What is the Mahjong Rating Standard?

A score tells one story.

The Mahjong Rating Standard (MRS) estimates playing strength from match placement, opponent strength, and uncertainty. It is separate from the points a hand earns. Mahjong implements a Taiwanese rating; the wider cross-variant standard remains a proposal.

The Mahjong Rating Standard (MRS) is the framework we are developing to do that. Its ambition is a shared, trustworthy language for player skill across the mahjong community, while respecting the rules that make each variant distinct.

The distinction starts at the table. Native scoring decides the result of your match. A separate rating model uses that result to update an estimate of your playing strength. Fan, tai, and the size of a payout are never simply converted into MRS points.

Follow a result into a rating.

Suppose a qualifying 16-hand match finishes with Alex on 560 points, Bo and Casey tied on 510, and Drew on 420. The four balances total the original 2,000 points. Alex places first, Bo and Casey share second, and Drew places fourth.

The rating model receives those placements alongside the players’ prior ratings and uncertainty. Sixteen hands supply one unit of match evidence. It does not add Alex’s 60-point table gain to Alex’s MRS: a numerical rating change cannot be calculated from these balances alone.

This is an illustrative result, not a live match or a prediction. The same ledger entered as an in-person game would remain unranked. See table scorekeeping and ranked versus practice play for the distinction.

Built for a four-player game.

MRS v1 proposes a Bayesian Plackett–Luce model. In plain language, it considers the order in which all four players finish, the strength of their opponents, and how much is already known about each player. It updates the four estimates together.

The same finish can mean different things at different tables. Second place against a much stronger field can be evidence of improvement. Second place when you were strongly expected to win can lead to a lower estimate. A fixed reward for finishing second would miss that context.

Genuine ties belong in the result as ties. Breaking them by seat position or by who happened to be recorded first would add information the match did not provide.

The proposed starting point is 1,500, on a familiar 400-point Elo-like scale. It is a starting estimate, not an assessment made before you have played.

Confidence belongs beside the number.

A rating should communicate what it does not yet know. MRS pairs its estimate with uncertainty, sometimes called rating deviation or RD. A new profile begins with high uncertainty; more evidence helps the estimate settle.

That is why a new player’s rating can move more quickly than an established player’s. The system is learning where the new player belongs. The v1 outline recommends an initial RD of 350 and readable confidence labels such as Provisional and Established.

Match length matters, too. Sixteen completed hands are the proposed reference for one unit of evidence. Shorter matches contribute less, and very long sessions have a capped weight. These parameters need calibration against simulations and real match data; they are not proof that luck has disappeared.

One framework. Distinct traditions.

Hong Kong and Taiwanese mahjong share a heritage, but a single blended number would assume that performance transfers between them in a known way. MRS does not make that assumption.

The outline proposes separate MRS-HK and MRS-TW ratings. Within each variant, every rated match should identify the exact version of the rules and rating algorithm used. Scoring options, dealer rules, and tie handling can change the meaning of a result; a label as broad as “Taiwanese” is not enough.

The Mahjong app currently implements MRS for its LRC Taiwanese rules, based on the December 2024 publication. Hong Kong support is part of the wider MRS proposal; it is not an available app feature being announced here.

A standard worth sharing.

We believe these principles make a strong foundation for a community standard: results understood in context, visible uncertainty, rules that can be identified, and a record of how each update was reached.

The proposed official standard requires verified rated matches. A durable match record should preserve the participants, final placements, completed hands, rules identity, algorithm version, and before-and-after ratings. Verification must also say what was checked: a result recorded by a local game engine is different evidence from a result checked by an independent server.

MRS is designed as a lifetime measure of skill. Seasonal rewards, experience points, purchases, and win-streak incentives should stay separate. A new season should not erase what the system has learned about your play.

Our aim is for MRS to become a leading standard for mahjong. That standing has to be earned through transparent specifications, independent scrutiny, reliable prediction, and community adoption. MRS is an initiative of Orion Designs LLC, not a ranking endorsed by a governing body or a claim that one tradition is superior to another.

For the player, the ambition is simpler: a number that helps you understand your progress, and a reason to return to the next hand with curiosity.