feat(ai): AI Rework v1.0: Enemy Command Selection

[innerthunder]

What are the changes the user will see?

The Enemy AI's command selection algorithm has been completely rewritten, improving its decision-making in the following ways:

• The AI now uses estimated damage calculations to inform its decision to switch out or use a move. It is now more likely to switch out if it can safely avoid a KO by doing so, and it's much less likely to switch out otherwise.
• The AI generally weighs the effects of Status moves and the secondary effects of Attack moves more heavily when choosing a move. It also penalizes attacks with drawbacks and steep conditions (e.g. Focus Punch) more harshly. However, it should still prioritize moves that are forecasted to KO the player's Pokemon at all times.
• Because variance is applied when scoring the move instead of during the final decision-making process, the AI should be much less likely to randomly choose an obviously useless or detrimental move action.
• The AI's move targeting in double battles is generally improved. Several instances where the AI would erroneously target its ally with an attack or Status move in double battles have been fixed.

Why am I making these changes?

The existing Enemy AI is – to put it bluntly – terrible. There are several glaring issues with it that detract from the game's battle system:

1. The AI values high-power attacks way too highly in most situations. This causes the AI to almost never use Status moves in the late game and, in some cases, spam moves with exploitable side effects or conditions such as Focus Punch and Giga Impact.
2. The AI for Wild Pokemon doesn't eliminate useless or detrimental moves before it makes a weighted random move selection. As a result, if 1/4 of the Pokemon's moves are useless, it has a ~6% chance of selecting that move anyway.
3. The enemy's HP ratio is too much of a factor in evaluating matchups, meaning that Trainer Pokemon can easily be coerced into switching out repeatedly while they're at low HP.
4. When evaluating matchups, Pokemon are evaluated based on their base types and not the moves that are known to the AI. Enemy Pokemon may decide to switch even if they have a super-effective move against their opponent.
5. Double battles are mostly an afterthought. The AI often attacks the wrong target, and sometimes even targets their own ally with negative effects.
6. Several move effects don't have benefit scores implemented and thus cannot be evaluated by the AI. The scores that are implemented are often not implemented with targeting in mind, not scaled properly with respect to attacks, and/or not documented at all.
7. The AI completely ignores ongoing field effects and other battle effects in its decision making. This is most notable with hazards like Stealth Rock, which are currently absurdly strong in key battles because of how frequently the Trainer can be baited into switching.

You get the idea. With the AI only having two real modes in its decision making – switching and attacking – there is a lot of wasted potential in the battle system.

What are the changes from a developer perspective?

All aspects of the Enemy AI's command selection have been rewritten, including:

• the entire EnemyCommandPhase
• related Pokemon and EnemyPokemon methods, including getNextMove and getMatchupScore
• Move's score calculation methods
• Effect Score functions for all Move Effect Attributes and other relevant Move Attributes
• Condition Score functions for Move Conditions (which have also received major structural changes)

This PR also features a new suite of tests performing statistical analysis on the AI's command selection in different scenarios. These tests were made with new utils and matchers targeting EnemyPokemon and their command selection methods.

For more details on the algorithms involved, see the Poketernity Enemy AI Proposal.

For details on newly implemented Move Effect Scores, see the Effect Scores Sheet.

This PR has been developed incrementally with several Addendum PRs, all of which are listed below.

Addendum PRs:

• [AI-Rework] Add starter code for move scoring functions #356
• [AI-Rework] Rewrite enemy command selection #403
• [AI-Rework] Add move-property-based scoring #442
• [AI-Rework] Add Move Attribute scoring parameters under MoveAttrOptions #463
• [AI-Rework] Add effect scores for ability-changing effects #834
• [Addendum][AI-Rework][Test] Add matchers + Improve test utils #1125
• [AI-Rework] Add effect scores for arena-tag-based moves #1158
• [AI-Rework] Rewrite MoveCondition + update Condition Score logic #1182
• [AI-Rework] Add effect scores for moves with BattlerTag effects #1219
• [AI-Rework] Remove appliesScoreOnKO option #1222
• [AI-Rework] Add caching for invariant scoring calculations #1258
• [AI-Rework] Re-implement Instant Tera commands #1283
• [AI-Rework] Add Effect Scores for Arena/Battler-tag-modifying effects #1303
• ai-rework: Add Effect Scores for Status Effect inflicting moves #1312
• ai-rework: Add Effect Scores for misc. status-effect-related attributes #1324
• ai-rework: Add Effect Scores for healing effects #1330
• ai-rework: Rewrite Ally Target scoring logic #1331
• ai-rework: Add scoring for StatStageChangeAttr + subclasses #1337
• ai-rework: Add Effect Scores for misc. stat-modifying attributes #1340
• ai-rework: Move ability- and move-revealing test utils to FieldHelper #1341
• ai-rework: Rework multi-target move scoring logic #1344
• ai-rework: Add Effect Scores for self-damaging move effects #1346
• ai-rework: Add Effect Scores for Terrain- and Weather-changing effects #1351
• ai-rework: Update Effect Score calculation for charge moves #1358
• ai-rework: Add Effect Scores for type-modifying moves #1362
• ai-rework: Add Matchup Score modifiers #1365

Screenshots/Videos

How to test the changes?

pnpm test ai/

Checklist

• ⚠️ If this is a PR for main (such as a hotfix), has the game version been updated (pnpm update-version:patch / pnpm update-version:minor?
• Otherwise: I'm using beta as my base branch
• There is no overlap with another PR?
• The PR is self-contained and cannot be split into smaller PRs?
• Have I provided a clear explanation of the changes?
• Have I tested the changes manually?

• Are all unit tests still passing? (pnpm test:silent)
  • Have I created new automated tests (pnpm test:create) or updated existing tests related to the PR's changes?
• Have I provided screenshots/videos of the changes (if applicable)?
  • Have I made sure that any UI change works for both UI themes (dark and light)?