Card-Forge
AI doesn't cast Batwing Brume
Describe the bug AI doesn't cast Batwing Brume when an opponent's creatures are attacked
To Reproduce Steps to reproduce the behavior:
- AI has 2+ untapped lands
- AI has Batwing Brume
- Player attacks with 3+ creatures
- AI doesn't cast Batwing Brume
Expected behavior AI should cast Batwing Brume for white and black mana (as possible) when an opponent's 3+ creatures are attacked or if this damage will be lethal. There should be multiple payment options for AI: for BW, for 1W (without B), and for 1B (without W).
Additional context I think all cards that have mana payment in the text have similar problems (https://scryfall.com/search?as=grid&order=name&q=%28oracle%3Awas+oracle%3Aspent+oracle%3Ato+oracle%3Acast+oracle%3Athis%29+%28game%3Apaper%29).
Desktop OS: Windows 10 Java Version: 1.8.0_202-b08 Forge Version: 1.6.60-SNAPSHOT-02.29
Screenshots
I don't understand these reports.
This card (like many others) is clearly marked as AI:RemoveDeck:All
You should be warned when you force that on it in a deck
@tool4ever, and I don't understand your outrage. Since the card is marked as AI:RemoveDeck:All, it can't be coded?
I'm not really outraged... If you don't appreciate me telling you you're wasting your time since us developers (well really anybody) can easily search for the affected cards then that's fine by me. Continue doing this for your pet cards by all means 🤦♂️
@tool4ever, I'm not marking cards that should in principle be difficult for AI (Prismatic Ending, Rakdos Charm, Shadow of Doubt, and so on). I'm marking the ones that intuitively should be understandable by AI, but somehow aren't. For example, AI makes great use of Springleaf Drum, but for some reason it only puts it on the battlefield with Urza's Saga. I get the feeling they've just been forgotten about.
They're not my "pet cards". Most of these cards are played in tier or popular decks, and often AI plays these decks poorly because of weird moments like this, which I don't think are hard to fix.
And I try to do what I can for the project. Now, for example, AI can fully play Aether Vial and Karn, the Great Creator, although there were some basic flaws before my requests. Now AI plays great with human decks and merfolk decks.
Continue doing this for your pet cards by all means
🤦♂️
Oh, that's so constructive and inviting to communicate (no).
But according to your reply, I'm really wasting my time here volunteering and making requests that even the description of the card at the stack stopped being corrected....
In theory everything can be coded but unique stuff might not be worth the effort (what is "easy" for the AI depends on a few factors).
But I don't think you're understanding my point:
Why are you ignoring this warning?
I know you're trying to help out but we don't need your reports for 2500+ of these cards...
While it is possible that some tagged cards have become playable, that would require additional/different research.
Why are you ignoring this warning?
I'm not ignoring the warning, I'm considering it. But I can see that AI easily using Springleaf Drum, and the whole problem is that it doesn't cast it from hand.
Batwing Brume doesn't feel like a broken card either. Can't it be possible to make the card have 3 payment modes, or at least one, so that AI uses it like Fog?
I know you're trying to help out but we don't need your reports for 2500+ of these cards...
@tool4ever, even if I close all ~13 requests with cards from the warnings (which is no guarantee at all that they will be noticed later), there are a ~73 requests left, ranging from the simplest #4372 to #4152 and #4266, all with different problems.
- For Springleaf Drum it is very likely the AI will cheat with it in some scenarios since mana abilities with these additional tap costs can make it miscalculate
- Batwing Brume is a good example of a card that might look simple, but is actually implemented by 3 separate effects in Forge (additional RepeatEach because of the tricky variable life loss). Those would all need to be synchronized to reach a valid play decision but to make it even more difficult you have shard conditions for each half - so even more logic is needed to make the AI "lookahead" based on what colors it has for paying. Considering there are plenty of alternative Fog-like cards to put in AI decks the decision to not spend extra time on this one feels like a no-brainer
@tool4ever, thank you for the detailed explanation! But what should I do with these requests? Should I close them as unrealizable? And the other ones too (like #4391)?
And what should I do with those ~73 requests that are not included in the warnings?
- scripts the AI can't play should probably be tagged after checking they're just not missing AiLogic params. For the last recent sets that part of "QA" hasn't really been enforced too strictly because there's just no way for our current manpower to keep up with the amount of cards & gimmick mechanics Wizards keeps throwing around
- already tagged scripts should (e. g. we can't prevent human from Donate a bad one) not cause bugs. Ideally a developer extending an effect would always address any relevant AI impact. But because of what I said above the occasional iteration over them could still be a justified activity (though without randomly picking you'd need some background knowledge to understand recent changes in API or script refactors)
This issue has not been updated in a while and has now been marked as stale. Stale messages will be auto closed.
![github-actions[bot] avatar](https://avatars.githubusercontent.com/in/15368?v=4 "Published by a pub.dev verified publisher"){kind=small}