The data
- 283,732 battles from 2026-09-16 to 2026-09-28 (UTC). The last day is partial. That’s about 22,521 games a day.
- Every game has a winner, a rating and a full six-mon team preview on both sides. Unrated games and games where the same player sat on both sides are dropped.
- Megas are counted as their base species. Cosmetic formes (Vivillon patterns, Alcremie creams, Furfrou trims, Florges colours, Gourgeist sizes) are merged for display.
Why these two win rates
On-team WR asks “did the side with this species in its preview win?” It counts every game the species was on a team, brought or not. Across all species it averages exactly 50.0%.
Full-reveal WR asks “did the side that brought it win?”, but only in games where both sides revealed all four of their picks. It also averages exactly 50.0%.
Why not raw “brought” WR
You only learn a mon was brought when it hits the field. Losers reveal more of their four (3.79 per side vs 3.54 for winners), because fainted mons force the back line out, while a winner’s fourth often never appears. About half of all games end in a forfeit, which widens the gap. So raw “brought” WR is biased low. It averages 48.1% rather than 50%, and popular species lose 2 to 4 pp to it. Restricting to full-reveal games removes the bias.
Mirrors
If both sides have the species, one of them must win, so the game tells you nothing about it. Those games are left out of that species’ (or that pair’s) win rate.
Confidence intervals and sufficiency
- Brackets are 95% Wilson intervals.
- A number is shown as ranked only if n ≥ 500 and the interval is no wider than ±3 pp. That usually takes about 1,070 games. Anything thinner is greyed and marked thin, and sorts to the bottom.
- Species pages exist only for species whose on-team and full-reveal WR both clear the bar at all Elo.
- Colour marks whether the interval sits entirely above 50%, entirely below it, or straddles it. A straddling number is not evidence of anything.
The skill confound and the Elo split
Below 1300, players on stock meta teams beat players on off-meta teams, largely because they are better players. So an all-Elo win rate partly measures who uses a species, not only how good it is. Meta staples such as Rillaboom and Sneasler drop about 2 pp at ≥ 1300, while some support or gimmick picks rise 2.5 to 4 pp.
The ≥ 1300 view keeps only games where both players are rated 1300 or more (about 23% of games). It is the better guide to strength, but it has a quarter of the sample, so more of it is greyed out. Elo-adjusted win rates are a possible later addition.
Who is in the sample
- The collector only spectates rooms with a ladder rating of about 1100 or higher. Games below that are absent, so the all-Elo view means “≥ 1100”, not the whole ladder.
- Coverage of the ladder is unknown. These numbers describe rated M-C games the collector caught, not every game played.
- Games are only saved when they finish with a winner, so abandoned games can’t be measured.
- The higher-rated player wins only about 54% of the time. Ratings in a new regulation are young and noisy.
Player clustering
Wilson intervals assume every game is independent. They aren’t: one player can play hundreds of games with the same team. Popular pairs are spread across thousands of players, but a niche pair can be carried by a few dozen. When that happens the true uncertainty is wider than the interval shows.
So pairs are only listed when they also have at least 50 distinct players. At this snapshot every pair that clears the sample bar also clears the player bar. Species tables do not yet carry a player count. Player-clustered or bootstrap intervals would be the fuller fix.
What’s next: a rolling window
This v1 is one cumulative snapshot of the regulation’s first 13 days. A new regulation’s meta moves, so the plan is a rolling 30-day window. At current volume, that keeps roughly 215 species and 700+ full-reveal pairs above the bar at all Elo. The ≥ 1300 view needs about another month of data before most niche species and pairs fill in.
Counters
A counter cell asks: when one side has species A in its preview and the other side has species B, how often does A’s side win? It is an on-team win rate, so it counts every such game whether or not either species was brought. Games where both previews hold both A and B are left out, because both sides would count and the cell would sit at exactly 50%. A cell is shown only with n ≥ 500, an interval no wider than ±3 pp and at least 50 distinct players on A’s side.
A counter is a matchup between whole teams, not a one-on-one duel. A low cell may mean B’s teams are built to beat A’s archetype, or that the players who choose B are better. The skill confound applies here too.
Leads
A lead pair is the two species a side has on the field at turn 1. Share is the percentage of sides that opened with that pair. Lead win rate is the win rate of the side that opened with it, excluding games where both sides opened with the same pair. Lead pairs use the same bar as pairs: n ≥ 500, ±3 pp and 50+ players.
Leads are chosen after team preview. Each player picks a lead after seeing the opponent’s six, so a lead win rate describes the opening in the games where a player chose it. It is not how that pair would do if it led every game. A lead that only comes out into favourable teams will look better than it is. Lead win rate is also partly a team win rate: an opening belongs to a team, and the rest of the team shares the credit.
Seen on the ladder: revealed sets
Species pages list the moves, items and abilities each Pokémon showed during play, across 281,162 ladder games (2,041,086 brought Pokémon). The unit is an appearance: one species brought in one game. Each number is the share of its appearances in which it showed that move, item or ability at least once.
These are floors, not set frequencies. A move only shows when it is used. An item shows only when it triggers or is removed: a berry eaten, Leftovers healing, Life Orb recoil, a Focus Sash breaking, a Knock Off, a Mega Stone on Mega Evolution. A Choice item or an Assault Vest almost never shows. An ability shows only when it activates, so Intimidate reveals nearly every time and a passive ability almost never. Across all species, 53% of appearances revealed an item, 45% an ability, and the average appearance used 1.7 different moves. Shorter games, forfeits and early faints all hide more.
- Moves count for the Pokémon that chose them. Moves called by another move (Copycat, Metronome) or ability (Dancer, Magic Bounce) don’t count; Sleep Talk and Instruct repeat the user’s own moves, so they do. Nothing counts while a Pokémon is Transformed.
- Items and abilities count for their holder, not for whoever triggered them: Rocky Helmet and Rough Skin credit the defender, Frisk credits the frisker and the item’s owner separately. An item taken by Trick, Thief or Symbiosis, or an ability copied by Trace, Skill Swap or Mummy, is not credited to the new owner. Abilities shown after Mega Evolution are listed separately.
- Illusion is undone when it breaks: the whole stint is credited to Zoroark. A disguise that never breaks stays credited to the disguise, a small leak.
- Open team sheets are not used. The 2,621 games with one (0.9%) are left out entirely, since the opt-in games are a different sample. Champions has no Terastallization; no game in the corpus shows one.
- A hand check of 20 random games found every one of 711 attributions correct. A further 16 games picked for Illusion, Transform, Trick, Thief, Symbiosis, Entrainment, Mummy and Copycat turned up one missed ability reveal, now fixed.
- No win rates by item or ability. Whether one is revealed depends on how the game went (a Focus Sash only breaks when its holder is about to faint), so a win rate by revealed item would mostly measure the game, not the item.
The lead and bring calculator
The calculator runs in your browser on count tables built from every game that reached turn 1 (all Elo). It ships only counts: per species, per same-team pair and per cross-team pair, and only cells with at least 100 sides.
- Their bring. Each of their species starts at its bring rate given it is on the team. That rate is shifted, on the log-odds scale, by how it changes next to each of their teammates (×0.5) and against each of your species (×0.5). Their exact four is then the most likely set of four under those six chances.
- Their lead. Each of the 15 pairs starts from how often it led when both were on a team, shifted the same way by the other teammates (×0.25) and by your species (×0.5), then normalised to sum to 100%.
- Your lead. The pair’s own lead win rate, shifted by how each of its two species’ lead win rate changes against each of their six, averaged over the two leads (×0.8).
- Your four. The average of the six pairs’ full-reveal win rates (the unbiased “brought” rate), shifted by how each species does against each of their six, averaged over the four (×0.8).
- Shrinkage. Every rate is pulled toward the level above it (a pair toward its two species, a species toward the format) by 50 pseudo-sides for bring and lead rates and 200 pseudo-games for win rates. A cell seen 50 times barely moves anything; one seen 5,000 times mostly speaks for itself. Species with under 100 team sides count as average.
- The weights in brackets were picked by likelihood on Sep 23–24 games, using tables built only from earlier games.
Backtest
Tables were rebuilt from games through Sep 24 and tested on 74,567 games from Sep 25 onward.
- Their lead: the top guess was right 22.3% of the time and the actual lead was in the top three 47.3% of the time. Guessing their most common lead pair gets 12.9% and 32.2%. Uniform guessing gets 6.7%.
- Their four: 74.5% of revealed mons were in the predicted four (top four by bring rate alone: 70.4%). In games where they showed all four, the exact four was right 17.8% of the time (baseline 9.3%).
- Your lead: sides that led the recommended pair won 55.6% (CI 54.8–56.4, n 15,958); every other lead won 49.3% (CI 49.1–49.6, n 132,844). Leads ranked 9th to 15th won 46.6% (CI 46.2–47.0, n 56,790).
- Your four: sides that brought the recommended four won 54.7% (CI 53.8–55.7, n 10,391); other fours won 49.4% (CI 49.1–49.8, n 83,085).
- Estimated win rates are roughly calibrated: in every 2 pp bin with 1,000+ test sides, the observed win rate is within 3 pp of the estimate and usually 1 to 2 pp above it, so the estimates are if anything conservative.
The “followed the pick” numbers are observational. Players who pick what the ladder data favours are probably better players, so these show agreement with winning play, not that the pick causes wins. For scale, the mimikyu model for Regulation M-B reported a lead top-1 of 23.3% and bring recall of 66.1%, on different data, a different format and a different split.
Smaller caveats
- Player 1 wins 49.5% of games, a small side skew. Species stats count both sides, so it cancels.
- Illusion (Zoroark) appears in about 1% of games and can create a false “brought” for its disguise. The effect is negligible except for Zoroark-Hisui itself.
Site analytics
This site counts visits with Umami, self-hosted on our own server. It sets no cookies and stores no IP addresses. It records the page, referrer, browser, device type and country, plus a few interface events: calculator runs (how many species are on each side, not which ones), the column a table was sorted by, the species a table filter matched, and which outbound link was followed. Nothing is shared with third parties or used for advertising.