ABOUT
About this site
Methodology, data sources, and underlying assumptions behind the ratings on this site.
Rating Algorithm
GLICKO-2
We use the Glicko-2 rating system (Glickman 2012), an Elo successor that tracks each wrestler's strength as a rating value, rating deviation (RD, uncertainty), and volatility (σ).
For each tournament, we apply Glicko-2 to all bouts in chronological order. The rating stored against tournament X is the value after tournament X's bouts have been processed. Ratings are therefore strictly causal — no future information is used at any point.
Initial Ratings
INITIAL VALUES
Each wrestler starts at a rating determined by their debut division. Higher initial RD (350) lets the system rapidly correct without large rating leakage to opponents.
| Debut division | Initial rating |
|---|---|
| Makushita TsukedashiElite amateur entry | 1500 |
| Sandanme TsukedashiMid amateur entry | 1400 |
| Tsukedashi (general)Other amateur entry | 1300 |
| MakushitaRare direct entry | 1500 |
| Sandanme | 1400 |
| Jonidan | 1300 |
| Jonokuchi | 1200 |
| MaezumoNew recruit pre-list | 1100 |
Banzuke Forecast
BANZUKE FORECAST
Top divisions (Yokozuna, Ozeki, Sekiwake) follow traditional promotion / demotion rules; everyone else is forecast by looking up past wrestlers with similar rank and record, taking the median of where they ended up. Slot assignment then enforces the canonical banzuke shape (East/West, rank numbering, division headcounts).
Importantly: wrestlers with a winning record (kachi-koshi) are never demoted past Sandanme; wrestlers with a losing record (make-koshi) are never promoted past Sandanme.
Yusho Forecast
MONTE CARLO
We run 10,000 Monte Carlo simulations per tournament. Each simulation generates 15 (or 7) days of bouts using dynamic matchmaking that resembles real Sumo Association practice: early days pair wrestlers by rank, later days by current wins, last days tighten around the contenders.
Bout outcomes follow the Glicko-2 expected score formula based on the pre-tournament ratings.
Data
DATA
All bout records since 1985 (Showa 60). Wrestler profiles, banzuke, and tournament results from public sources.
Forecasts may differ from actual decisions by the Sumo Association's Banzuke Committee. Use this site for fun, not for betting.
FAQ
FAQ
What is Glicko-2?
Glicko-2 is a rating system developed by Prof. Mark Glickman, succeeding Elo. Each player's strength is represented by three values: rating, rating deviation (RD, uncertainty) and volatility (sigma). This site processes all official tournament bouts since 1985 in chronological order and stores the rating as of the end of each tournament.
How is the yusho (tournament championship) probability computed?
We run 10,000 Monte Carlo simulations per tournament. 15 days (7 for lower divisions) of bouts are generated by dynamic matchmaking — early days by rank, middle days by current wins, last days tighter around the contenders. Each bout outcome is decided probabilistically using the Glicko-2 expected score.
How is the banzuke (ranking sheet) forecast computed?
Top ranks (Yokozuna, Ozeki, etc.) follow traditional promotion/demotion rules. Others are forecast by looking up past wrestlers with similar rank and record and taking the median next-tournament rank. Final slot assignment respects per-division size limits and the rule that wrestlers with winning records cannot demote (above Sandanme), and vice versa.
How far back does the data go?
All official bouts since 1985 are covered. Wrestlers who retired before then have profile and banzuke history but no rating curve.