The matchup simulator is a decision-support estimate, not a full battle engine (docs/07). Every number comes from Monte Carlo games that use the engine's damage calcs, with a simple model of how each side plays. This page lists what the model leaves out or simplifies.

Team preview and plans

• Each side brings four and leads with two. With Auto bring-4, a quick pre-sim ranks every plan (90 for six Pokémon) in rounds that keep the best third: 1/5 of the pre-sim games per plan, then 2/5, then the full count for the best ninth. Each side ranks its plans against an even mix of the other's, so neither knows what the other will bring. Within a round every plan meets the same opponent plans with the same dice.
• Each side then plays its best ten plans by softmax over its final pre-sim win rates. With a fixed bring-4 you play that plan and the opponent brings the same as with Auto. The pre-sim is seeded from the two teams, so the plans don't change with the run's seed.
• There is no real team-preview reasoning, no knowledge of the opponent's sets beyond their most common ones, and no adaptation between games.

Decisions

• Both sides use docs/07's policy: every legal move × target gets log(usage prior + 0.05) + β × a tactical score, and the move is sampled by softmax. β is 1.5 (usage-weighted), 6 (greedy), or 1.5 with a Protect bonus (cautious). Your side is always usage-weighted.
• The usage prior is the move's share of the species' sets. A move outside the species' usual moves (its most used, as many as the set has) stands in for the most used of those the set leaves out, and takes that share when it's larger: a set built for a plan uses its plan move about as often as the move it replaced, not as rarely as the ladder does.
• The tactical score is the expected damage share with KO bonuses and a focus-fire bonus, plus heuristics for Fake Out, Protect, speed control, redirection, Helping Hand, status and setup moves, hazards (what they'll take from the Pokémon still to come in), Perish Song (the foes it will take, less your own), trapping (worth it only with a perish count to follow) and self-KO moves (less the user). Nobody predicts the opponent's Protect or switch, and partners coordinate only through the focus-fire bonus.
• The only switch made by choice is Perish Song's: a Pokémon whose perish count ends this turn switches out first when nothing holds it in and a Pokémon is left on the bench (with one left for two, the healthier goes). Otherwise Pokémon leave only when they faint (the replacement with the best matchup comes in, chosen by both sides before either is sent in), with U-turn, Volt Switch, Flip Turn and Parting Shot (the back Pokémon with the best matchup comes in, and Intimidate users avoid Defiant and Competitive foes), or when Dragon Tail or Circle Throw drags them out (a random back Pokémon comes in).
• A Pokémon Mega Evolves the first turn it can. Only one per side does.

Moves never chosen

• Encore, Disable, Taunt, Torment, Imprison, Substitute, Yawn, Octolock, Trick and Switcheroo, Transform, Metronome and similar volatile or disruptive moves.
• Side conditions other than Tailwind, Reflect, Light Screen, Aurora Veil, Wide Guard, Quick Guard and the four hazards, and hazard removal by status moves (Defog, Tidy Up, Court Change).
• Semi-invulnerable moves (Fly, Dig, Dive, Bounce, Phantom Force, Shadow Force), recharge moves (Hyper Beam, Giga Impact), delayed moves (Future Sight), OHKO moves, Counter, Mirror Coat, Metal Burst, Comeuppance and Beat Up.
• Roar and Whirlwind, and status moves that faint the user (Memento, Healing Wish).
• Two-turn moves other than Electro Shot, Meteor Beam, Solar Beam and Solar Blade. Those charge for a turn unless their weather or a Power Herb makes them instant.

Hazards, Perish Song and trapping

• Hazards work as on Showdown. Stealth Rock takes 1/8 of a Pokémon's max HP as it comes in, times Rock's effectiveness on it. Spikes take 1/8, 1/6 or 1/4 by layers. Toxic Spikes poison, or badly poison with two layers, and a grounded Poison type coming in takes them away. Sticky Web lowers Speed by 1, which Defiant and Competitive answer. All but Stealth Rock miss a Pokémon that isn't grounded (Flying, Levitate, Air Balloon). Heavy-Duty Boots avoid them all, and Magic Guard their damage. Stone Axe and Ceaseless Edge lay them as they hit, and Toxic Debris when a physical move hits its holder; Rapid Spin and Mortal Spin clear the user's side. A Pokémon that faints to hazards on its way in is replaced straight away.
• Not modelled: Magic Bounce sending them back, and anything else that grounds a Pokémon (Gravity, Iron Ball, Smack Down). Replacements are still chosen by matchup, whatever the hazards will take from them.
• Perish Song gives every Pokémon on the field that hears it (not Soundproof, but the singer always) a count: each faints at the end of the third turn after, unless it has switched out. It switches out only on its last turn, as above.
• Trapping: Shadow Tag (and Arena Trap and Magnet Pull); Mean Look, Block, Spirit Shackle and Jaw Lock while the user stays in; binding moves (Infestation, Whirlpool, Fire Spin, Sand Tomb, Bind, Wrap, Snap Trap: 4 or 5 turns at 1/8 of max HP a turn, 1/6 with Binding Band, 7 turns with Grip Claw); and Fairy Lock for the next turn. Ghost types and Shed Shell holders always get away. Since the only switch by choice is on a last perish count, trapping matters only then (a binding move's damage always applies). "Why you lose" names the Pokémon whose trap took yours.
• Self-KO attacks: Explosion, Self-Destruct and Misty Explosion take the user down as they're used, even into Protect, and Final Gambit when it hits (it deals the user's HP). When a side's last Pokémon explodes, that side loses, as on Showdown, where its faint comes first.

Effects not modelled

• Status: confusion, infatuation, Leech Seed, Salt Cure, Curse and Nightmare. Sleep, freeze and paralysis follow Champions (sleep 1–2 turns, freeze thaws 1/4 per turn and by the third, full paralysis 1/8). Burn, poison and toxic damage apply.
• Abilities: Magic Bounce, Trace, Download, Protean and Libero, Illusion, Imposter, Cursed Body, Effect Spore, Emergency Exit, Wimp Out, Pressure, Truant, Healer, Harvest, Cud Chew, Zero to Hero, Commander, Moody, Flash Fire's boost, Storm Drain and Lightning Rod redirection (their boost and immunity do apply), Anger Point, Mega Sol, Piercing Drill and Spicy Spray. Any ability the damage calc handles still applies to damage.
• Items: Eject Button, Eject Pack, Red Card, Mental Herb, Quick Claw, Custap Berry, Lum, Chesto and other status or pinch berries, Flame and Toxic Orb, Mirror Herb, Room Service, Blunder Policy, Booster Energy and the Metronome item's counter. Any item the damage calc handles still applies to damage.
• Field: Gravity, Wonder Room, Magic Room, primal weather, Cloud Nine and Air Lock.
• Moves: Ally Switch, Spotlight, After You, Quash, Instruct, Pollen Puff on an ally, multi-target Dragon Darts, and the HP Steel Beam costs its user.
• Speed: Quick Claw, Stall, Lagging Tail, Mycelium Might and priority-changing items.
• Struggle: a Pokémon with no PP left does nothing.

Game rules

• Doubles at level 50 under Champions rules: no Terastallization, stat points and Champions PP (Protect 8).
• A consecutive Protect-style move succeeds 1/3 as often as the one before (Showdown's Champions mod keeps the mainline rule). It fails if nothing moves after it.
• Speed order is recalculated after every action, as in Gen 8+. Speed ties are random.
• Every hit of a multi-hit move rolls its own damage and critical hit. A 2–5-hit move hits 2 or 3 times (35% each) or 4 or 5 times (15% each); Skill Link makes it 5, Loaded Dice 4 or 5.
• A game still going after 30 turns goes to the side with more Pokémon left, then more HP left. If both sides' last Pokémon faint together, the game is a draw and counts half.

Checked against real games

• Each week the simulator plays the Bo3 ladder games whose players showed their open team sheets, and its estimates are set against who won. The Method page shows the result under "Checking the simulator".
• sim@3 (29 Sep 2026, 265 games): the favourite won 56% of the games. On the 256 of them sim@2 also played, sim@3's estimates are closer to the results (Brier score 0.255 against 0.262) and its favourite won more often (55% against 54%).
• Since sim@2 neither side leans: sim@1 gave the side simulated as yours about 7 points too many, because only yours picked its four knowing what the opponent tends to bring.
• Real games are closer than simulated ones. The simulator knows the teams, not the players: its edges have held up only about half as strongly in real games. Under each estimate the simulator shows its real-game reading, the estimate pulled towards 50% by how strongly edges have held up so far. Read the estimate itself as a comparison between plans and teams, and the reading as the odds.

The gauntlet

• The gauntlet plays your team against each of the ladder's ten most used teams: the same six used by the most players in the week's ladder games, with their newest open team sheet's sets, or each Pokémon's usual set when no sheet was shown. Their players may have changed a set since.
• Each team gets one run of 1,000 games after the usual pre-sim, as the simulator is checked against real games, so each team's estimate is good to about ± 3 points. The summary counts each team by its players; its interval counts the dice only, not what this page lists.
• Ten teams are a sample of the ladder, not all of it: a team that isn't among the ten doesn't count, however often your team would meet it.