. It should never be used before the declare attackers step. (take advantage of vigilance)
19. Once the AI attacked with a 2/2 Quiron Dryan into my 2/3. It had Wax // Wane in hand but no mana to play it - all 3 lands were tapped.
21. Making Pulse of Llanowar, Elfhame Sanctuary and Fertile Ground AI playable would likely increase the AI playability of the block a lot (currently none of these cards are included in AI decks)
23. The AI used a counterspell to counter a spell targeting Glimmering Angel instead of activating the ability
25. The AI activated Crosis the Purger when I had zero cards in my hand.
26. Creatures with the "tap target creature" ability like Benalish Trapper and Stormscape Appreneic should not attack if the opponent has creatures that could be tapped next upkeep to prevent them from attacking.
(this seems to be a generic problem - small creatures with important noncombat abilties should be marked "stay home" and not attack unless the enemy is so low on health they can be defeated by 3 swings from the creature. This was one of the main problems
in Tempest block, for example Hermit Druid, Tradewind Rider and Ertai Wizard Adept. I really wanted to implement AI logic for this myself but...)
27. Yawgomoth's Ageanda should also be marked as a spell where the AI should never play a second copy (also it shouldn't be played without a large graveyard but I don't think the AI did that even once so maybe that's already implemented?)
33. The AI always picks "lose 1 life" on Putrid Warrior even if the opposite would be better.
36. AI played played a creature before Desolation Giant in the same turn.
39. AI doesn't seem to pay a kicker cost for Verdeloth the Ancient even if it has mana to do so.
40. AI uses Nightscape Master to bounce cards it could be killing with the other ability instead.
41. AI probably shouldn't use Recoil to bonuce own cards.
AI major :
1. AI should not play gating creatures if it would return another gating creature to hand, unless it's triggering kavu lair or sparkcaster ability.
7. The AI shouldn't sacrifice Fires of Yawimaya to deal 2 more damage when I still have 20 life. This should only be used to save creatures, or deal leathal damage to a blocker or player.
7b - it especially shouldn't do this is that results in the buffed creature no longer having haste and failing to attack!
8. The AI uses Rout as an instant during its own upkeep instead of waiting until main 1 and saving 2 mana (or possibly attacking first and doing it in main 2), this should never be used as an instant during the AI's own turn.
10. The AI seems to consistently fail paying both kicker costs on volvers even if it can afford doing it - it only pays one of them.
12. AI might be unable to pay the kicker for Dralnu's Pet - it had the mana and creature in hand but didn't do it. (and the creature was of a color the AI cannot play from hand)
14. Like Rout, Ghitu Fire also shouldn't be played as an instant during the AI's own upkeep. In fact, the AI generally should not use instant effects or cards during its own upkeep, it's very rarely the correct move, if ever.
22. The AI seems to be unable to play the card "Meddling Mage" even though it's marked with a "?" not an "AI" tag. I wouldn't expect it to play "fair" and pretend it has no idea what cards are in my deck (it guesses the color of my cards for other choices anyway) but in worst case it should at least play it as a 2/2 for 2 and name any random card it has no copies of in its deck.
24. The AI does not consider the board when picking colors for Voice of All. Even if I have two big white flying creatures in play, it still picks black because my deck has more black cards than white.
28. Once the AI had Temporal Spring and AEther Mutation in hand, and 4 life while I had 3 2/x creatures and it didn't play either bonuce spell. Maybe it was holding mana to cast Fact or Fiction later? (it had 5 lands, 3 forest 2 island)
(surprisingly it didn't even try to play Fact or Fiction before I won. Bug?)
29. The AI probably shouldn't play Phyrexian Arena when at low (below 10?) health.
30. AI didn't activate Pernicious Deed to destroy my 3/3 animated swamp that was killing them, even though it only takes an activation cost of 0. It also didn't activate it when I had a 4/4, they had a 4/4 and I had a 2/1 flying and they had 2 life.
After further games : In general the AI does not use Pernicious Deed very well (on top of playing 2-3 copies), and lets me hit them each turn with creatures and even dies, as long as I don't play too many at a time.
While I won't rate decks unplayable for just this, I'm pretty sure the decks including the card perform significantly worse than they could be and the card is borderline unplayable in the sense the AI obstinently refuses to activate it to save itself.
I suspect there is a threashold for how much advantage it expects from the card, and will not use it otherwise, even if not using it means losing the game. (even if I would be the only one losing permanents!)
31. The AI doesn't attack with units enchanted by Armadillo Cloak, and keeps them as blockers, even if it could gain several times the life it would lose to not blocking by the attack (AI had a 7/7 trample with it, I had two 2/x to attack and a 2/1 flying)
32. On the long term some sort of logic to decide whether to waste a kicker card immediately or wait for mana to use it with kicker would be nice, but it'll likely be hard to make one. (This is especially important for creatures, like Kavu Titan, volvers, etc)
34. When the AI has a 1/1 creature enchanter with Quicksilver Dagger, it still uses it to attack instead of tapping it to deal damage AND draw a card. (and this actually risks losing the creature to a "destroy target attacking..." spell)
35. The AI should play Warp Devotion before the bounce spells. As is, that deck works poorly.
37. AI had a Bloodfire Colossus in play, had active red mana, I had 3 life, it had 12, didn't activate to win, instead used Rout to destroy all creatures, and it didn't even sac it in response!
(I thought we improved the DamageAll logic? Is this card not using it?)
38. AI doesn't recognize Gerrard Capashen's ability to tap when attacking, so it fails to notice it could win the game as I only have 3 life and 1 blocker which could be tapped.
AI severe :
16. The AI seems to be unable to play Orim's Thunder at all, even if I have both a creature and an artifact and the creature is weak enough to be killed by the kicker effect. It can't even play it without kicker.
(I'll rank the deck with this card 1 higher to account for this bug getting fixed in the future - most of the AI's losses came from holding these and not using them to kill my creatures.)
20. The AI is unable to play Harrow and Allied Strategies even if it has no other cards to play and has mana. I unfortunately have to conclude the "Domain" deck is not AI playable as is - land searching and drawing are essential for it.
Note : I did see the AI play Harrow once in a blue moon, for example when it was about to die in combat, but it usually does not play the card.
28. AI used Dromar to return itself to hand by picking white (I had no white creatures). As all 5 dragon legends are actual opponents with their cards in the deck, and Dromar's doesn't have another creature, this is quite a big problem.
