Read the explanation
Battle AI system

How PokeSynergy chooses a Battle AI action

PokeSynergy starts only from what the acting side can know, scores every legal joint action against plausible opponent actions, spends extra simulation only on tactically distinct close decisions, then returns a deterministic choice with an auditable score breakdown.

Read the explanationReviewed · PNG download · Updated 21 October 2026
How PokeSynergy chooses a Battle AI action: stages and decision branches, explained in text below.
Revealed information in · every legal choice compared · one explained action out · Read the equivalent explanation below, or open the PNG to inspect every label.

Scope and inputs

Intended reader: Competitive Pokémon players who understand turns, moves, switching, and hidden information but not search algorithms.

Revealed information in · every legal choice compared · one explained action out.

Follow the stages below. Branches name their destination; a loop returns to the named stage.

Stages and branches

  1. Use Only Revealed Information

    Opponent sets stay hidden until revealed.

  2. Read the Tactical Board

  3. Model Five Plausible Replies

    Usage-supported, never hidden sets.

  4. Score Every Legal Choice

    Compare value, cost, and risk.

  5. Reject Before Scoring

  6. Keep Distinct Contenders

    Retain up to 12 tactical plans.

  7. Is Deeper Search Worth It?

  8. Combine and Rank Scores

    Stable choice ID breaks ties.

  9. Return Choice and Reasons

Outputs

  • Return Choice and Reasons

Limits and uncertainty

  • Illegal = rejected · Ineffective = legal, lower score · Deepened ≠ selected · Rank one wins
  • Rejected by the request/protocol boundary or absent from the legal set; never a score band.
  • Still legal, but exposure, weak progress, repetition, or setup risk lowers its score.
  • Evaluated but not deterministic rank one; shallow alternatives may become ineligible after deep search.
  • Received bounded continuation; this does not guarantee selection or perfect play.

What this does not prove

A bounded evaluation is not perfect-play proof or a guarantee of results outside its declared protocol.

Glossary

joint action
both active Pokémon's choices taken together
hypothesis
a weighted plausible opponent action, not hidden knowledge
policy selection probability
the deterministic policy's rank-one selection, not win probability

Sources and review

Reviewed for publication . Source observations retain their original dates.

  • Battle frame · observed 2026-08-27
  • Known legal actions · observed 2026-08-27
  • Turn scoring · observed 2026-08-27
  • Policy profile + diagnostics · observed 2026-08-27
Source references and revision

Source SHA-256: 2a8f960d037b7add4a076af8fce68d28a05c5542b5b79bb0fd9c50426bf9c5f5

Originating chart: POK-98

  • Battle frame: src/server/battle/championsBattleProtocol.ts; src/domain/battle/protocol.ts
  • Known legal actions: src/server/battle/policy/knowledge.ts; src/server/battle/policy/legalCandidates.ts
  • Turn scoring: src/server/battle/policy/turnPolicy.ts; src/server/battle/policy/scoring.ts; src/server/battle/policy/boardEvaluation.ts
  • Policy profile + diagnostics: src/server/battle/championsBattlePolicyProfile.ts; src/domain/battle/diagnostics.ts