The Game Theory of Competitive Pokémon

By Aaron Trailer · VGC World Championship Top 8 · PhD Computer Science

"If you solved that, I would give you a PhD on the spot."
— A Professor of Computer Science, hearing about team building for the first time

Why Is Competitive Pokémon Hard?URL copied

If you've only ever played the single-player campaign, competitive Pokémon seems easy. Teach your level 100 Charizard Earthquake and win, right?

As it turns out, winning at multiplayer Pokémon - beating another human player, becoming world champion - is extraordinarily difficult. Answering why is also difficult, because competitive Pokémon is so complex it's hard to even put into words.

The strategy is like a tangled web: answer one question and many more follow. This document untangles it by identifying the fundamental building blocks that make competitive Pokémon what it is as a strategic game.

The Two PhasesURL copied

Competitive Pokémon consists of exactly two phases. They are deeply coupled, and you can't understand one without the other.

Team Building: Choose and fully customize six Pokémon - species, moves, held items, stats, and tera types. The art side.

Battling: Use those six Pokémon to reduce all of your opponent's to zero. The science side.

flowchart LR
    TB["🎨 Team Building<br/>(creative, iterative,<br/>open-ended)"]
    B["⚔️ Battling<br/>(tactical, time-pressured,<br/>opponent-driven)"]

    TB -->|"fields a team"| B
    B  -->|"win/loss reveals gaps"| TB
    B  -->|"metagame shifts"| TB

    style TB fill:#003a70,color:#ffcb05,stroke:#ffcb05,stroke-width:2px
    style B  fill:#cc0000,color:#fff,stroke:#ffcb05,stroke-width:2px

Strategic Element 1: Long-Term PlanningURL copied

Resources shrink as a battle progresses. Once a Pokémon faints, it almost never returns. (The move Revival Blessing exists but appears on essentially no competitive teams. For all practical purposes, fainted = gone.) This creates a hard asymmetry between the early game and the late game.

The early game is an investment, not just an exchange. Knocking out your opponent's Arcanine on turn two means your Venusaur has one fewer Fire-type threat to worry about for the rest of the game. That's not damage - that's positioning.

The Whimsicott GameURL copied

This is a real tournament game that demonstrates long-term planning in full. Whimsicott has terrible base stats - but its Prankster ability gives all non-attacking moves +1 priority.

sequenceDiagram
    participant A  as Aaron<br/>(Whimsicott + Miraidon)
    participant Op as Opponent<br/>(Flutter Mane + Chiu)

    Note over A,Op: Turn 1 — 4v4 Pokémon
    A  ->>  Op: Tailwind (priority via Prankster) ✓
    Op ->>  A:  Dazzling Gleam + Heatwave
    Note over A,Op: Both of Aaron's leads faint.<br/>Count: Aaron 2, Opponent 4.

    Note over A,Op: Turn 2 — Tailwind active!<br/>Aaron's team now doubles in speed.
    A  ->>  Op: Surging Strikes → KOs Flutter Mane
    A  ->>  Op: Ivy Cudgel → KOs Chiu
    Note over A,Op: Count: Aaron 2, Opponent 2.

    Note over A,Op: Turn 3
    A  ->>  Op: Surging Strikes → KOs Flutter Mane (sent back in)
    A  ->>  Op: Ivy Cudgel → KOs Zam mazenta
    Note over A,Op: Opponent has no Pokémon left. Aaron wins.

This is where Pokémon most resembles chess. In both games, resources decrease over time, and early sacrifice for positional advantage is a legitimate - sometimes optimal - strategy. As in chess, the late game is won or lost by decisions made in the opening.

Strategic Element 2: Simultaneous Action SelectionURL copied

In chess, players alternate. After your move, the board is a known state. Pokémon offers no such luxury.

Both players choose their moves at the same time. Then the game executes them in speed order. This one mechanic changes the entire character of the game.

The Decision MatrixURL copied

Annihilape (Ghost/Fighting) vs Gengar (Ghost/Poison). Ghost attacks are super effective against Ghost types. The faster attacker likely wins outright.

This feels like rock-paper-scissors. And in this isolated late-game example, it kind of is. But in a real double battle on turn three:

• Each player has 2 active Pokémon with 4 moves each
• Both can switch to 2 benched Pokémon
• Every combination produces a different probability distribution of outcomes

The actual decision grid isn't 2×2. It isn't even a grid - it's a wall of combinatorial explosions. No human can enumerate it fully. This is why skilled play is about avoiding the mind game entirely: navigating toward board states where you have a guaranteed win condition regardless of what the opponent does.

The Tailwind game above was one such state. With Tailwind active and a healthy attacking pair, there was literally nothing the opponent could do. Aaron had escaped the mind game through long-term planning.

Strategic Element 3: Imperfect InformationURL copied

In chess, both players see all pieces. In rock-paper-scissors, you know the three possible choices. Pokémon usually guarantees neither.

Assumption TrapsURL copied

You see Arcanine (Fire type). You think: Fire type → fire moves → my Water-type Gyarados is safe.

flowchart LR
    Obs["You see: Arcanine"]
    Obs --> Assume["Assume: only Fire moves"]
    Assume --> Dec["Send in Gyarados 'safely'"]
    Dec --> Reality{Wild Charge?}
    Reality -->|No — correct| Safe["Gyarados survives ✓"]
    Reality -->|Yes — surprise| KO["Gyarados KO'd ✗"]

    style Safe fill:#15803d,color:#fff
    style KO  fill:#cc0000,color:#fff

Wild Charge is an Electric-type move that Arcanine can legitimately learn. Your logic was sound. Your assumption was wrong.

The amount of unknown information decreases as the battle progresses and each player's Pokémon take actions, revealing their moves. But you still have to make decisions before you have full information - under time pressure, against an opponent doing the same thing.

Strategic Element 4: Probability ManagementURL copied

Even with perfect information and perfect reasoning, random outcomes remain. Variance is built into the engine.

Move Accuracy: Most moves are 100% accurate - but not all. Will-O-Wisp burns at 85%. Miss at the wrong moment and an entire strategy collapses.

Secondary Effects: Flamethrower deals damage and has a 10% chance to also burn the target. Free value; never guaranteed.

Critical Hits: Every attack has a $\frac{1}{24} \approx 4.17%$ chance to crit, dealing $ 1.5 \times$ base damage and ignoring your attack drops and their defense boosts.

Damage Rolls: The deepest source of variance. Every attack's final damage is multiplied by a random integer $r$ drawn uniformly from ${85, 86, \ldots, 100}$, producing up to 16 distinct damage values per move.

The Damage FormulaURL copied

$$ \text{Damage} = \left\lfloor \left\lfloor \left\lfloor \frac{2 \times \text{Level}}{5} + 2 \right\rfloor \times \frac{A}{D} \times \text{BasePower} \div 50 + 2 \right\rfloor \times \text{Modifiers} \times \frac{r}{100} \right\rfloor $$

Where $r \sim \text{Uniform}{85, \ldots, 100}$.

Simulated Damage DistributionURL copied

The turn_resolution.py script in this folder computes the exact distribution empirically. Core logic:

"""
Competitive Pokemon: damage roll simulation.

Demonstrates the 16-value damage range that makes
every KO threshold a probabilistic, not deterministic, fact.
"""

import random
from collections import Counter


def damage(base_power: int, attack: int, defense: int, level: int = 50) -> int:
    """Apply the core damage formula with a random roll in [85, 100]."""
    roll = random.randint(85, 100)
    raw = ((2 * level // 5 + 2) * base_power * attack // defense) // 50 + 2
    return int(raw * roll / 100)


def roll_distribution(base_power: int, attack: int, defense: int,
                      trials: int = 100_000) -> dict[int, float]:
    counts: Counter[int] = Counter(
        damage(base_power, attack, defense) for _ in range(trials)
    )
    total = sum(counts.values())
    return {v: round(c / total * 100, 2) for v, c in sorted(counts.items())}


def ko_probability(base_power: int, attack: int, defense: int,
                   target_hp: int, trials: int = 100_000) -> float:
    """P(one-hit KO) for given target_hp."""
    kos = sum(
        1 for _ in range(trials)
        if damage(base_power, attack, defense) >= target_hp
    )
    return round(kos / trials * 100, 2)

Accuracy Reference TableURL copied

The best players choose moves with the highest floor, not just the highest ceiling. Missing Dynamic Punch at a critical moment is not a freak event - it's 50% of the time.

The Strategic WebURL copied

The four elements don't operate independently. They compound.

mindmap
  root((Competitive<br/>Pokémon))
    Long-term Planning
      Resource shrinkage
      Type synergy chains
      Sacrifice for position
      Early vs late game
    Simultaneous Actions
      Decision matrix
      Opponent modeling
      Mind games
      Guaranteed win conditions
    Imperfect Information
      Hidden movesets
      Unknown held items
      Bench Pokémon
      Bluffs and traps
    Probability Management
      Move accuracy
      Secondary effects
      Critical hits
      Damage rolls
      KO thresholds

Interactive Strategy MapURL copied

How Pokémon ComparesURL copied

direction: right

elements: {
  ltp: { label: Long-term Planning }
  sim: { label: Simultaneous Actions }
  imp: { label: Imperfect Information }
  prob: { label: Probability Management }
}

games: {
  chess: {
    label: Chess
    icon: https://api.iconify.design/mdi:chess-king.svg
  }
  rps: {
    label: Rock Paper Scissors
    icon: https://api.iconify.design/mdi:hand-back-right.svg
  }
  poker: {
    label: Poker
    icon: https://api.iconify.design/mdi:cards-playing-spade-multiple.svg
  }
  mtg: {
    label: Magic the Gathering
    icon: https://api.iconify.design/mdi:cards.svg
  }
  pokemon: {
    label: Competitive Pokémon
    icon: https://api.iconify.design/mdi:pokeball.svg
  }
}

games.chess   -> elements.ltp
games.rps     -> elements.sim
games.poker   -> elements.ltp
games.poker   -> elements.imp
games.poker   -> elements.prob
games.mtg     -> elements.ltp
games.mtg     -> elements.imp
games.mtg     -> elements.prob
games.pokemon -> elements.ltp
games.pokemon -> elements.sim
games.pokemon -> elements.imp
games.pokemon -> elements.prob

Only competitive Pokémon connects to all four. This isn't a coincidence. The design of simultaneous move selection, combined with customizable teams, imperfect information, and deep probability variance, produces a game that cannot be fully solved by any one strategic framework.

Team BuildingURL copied

If battling is science, team building is art - and art has no solved strategy.

Speed Creep: One Number Wins TournamentsURL copied

Tournaments with prize pools in the thousands of dollars have been decided by training a Pokémon to be one speed stat point faster than the opponent's Pokémon.

Why? If a Pokémon is knocked out before it can act, its move never executes. Speed is not just "act sooner" - it is the gating condition on whether you get to play at all.

$$ \text{Final Speed} = \left\lfloor \text{BaseStat} \times \frac{\text{Nature}}{1} \times \frac{2 \times \text{IV} + \text{EV}/4 + \text{Base} \times 2 + 5}{100 + 5} \right\rfloor $$

A single EV allocation decision - 4 EVs is 1 stat point - can flip an entire matchup. Every speed tier in the metagame is a known threshold, and players deliberately park their Pokémon one point above the most common benchmarks.

The MetagameURL copied

Metagame
: The game about the game. What teams are other players running? What are the dominant strategies? What counters them?

The Iterative LoopURL copied

Team building and battling form a feedback loop that never fully closes.

stateDiagram-v2
    [*]          --> Hypothesis : Have an idea
    Hypothesis   --> Build      : Finalize 6 Pokémon
    Build        --> Ladder     : Test on ranked
    Ladder       --> Win        : Strong matchup
    Ladder       --> Loss       : Weak matchup
    Win          --> Analyze    : What worked and why?
    Loss         --> Analyze    : What failed and why?
    Analyze      --> Hypothesis : Revise theory
    Analyze      --> Build      : Adjust team
    Build        --> Tournament : Submit team
    Tournament   --> [*]        : Outcome recorded

What makes this worse: practice games are usually against known opponents. Tournament opponents make decisions in entirely different ways. And if you happen to freeze several opponents in a row with Blizzard in testing, your six Ice-type team looks like a genius idea - right up until the tournament, where you play against Fire and Rock teams and your Pokémon are stationary boulders.

ConclusionURL copied

Competitive Pokémon is unique because no other mainstream game combines all four strategic dimensions in the same package:

• Task
  Long-term planning across 8--200 turns
  highBattling
• Task
  Simultaneous decision-making without mind game losses
  highBattling
• Task
  Processing incomplete information about an opponent's team
  mediumBattling
• Task
  Managing probability: accepting variance, minimizing avoidable risk
  mediumBattling
• Task
  Build a winning team from $ 6.08 \times 10^{218}$ possible options
  criticalTeam Building
• Task
  Understand the metagame as a social, not just mechanical, construct
  mediumTeam Building

When somebody asks why is competitive Pokémon hard - this is what we mean. Not just "lots of Pokémon" or "it takes practice." The mechanics themselves produce a game where:

None of these games or any other, as far as we know, does all five quite the same way.

SourceURL copied