What's Wratings?
Wratings is an evidence-based seeding tool and performance metric for esports. Our initial focus is Super Smash Bros. Ultimate and Melee, with additional games to follow.
A player’s Wrating is a public, rounded score derived from a conservative OpenSkill estimate. Wratings processes eligible recorded Set results from Start.gg. A Set must have a deterministic recorded result and fall within the selected calculation’s date, connection, and Event-entrant boundaries. Wratings does not add subjective event weights or ignore a result merely because it was played at a small eligible event.
The Engine: OpenSkill & Bayesian Inference
Wratings uses OpenSkill, an open-source implementation of the Plackett–Luce rating model. The model maintains two values for each competitor in a selected calculation:
- μ (Mu): the model’s current skill estimate.
- σ (Sigma): the model’s uncertainty about that estimate.
A player with fewer eligible recorded Sets usually has more uncertainty. As the selected calculation gains reliable connected evidence, the estimate can become more precise.
The public score begins with OpenSkill’s conservative ordinal, μ − 3σ. This discounts estimates with greater uncertainty. It is a conservative model score, not a literal claim that every displayed value has a separately calibrated 99.7% confidence guarantee.
Scaling and Rounding
Wratings then scales and rounds the conservative ordinal within each Pond. Before rounding, each Pond is scaled to a mean target of 1500. This makes the public score easy to read within that selected calculation. It does not make 1500 a universal player baseline or make scores from disconnected Ponds directly comparable.
Graph Theory: Ponds and Bridges
Wratings separates disconnected groups of players into Ponds rather than forcing them into one unified list. This avoids making comparisons where the recorded evidence contains no connection.
A Pond is a connected component in the Set graph. Players belong to the same Pond when eligible recorded Sets create a direct or indirect path between them.
A Bridge is any eligible recorded Set that connects two previously separate groups. It can arise through travel, online competition, or another valid cross-network encounter. Once connected, the selected calculation can use that evidence to estimate the groups together.
Pond 1 means the largest connected component in the selected calculation. It is a measure of connected-player count, not a skill tier or a claim that every player in it is stronger than every player in another Pond.
Time Windows and Available Evidence
Each Game, Connection, and Time Window combination is a separate OpenSkill calculation. Maximum Available uses the declared available-evidence start through the selected output’s actual calculation date; it is not an unbounded all-time claim. Standard buffered choices are offered only when the declared available data covers their full span.
The standard formula is nominal months × 30 + 14 days for 1, 3, and 6 months (44, 104, and 194 days), and nominal years × 365 + 181 days for supported full-year choices (546 days for 1 year, 911 days for 2 years, and so on). Only choices within the available evidence span are offered. The controls disclose actual inclusive dates and the calculation’s as-of date, which may be earlier than today. A buffer does not add unrecorded results or give extra rating weight to any event.
The selected window determines which eligible recorded Sets enter the mathematics. It is not merely a display filter: changing it produces a separate rating calculation, Pond structure, and matchup context. Maximum Available expands from the declared coverage start as new daily calculations are published; it is labeled All-Time only after coverage has been separately established. When maximum and a standard choice have exactly the same span, one choice carries both roles.
Why Wrating Values Shift
A Wrating can change after a refresh even when a player did not compete that day. That behavior follows from using the complete eligible evidence inside a rolling calculation window.
1. The Rolling Evidence Window
As time moves forward, older eligible Sets can leave the selected evidence window and newer eligible Sets can enter it. The evidence available to a calculation therefore changes over time.
2. Connected Results
The model evaluates the eligible Set network in the selected window. A new result can affect estimates for other players connected through that network, even if those players did not compete in the newest event.
3. Uncertainty and Activity
Wratings does not apply a separately described public inactivity penalty. A player’s score can still change when the selected rolling window contains less, different, or newly connected evidence. If no eligible current evidence remains for a player, that player may no longer appear in that calculation.
A Living Record of Competition
Wratings is a recalculated record of eligible competition, not a static list. Each qualifying recorded Set contributes evidence to the connected player network without subjective panel judgments or event-size weighting.