← Back to About

cEDH Elo Methodology

Derived from the TopDeck cEDH Skill Rating framework by Charles Lien and Alex Lien.

Model Summary

We model each game as a 4-player zero-sum event. Players have one global rating and converted equity. Expected outcome is proportional to equity, and rating updates follow Elo-style gradient steps.

Player equity: 2^(R/200)

Expected share: Eᵢ = equityᵢ / Σ equity

Result share: Sᵢ = 1 (win), 0 (loss), 1/n (draw with n players)

Update rule: R' = R + K × (S - E), with K = 48 in current implementation.

Implementation Notes

Initial rating is 1500 for unseen players.

Games are processed chronologically across the full included dataset.

Draws are handled as fractional wins to preserve zero-sum accounting.

Non-4-player games are included and draw value scales to player count.

State leaderboards do not recompute a separate Elo. They rank players by global Elo after assigning each player to a primary state from recency-weighted local activity.

Reference

TopDeck document: https://topdeck.gg/elo/edh

We also preserved a local methodology copy at docs/methodology/cedh-skill-rating.md.