Ranked-picktool

The ranking comes from the same aggregate the ranked tier list uses: archived ranked battles for Knockout · Flaring Phoenix over the last seven days, with a floor of 200 picks per brawler before it is scored at all.

Not on this selection, though. Only 4.289 ranked battles have been archived here for it, short of the 5.000 needed to measure a list, so every row below — win rates, pick rates and the sample size alike — comes from the deterministic generator rather than from play. Treat the ordering as illustrative.

Each brawler's raw win rate is first shrunk toward the global win rate of the pool by adding 1,500 pseudo-battles — an empirical-Bayes prior. Without it a brawler with 210 games and one lucky streak would outrank everything. A log-scaled popularity term is then added with a small weight, because a brawler nobody plays is rarely the safe first pick. The result is the Score column; tiers are standard-deviation bands around the mean score.

Ban value is derived, not measured. It is the same score plus the brawler's pick rate relative to the most-picked brawler left in the pool, weighted ×2.0. It answers “what would hurt most if the enemy took it”, which is not the same question as “what is strongest”.

What this does not do: There is no synergy or counter model here. Every suggestion ranks a brawler against the map alone — it does not know that your team already has two throwers, and it cannot tell you what beats the enemy's last pick. Removing picked and banned brawlers from the pool is the only way the draft affects the list.