Which Champions Are Best When You're Off Role?
10 cells • Python 3
Which Champions Are Best When You're Off Role?
Everyone has an emergency queue game:
- you did not get your main role
- you still want a real chance to win
- you need a champion that asks less from pure role comfort
This notebook looks for champions that overperform in exactly that spot.
Method
There is no direct "autofilled" flag in the notebook tables, so this notebook uses a practical proxy:
- take players from
summoner_leaderboard - use each player's
main_roleas their primary role identity - treat any recent ranked game in a different role as an off-role game
Then, for each target role, compare off-role champion performance, against that champion's normal role baseline across the latest three ranked patches.
The main ranking metric is bayes_off_role_win_rate:
- start with the champion's win rate in off-role games
- shrink it toward that champion's normal role win rate
- rank by the smoothed off-role win rate, not by pure edge
We still keep gap_vs_role_baseline_pp for context, but the main question here is simple:
*if you are off role, which champions actually win the most with a real sample behind them?
Highest smoothed off-role win rate in each role with a large sample
The first chart gives one answer per role, but the more useful output is still the short list behind it.
An off-role pick should usually do three things well:
- clear the floor on execution
- stay useful without perfect laning
- post enough sample that the result is believable
So the next table keeps the top five picks in each role, not just the single winner.
That is usually the right read for a three-patch notebook: fewer one-patch spikes, more stable emergency options.
Best large-sample off-role picks in BOTTOM across the latest three patches
Verdict
This notebook is not trying to find the highest-skill ceiling pick for an off-role game.
It is trying to find the best emergency pick:
- enough sample to trust
- high actual off-role win rate
- enough role forgiveness to survive lower familiarity
The practical way to use it is simple:
- take the winner table for a one-pick answer
- use the top-five table for a real emergency pool
- use
gap_vs_role_baseline_ppas a sanity check, not the main ranking metric