In-Game Decision Making

01At a glance

In-game decisions reduce to a small set of repeatable questions. Knowing the questions — and answering them in order — is the entire game. The questions don't change between turns; only the inputs do.

02The per-turn decision loop

Every turn, run the same loop: assess board state, predict opponent, evaluate moves, commit to action. 10-15 seconds per turn. Done consistently across 50 turns, this loop is what wins games.

The 4-step turn loop

1. Assess (3 sec): board state — your HP, their HP, hazards, weather, terrain, statuses, current win condition.
2. Predict (3 sec): opponent's 2 most likely moves. Use last 2 turns + your scouting + their threat profile.
3. Evaluate (5 sec): your 2-3 candidate moves. For each, expected value against their predicted moves.
4. Commit (2 sec): pick the highest-EV move. Click. Don't overthink — better to commit on Q3 evaluation than on perfect math.

When the loop breaks

• Time pressure: tournament timer running out. Skip Q5; trust your matrix.
• Tilt: a missed crit just cost you the game. Q3 evaluation suffers — bias toward safer plays until composed.
• Information overload: surprise set, surprise spread, surprise Tera. Reset to default plan; don't panic-predict.

03Risk vs reward — the core trade-off

Every move is either preserving your win condition or pushing for the kill. These two modes have opposite risk profiles. Knowing which mode you're in is the decision's prerequisite.

The 70/30 rule

Default to safe play 70% of the time. Aggressive 30% of the time. Reverse those numbers and you'll lose to better predictors. This is the consensus heuristic among 1900+ ELO players.

04Prediction trees

Predicting the opponent's next move is the hardest skill in competitive Pokémon. The technique: enumerate their 2-3 most likely options, pick your move based on expected value across that distribution.

Building a prediction tree

1. List opponent's 2-3 most likely moves: based on their threat profile, current HP, item, last turn's behavior. Cap at 3 — going deeper is wasted effort.
2. Assign rough probabilities: 60% / 30% / 10% is a typical split. Don't obsess over exact numbers — gut feel is fine.
3. For each of YOUR candidate moves, compute expected value across the 3 branches: win the branch × probability + tie × probability + lose × probability.
4. Pick the highest expected value move. If two are tied, pick the one with the lowest worst-case (loss aversion is rational here).

Worked example: Iron Bundle vs your Kingambit

Their Iron Bundle (Choice Scarf, max HP) is at 50% HP. Your Kingambit at 100% HP. You can:

• A — Sucker Punch: hits if Bundle attacks; misses if Bundle switches. Sucker on Scarf-locked Bundle is high-value if attacking, 0 if switching.
• B — Iron Head (no priority): KOs if Bundle stays in. If Bundle switches, hits the switch-in (often 0 damage if Steel-resist).
• C — Switch out to a Bundle-counter (Iron Hands): saves Kingambit, brings in something that walls Bundle. Good if Bundle stays in.

Prediction: Bundle is Scarf-locked into a move. If they Hydro Pump (its strongest move), it does ~85% to Kingambit. They likely commit. Probability: 70% Bundle stays in + Hydro Pump, 20% switches to Hydro Pump teammate, 10% other.

• A (Sucker Punch): KOs Bundle 70% of the time, 0 if switch (30%). Expected value: positive.
• B (Iron Head): KOs Bundle if stay (70%), 0 if switch (30%). Same EV as A but loses the priority.
• C (Switch): 0 damage but keeps Kingambit alive. Useful if you don't need Kingambit's damage now.

Pick A. Sucker Punch's priority means even if Bundle is faster, Sucker hits first, then if they switched, you took 0 damage but they lost tempo. Highest expected value.

05Win-condition tracking

Your win condition is the Pokémon that wins the game when you reach late game. Every turn, every decision, ask: does this preserve or break my win condition? If preserve, play safe. If break, you may need to accept a bad position to save it.

Identifying your win condition

• Setup sweepers: Volcarona after Quiver Dance, Garchomp after Dragon Dance, Iron Valiant after Swords Dance.
• Late-game cleaners: Kingambit with Sucker Punch, Dragonite at +1 from Multiscale.
• Strong wallbreakers: Iron Hands Belly Drum, Roaring Moon at +2 from Dragon Dance.
• Pivot win conditions: Slowking-Galar + Future Sight + status spam — slow grinding win.

Win-condition states

Multi-win-condition teams

Strong teams have 2 win conditions — main and backup. Your Iron Valiant Swords Dance is the main; your KingambitSucker Punch cleanup is the backup. If main goes down, you fall to backup. Tracking both means you don't panic when main dies.

06The 50/50 trap

A 50/50 is a position where two of your moves win 50% of the time, lose 50% of the time, and you can't reliably outpredict. Strong players AVOID 50/50s by building positions that have a dominant choice. Weak players take them, hoping to outpredict.

Why 50/50s are bad

• Variance dominates: half the time, you lose a critical position. Across many 50/50s, half your "perfect plays" lose anyway.
• Mental fatigue: 50/50s are stressful. Across a Bo3, 3-4 50/50s drain mental energy faster than 30 simple decisions.
• Compound effect: losing 1 of 2 50/50s in a game = playing from behind. Losing 2 of 2 = match over.

Avoiding 50/50s

1. Build positions where one choice dominates: switch in the right Pokémon ahead of the matchup. Set up at the right moment so the opponent has no good response.
2. Force the 50/50 ON THEM: maneuver into a position where the OPPONENT has the 50/50, not you. Your safe play wins regardless of their choice.
3. Pick Pokémon / movesets that resolve 50/50s: Sucker Punch eliminates the "they switch / they attack" 50/50. Wide Guard eliminates the "spread vs targeted" 50/50.
4. Accept neutral outcomes: a tied turn isn't a loss. If the only options are 50/50 or neutral, neutral is correct.

07Positional play vs aggressive play

Two playstyles: positional (manipulate the board to force good matchups) and aggressive (push for damage at every opportunity). Both win at high level; the question is which fits your team and your read on the opponent.

Style adaptation

Top players adapt their style to opponent. Against a positional opponent, ramp aggression to break their patience. Against an aggressive opponent, slow down and let them overcommit.

08Calc discipline — never guess damage

The single most common decision-quality leak: estimating damage instead of calculating it. "That should KO" → it doesn't. "That should survive" → it doesn't. The damage calculator is a 5-second check; ignoring it is an unforced error.

When to calc

• Pre-game: calc your team vs every meta threat. Know all the 1HKO / 2HKO thresholds.
• Pre-match: re-calc against the specific opponent's likely sets if scouting revealed unusual spreads.
• Mid-game: when in doubt about a kill or a survival roll, calc. Even a 5-second check changes the decision.
• Post-game: calc unexpected outcomes — if a hit didn't KO when expected, your spread or assumption was wrong.

Internalizing damage rolls

Top players know damage rolls by heart for the 5-10 most common matchups in their format. Heatran Magma Storm vs Iron Hands = ~38-45%. Iron Valiant +1 Close Combat vs Skeledirge = ~75-90% (likely OHKO with hazards). Knowing these by memory speeds up decisions and frees mental cycles for prediction.

09Tilt control — recovering from a misplay

You'll misplay. Top players misplay too. The skill is recovering from the misplay without spiraling. Tilt = emotional state where decision quality drops because the previous turn went badly.

Recognizing tilt

• You're replaying the previous turn in your head while making the current decision: signal of regret-driven play.
• You're biased toward aggressive plays to "make up" for the misplay: revenge desire is the tilt indicator.
• You're skipping prediction tree work because "it doesn't matter, I'm losing anyway": emotional shortcut.

Recovery protocol

1. Pause: take 10 seconds. Don't click yet.
2. Reset: re-read your written mental model. Re-confirm win condition.
3. Refocus on this turn: previous turn is sunk cost. Current turn is what matters.
4. Default to safe: when tilted, bias toward the safe play even if you think aggressive is better. Bias correction.
5. Breathe: 4-second inhale, 4-second exhale. Resets autonomic state.

10Common mistakes

• Estimating damage instead of calculating — "that should KO" without checking. Calc takes 5 seconds; misplay costs 4 turns.
• Going aggressive by default — most players bias toward aggressive. The safe play is harder to pick because it feels passive — but it's correct more often.
• Predicting too deep — projecting 4 turns out is wasted. Opponent will react; your prediction must update each turn. 1-2 turn depth is correct.
• Ignoring opponent's tendencies — you scouted them. Use the scouting. If they Tera turn 2 every replay, expect Tera turn 2 in your match.
• Taking 50/50s — when you have a safe alternative, take it. 50/50 = coin flip; the coin doesn't care about your skill.
• Tilting after a misplay — the previous turn is sunk cost. Don't let it influence the next 5 turns of decisions.
• Tunneling on a single threat — focusing on opponent's wincon means missing positional opportunities elsewhere. Stay aware of full board state.
• Failing to update mental model — your scouting said X, but in-game they did Y. Update the model; don't insist on the prep.

11Where to go from here

In-game decisions execute the prep done in match preparation. The Bo3 page extends this to multi-game adaptation. Both build on the foundations covered in mechanics + teambuilding.

• Match preparation — Match Preparation Workflow covers what feeds into in-game decisions.
• Bo3 strategy — Bo3 Tournament Strategy covers multi-game decision adaptation.
• Reading the meta — Reading the Meta covers format-level threat knowledge.
• Damage formula — Damage Formula covers the math the calc computes.
• Speed mechanics — Speed Mechanics covers priority brackets and Speed-modifying decisions.
• Live tools — Damage Calculator, Replay History, Gen 9 OU live data.