What's Wratings?

Wratings is an effort to provide 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 estimates the probability that the player will win a match against a given opponent, based entirely on observed data of tournaments of 8 or more entrants. We do not curate results or ignore matches; if a set was played, it is mathematically accounted for. By utilizing modern Bayesian inference and graph theory, we calculate the true mathematical probability of match outcomes, rather than relying on subjective panels or win/loss ratios.

The Engine: OpenSkill & Bayesian Inference

At the core of Wratings is OpenSkill, an open-source implementation of the Plackett-Luce model (conceptually similar to Microsoft's TrueSkill). Traditional ranking systems assign a single number to a player. OpenSkill, however, tracks two distinct variables for every competitor:

• μ (Mu): The system's estimation of the player's actual skill.
• σ (Sigma): The system's level of uncertainty regarding that skill.

When a new player enters the system, their uncertainty (σ) is extremely high. As they play more matches, the system gathers evidence, and σ decreases.

Because Wratings is designed as a professional seeding tool, it must be conservative. We calculate the final leaderboard wrating using the formula μ - 3σ. This represents a 99.7% confidence interval. The system assumes a player is at the absolute lower bound of their estimated skill until they consistently prove otherwise through volume and performance.

Scaling and Rounding

To make Wratings intuitive and easy to read, we scale then round the raw OpenSkill output. While the underlying math operates using small decimals, the final Wrating is mathematically translated so that the baseline average player sits at exactly 1500. This provides a familiar, traditional rating scale while maintaining the rigorous Bayesian probabilities underneath.

Graph Theory: Ponds and Bridges

Wratings separates players into Ponds, rather than forcing all players into a single, unified list. This is to prevent mathematically invalid lists.

Imagine two isolated local tournament scenes—two separate ponds. If no player from Pond A has ever competed against a player from Pond B, there is zero mathematical evidence to compare them. Any system that ranks a player from Pond A above or below a player from Pond B is guessing.

Wratings solves this using graph theory. We map every match as a connection between players. A Pond in Wratings is a "disjoint connected component"—a mathematically isolated group of players.

For two Ponds to merge, players must travel between them. If a competitor from Pond A travels to a major tournament and plays someone from Pond B, they create a Bridge. The OpenSkill algorithm uses that bridge to calibrate the relative strength of both ponds, allowing them to be accurately ranked on the same list. If you are in Pond 2, it simply means you share no common opponents (direct or indirect) with the players in Pond 1.

Wratings presents the Ponds sorted in descending order, and ranked. Competitively, a player would aim to be the biggest fish, the top player, in the biggest Pond, Pond 1.

Dynamic Systems: Why Wrating Values Shift

Sudden shifts in the leaderboards—sometimes occurring overnight, even when a player hasn't entered a tournament—are both common and expected. This is the intended behavior of a dynamic Bayesian system, driven by three main factors:

1. The Sliding Time Window

Competitive form is not static. Wratings calculates performance across specific timeframes (e.g., a 100-Day Season). As time moves forward, the window slides. If a player won a major tournament exactly 101 days ago, that data point falls out of the current window today. The sudden loss of that positive evidence will cause their rating to adjust downward, reflecting only their recent activity.

2. The Ripple Effect

In a connected graph, every match affects the entire network. Suppose you defeat Player X at a local tournament. A month later, Player X travels to a major and defeats the world champion. The OpenSkill engine retroactively realizes that Player X is much stronger than initially estimated. Because Player X's μ increases, your past victory over them becomes mathematically more impressive. Your rating will increase today, based on a match someone else played yesterday.

3. Empirical Skill Decay

Most rating models include a "decay" variable, which artificially increases a player's uncertainty (σ) during periods of inactivity. To ensure mathematical rigor, we tested this assumption against years of historical Super Smash Bros. Ultimate data. The empirical evidence revealed that SSBU competitors experience remarkably little actual skill decay during typical absences. As a result, Wratings optimizes the OpenSkill decay parameter to near-zero. Your rating is driven by your actual match results, not by arbitrary time penalties.

A Living Record of Competition

Because of these mechanics, Wratings is not a static leaderboard—it is a living ecosystem. Every set played at a local weekly, a regional qualifier, or an international super-major adds a new thread to the web, refining the probabilities for everyone connected to it.

Acknowledgements

Wratings is made possible by standing on the shoulders of giants. We extend our deep gratitude to the following platforms, projects, and technologies:

• Start.gg: For providing the comprehensive tournament data and API infrastructure that makes this project possible.
• OpenSkill: For the open-source implementation of the Plackett-Luce model that powers our core mathematical engine.
• Manus AI: For autonomous agent assistance in architecture, coding, and deployment.
• Python & NetworkX: For the data science and graph theory ecosystem.
• Jsoftware.com: The Wratings project originally began as a local, then global Elo calculator written in the J programming language and run on a laptop before migrating to the Cloud in Python.
• Google Cloud: For the robust, scalable serverless infrastructure (Cloud Run & BigQuery) hosting our V2 Railway architecture.
• Cloudflare: For keeping the platform secure and performant.