How we count

Clash Royale's official API has no endpoint for global deck statistics. It serves one player at a time, so a site like this has to build its own sample by reading many players' battle logs and computing everything itself.

We seed from Path of Legends rankings and from clan rosters, then follow opponents one hop. Clan rosters matter: matchmaking pairs players of similar skill, so following opponents from top players only ever reaches more top players. Without clan seeding the sample was 91% Path of Legends and had almost no ladder data at all.

This is a sample, not a census. Every table says so in its header: the population is sampled ranked players and their opponents, not everyone who plays.

Only 1v1 ladder and Path of Legends battles. Everything else is excluded because its win rates describe a different game:

• Friendly and challenge modes, where nothing is at stake
• Draft and pick modes, where you do not play your own deck
• Modified-rules events such as rage or triple elixir
• Clan war battles, which use different deck sizes
• 2v2, where the deck is only half the story

Roughly a quarter of everything the API returns is excluded on these grounds. We check the game mode rather than trusting the battle type, because some modified-rules events arrive labelled as ordinary matches.

A variant is an exact deck: eight cards, which of them are evolved, and the tower troop. A family is the same eight cards ignoring evolutions and tower troop.

Family pages carry the headline numbers because splitting every evolution combination apart would leave almost every deck below the reporting threshold. The variant table on each deck page shows what people actually run.

A win rate is wins divided by decided battles. Draws are excluded from both the top and the bottom of that fraction and reported separately. Counting a draw as half a win would quietly move every number on the site.

Each battle contributes one observation to each side's deck, so a battle between two decks counts once for each of them.

One player can contribute at most 20 battles per deck per day. Without that cap, a single person grinding one deck could move a published win rate on their own.

Every win rate is shown with a 95% Wilson score interval. Wilson is used rather than the textbook normal approximation because it stays inside 0-100% and behaves sensibly at small sample sizes, which describes most of this dataset.

The interval is asymmetric — the distance above the rate is not the distance below it — so it is always shown as a range and never as a plus-or-minus, which would be wrong exactly where the interval matters most.

A deck showing 63% over 130 battles has an interval of roughly 55-71%. The honest reading is “probably better than average, and we are not yet sure by how much”.

Tiers are computed from the lower bound of that interval, not the raw win rate. A deck at 60% over twelve games has a lower bound near 32% and does not tier highly; the same rate over 1,200 games does.

Matchup lists rank the same way, so a 3-0 record cannot be presented as a deck's best matchup ahead of a 550-450 one.

Below 100 decided battles a deck receives no tier at all and is listed under “still gathering data”, ordered by how often it is played rather than by how well it performs.

We measured deck concentration separately for each mode. In Path of Legends the 500 most played decks account for about 64% of play — a real, analysable meta. On ladder the 500 most played decks account for under 20%, and even at high trophies the top 100 decks cover around 12%.

A ladder tier list would therefore describe roughly a tenth of what ladder players actually do. We collect ladder battles and use them for card statistics and personal records, where the sparsity does not apply, but we do not publish ladder deck rankings.

A card pair's raw win rate says almost nothing. A popular staple appears in dozens of decks, so its pair rate regresses to the overall average: on the current patch the most played pairs all sit between 50% and 51.5%. Ranking by that would produce a list of the most common cards labelled as synergy.

So we report the difference between how a pair performs together and how its two cards perform on their own. Both figures are computed over the same population — decks that contain both cards and have cleared the publication floor. That matters: comparing against each card's record across all decks, including the long tail of rare ones, made almost every pair look positive, because established decks win more than rare ones. That is a property of establishment, not of the pair.

A pair qualifies only if it appears in at least three established decks. Below that, the pair's record is one deck's record attributed to two of its cards. We also rank only pairs whose interval separates them from their own expected rate, so the ordering is not driven by pairs that simply have not been played much.

This is observational. Cards are not assigned to decks at random, so a pair can look strong because it appears in strong decks, because both cards suit the same archetype, or because stronger players favour it. The comparison narrows that question; it does not settle it. Treat synergy as a hint worth testing, not a finding.

Observations are clustered. The same players recur, and matchmaking correlates which decks meet which. A confidence interval computed as though every battle were independent is narrower than the truth. The contribution cap reduces this; it does not remove it.

The sample is not the whole player base. It is built from ranked players, clan members, and the opponents they happen to face.

A win rate is not a recommendation. It reflects the players who chose the deck as much as the deck itself. A deck played mostly by strong players will look strong.

Card levels are not modelled. Two players running identical decks at different levels are counted the same way.

Battles are ingested every 15 minutes and pages are rebuilt on the same cadence. The timestamp on each table is the age of the data, not of the page render — a freshly rendered page built from an hour-old ingestion cycle says one hour, because that is how old the numbers are.