Bot ranking (all-time)

General principles — deliberately without fine-tuning details.

• Seasonal Elo: each map moves the rating — map result plus lane dominance. Published from the first map of the window. Scale centred on 1500.
• On 1 January, partial carry-over of the rating from one season to the next.
• Season (default) = Elo of the calendar year. Overall = average of season ratings, with a safety net for short careers. Current = Elo on ~6 months. Peak = best published season. 12 months / 2y / 5y = season average on that window, same safety net.
• This is not a match predictor — team Elo still wins there. A win rate mostly describes the club; the rating follows the player when he changes shirts.
• Only players with at least 40 recorded games are listed for overall.

What this ranking cannot tell you. Worth reading before quoting it.

• This is not a match predictor. Evaluating a player and predicting a game pull in opposite directions.
• A player seen mostly with the same team-mates is harder to separate from them — the model pulls him toward average.
• Pre-2014 games have lineups but little individual stats: domination leans on the years that have them.