How the meta score is calculated

Every two hours the sampler walks the global leaderboards and reads the battle logs of the players it finds. A battle log holds a player's last ~25 battles and there is no history endpoint, so anything played between visits is lost permanently — the two-hour interval is set by that, not by cost.

The Ranked list counts competitive battles only. Ranked matchmaking pairs comparable opponents, so what is left reflects the brawler rather than who was holding it. The trophy list counts ladder battles, showdown included.

A raw win rate is not comparable across brawlers, because modes are not comparable. Finishing top four of ten in showdown caps the achievable rate near 40%; the same cohort wins closer to 78% in Brawl Ball. Ranking both on one number puts every showdown brawler at the bottom regardless of how good it is.

So each brawler is measured against the average for the mix of modes it is actually played in, and the difference is re-centred on 50%. An adjusted rate of 53% means “three points better than average, given where this brawler gets played”.

A 90% win rate over 50 battles is not better than an 86% rate over 1,300, but a naive ranking says it is. Every rate is therefore blended with the average, weighted by 50 pseudo-battles: a brawler needs roughly that many decided battles before its own record outweighs the prior, which is about where the noise stops dominating.

Below 20 decided battles a brawler is not rated at all. “We have not measured this” and “we measured this and it is bad” are different claims, and collapsing them is how tier lists end up implying the second about brawlers they have three battles for.

A win rate says the side holding a brawler won. It does not say the brawler is why. Strong players gravitate toward particular brawlers and carry their results with them, and step 2 corrects for where a brawler is played, not who is playing it.

So each brawler is also measured a second way: every player's win rate with it, minus that same player's win rate without it. Comparing a player against themselves removes skill exactly. The published figure is the average of the two estimates, because they are wrong about different things — the first carries the skill of whoever played it, the second compares a showdown specialist's showdown games against their own Brawl Ball games.

The two agree closely (r = +0.77) and disagree where it matters. Bibi ranked 96th on win rate alone and 16th once every player was compared against their own average, because she is played mostly by weaker accounts and their results were being read as hers.

Win rate alone cannot tell a genuinely strong staple from a niche pick that happens to win. A brawler played in 0.13% of battles at 52% and one played in 3.4% at 52% score the same, even though only the second is shaping the meta.

Pick rate is log-scaled, because the roster spans two orders of magnitude of usage. On a linear scale everything outside the top handful collapses into the same value.

Both measurements are mapped onto 0–1 between fixed anchors, so a score means the same thing from one week to the next rather than being rescaled by whoever happens to be present. The anchors sit just outside each list's own 5th–95th percentile:

Fixed is not the same as permanent, and that distinction has already cost these lists their calibration once. The anchors were set before the sampling rate rose eighteenfold in August 2026 and nothing re-derived them, so by late September 42% of the Ranked roster sat in D while 41% of the trophy roster sat in S or A. Nothing failed and nobody was told. They are now re-measured on every sampler run, and drifting back outside the scale is reported rather than absorbed.

The cut-offs are plain thresholds on the score:

• S7.5+
• A6.5+
• B5.5+
• C4+
• D3+
• E2+
• Fbelow 2