Everything on this site comes from the official Brawl Stars API — no estimates, no third-party data sets.
What follows tells you how far the collection reaches and where it stops.
14,334,169Ranked battles
30Maps
1,681,362Players read
0Players at Pro tier
86,005,014rows analysed
How the data is collected
For each player the crawler fetches their battle log — the last 25 battles or so.
Each of those battles names up to nine further player tags, and those go into the queue.
So the collection grows out of itself, starting from the leaderboards of the top players.
18,959,320Players known
17,277,958of those never read
3,967,500play ranked
1 : 11read : known
1Rosters in memory
1Histories in memory
That is the flip side of the snowball method: in the last pass 2,000 players read brought in 66,840 new tags, about 33 each.
The queue therefore grows faster than it can be emptied — catching up is impossible, and that is not the point either.
What matters is not how many players get read, but which ones.
So the selection takes players who play ranked at all first, and among those the most active:
right now the queue holds 4.24 million without a single ranked battle — and only 80,739 with eleven or more.
How far back the numbers go
Ranked battles per day, since the start of the season being analysed.
The early days are thin — the collection was only just starting up, and statements about them carry correspondingly little.
27.07.
28.07.
29.07.
30.07.
31.07.
01.08.
02.08.
03.08.
04.08.
05.08.
06.08.
07.08.
08.08.
09.08.
10.08.
11.08.
12.08.
13.08.
14.08.
15.08.
16.08.
17.08.
18.08.
19.08.
20.08.
21.08.
Last crawled: 8/21/26, 9:56 PM. Aggregates last computed: 8/21/26, 9:54 PM.
Is the draft prediction right?
The draft helper predicts an edge for a line-up. Whether that number holds up is testable —
against 400,000 real battles whose outcome is known. Each dot is a group of battles with a similar prediction;
horizontally what the helper said, vertically how often that side actually won.
worse draft50 %better draft
The direction is right — the dots rise from bottom left to top right, across the whole range, without outliers.
So the draft really does say something about the outcome.
The scale was wrong, and that is the real result: of one accumulated prediction point, only
22% actually comes through. That is not a bug but expected — the nine duels and six pairings
of a line-up partly explain the same thing and get counted several times when added up. The raw sum reaches past 130 points;
read as a win rate that would be nonsense. So the win chance in the draft helper is calibrated with this factor, not the raw sum.
What that means for the numbers in the draft has a page of its own.
What the test cannot do. The aggregates are built from the same battles that are being checked here — a clean
approach would use a time cut, but the duel and pairing tables carry no day column. The resulting error is small and boundable:
a duel cell contains several thousand pairings, a single battle contributes about a thousandth and cannot predict itself.
What remains is a slight self-confirmation — the measured scale is more likely too generous than too strict.
That is enough for “direction and order of magnitude are right”; not as proof of the last decimal place.
Calibration runs across all rank tiers together. Measured against 400 identical line-ups, the scale differs between
“all ranks”, Mythic and Legendary by less than 6% (34.5 / 35.1 / 36.4 points mean distance) — so the single factor holds
across the tiers. Calibrating per tier would be more precise, but costs a separate run for each one.
What is not in here
Deliberately left out
Friendly games never count — nobody plays them seriously.
Showdown is missing from the ranked analysis because ranked does not exist in that mode.
In ranked the current season counts; older battles sit under different balance conditions and would water the picture down.
The previous season stays stored all the same and stands in for individual rows while the new season is still too thin for them —
those rows are marked "prev. season". Trophies have no such cut: there a rolling window of 28 days counts,
because the meta shifts more slowly and the rotation has far too many maps to start from zero.
Not provided by the API
What someone played with — star power, gadget, gear, hypercharge — appears in no battle log entry.
Per player the API gives only tag, name and brawler. So there can be no analysis of those here, not even an estimated one.
The raw data is cleaned up after processing: individual battle participations after a day, battle IDs after two weeks.
The analysed numbers are untouched by this — they live in the aggregates, not in the raw data.
What can no longer be recomputed from them is a retroactively changed analysis across the whole period.