26 October 2026
Things the sampled battles say that a ranked table does not. Every one is the largest gap of its kind in the data that day — not an opinion, and not a list sorted by the obvious column.
SproutThe secret pick
53.7% adjusted win rate on just 0.6% of picks — one of the strongest records on the roster, from a brawler most drafts never consider.
From 46,262 sampled battles.
See the full record53.7%
adjusted win rate
GriffThe meta trap
Picked in 3.2% of sampled battles and returning 48.3% adjusted — below the average of the very players picking it.
From 245,676 sampled battles.
See why it underperforms3.2%
of all picks
Ruffs
ShellyThe giant killer
78.7% against Shelly specifically, +26.6 pts above Ruffs's own record. The most lopsided matchup in the sample.
From 4,806 sampled battles.
See every matchup+26.6 pts
vs its usual form
Shelly
CosmoThe secret duo
Teamed up, Shelly wins 72.6% — +38.5 pts better than it does otherwise. Neither is picked for the other.
From 7,592 sampled battles.
See who else pairs well+38.5 pts
when paired
JackyThe map surprise
55.2% here against 47.6% across every other map — the widest gap between a brawler and one piece of terrain.
From 2,725 sampled battles.
See the map+7.6 pts
on Triple Dribble
BonnieThe overnight rise
Meta score 4.5 to 5.6 in a single day, on enough battles for the move to be real rather than noise.
From 1,122 sampled battles.
See what changed+1.1
meta score, 1 day
How these are chosen. Each card is the argmax of a stated quantity over a stated population — the highest win rate among brawlers under 1% usage, the widest gap between a brawler and one map, and so on. The same data always produces the same set, so nothing here is written by a model or picked by hand. Win rates are adjusted against the sample average, so a week where the sampled players are stronger does not make every brawler look buffed.
Every day is kept at its own address — browse the archive or subscribe by RSS. These findings cannot be recomputed later, so the archive starts from the day it was built rather than reaching backwards.