About this chart
What the family tree shows, where its information comes from, and where it gets the history wrong.
What this is
A collection of material gathered from around the web and filed by variant, so you can check where each claim comes from. There are 66 entries, one page each, and each page lists what each source says about that ruleset — its tile count, its hand size, how it handles flowers, how it scores, who pays whom — beside the citation. The goal is to eventually write proper rulebooks from checked material.
The tree records descent as the source chart claims it. A variant's place in the tree is what the chart says, not a proven fact. Where the evidence is thin, the page says so.
Sources, and what was written from them
Every page keeps two kinds of material apart. Sources are the documents themselves: a citation, a link, and the date it was read. They are matters of fact in the plain sense that anyone can open them and see what they say.
AI-generated material is everything an AI model wrote from those sources: each variant's confidence (how good its sources are) and evidence (whether they agree), the cross-comparison of what each source says, the reading of it, the findings, the summaries of sources added while cross-checking, and the explanations further down this page. It is a reading, and it can be wrong. It always sits below the sources it came from, on a dashed rail, so that it can be checked against them and never mistaken for them.
Nothing here is final. Every page lists its sources, and corrections are welcome. The findings collect the places where sources contradict each other or say nothing; the references list every source used, with links to the variants each was cited for.
Where the material comes from
The tree was derived from a single compiled chart, kept in the repository beside the data generated from it. The chart draws on two main sources. Tom Sloper's variant catalogue [1] supplies the measurable facts for most leaf nodes: the tile counts, hand sizes and flower handling that tell one named variant from another. The Mahjong Wiki's family tree and timeline [7] supplies the branch structure and the decade-by-decade chronology this tree's shape follows.
The wiki is the weaker of the two. The timeline the branch structure rests on carries no citation on any page of the wiki: the 1950s adoption of Clear Chapter, the 1970s three-faan system and the late-1980s move to additive scoring are asserted and never sourced. What the wiki does credit is Cantonese — one named book [27], one named ruleset [28], and three Cantonese fan-list traditions with almost no English-language presence [29]. So the sourcing varies by region: the Hong Kong material is traceable, while most of the Japanese, Nanyang, Korean and mainland material is not.
How to read the tree
Colour marks the family a ruleset belongs to, and a link between two nodes means the source chart claims descent between them. Two of those links were drawn dashed in the chart, meaning no source establishes the descent at all: Nanyang variants and Korean traditional. Both are filed where the chart filed them, and tagged as guesses.
The confidence the analysis gives the sourcing behind the 66 entries breaks down as follows.
| Confidence | Meaning | Entries |
|---|---|---|
| documented | a published ruleset or rule description exists and is cited | 33 |
| catalogued | listed in a variant catalogue with tile counts and hand sizes, but no rules text verified | 26 |
| single-informant | the whole entry rests on one report from one person | 4 |
| disputed | the descent shown is contested; the chart drew this link dashed | 1 |
| placeholder | no source establishes the descent at all; the parent link is a guess | 1 |
| lost | no rules text survives | 1 |
The method page explains how the compendium is put together and what it leaves out.
Known limits of this tree
A tree can't show everything. Each of the chart's own warnings is quoted below, with the sources that bear on it, and then explained. The explanation is AI-generated.
Influence ran both ways between Hong Kong and Taiwan
Guangdong was influenced first by Hong Kong, later by Taiwanese; the arrow should run both ways.
Entries affected: Guangdong style, Taiwanese style · Sources: [42] [8] [28]
The chart gives Guangdong style one parent, Hong Kong old style, because a single parent is the only shape a tree can hold. The account behind it is messier: Guangdong play was shaped first by Hong Kong, and then by Taiwanese play once the 16-tile game had established itself across the strait.
The entry keeps the single arrow and records the rest as a caveat, so the parentage drawn here is a sourced claim rather than the whole history. The late-1980s move to additive scoring belongs to that second phase, and is asserted on the wiki's timeline without a citation anywhere — which is why the node can be well documented for its rules and still carry a warning about its chronology.
The mainland branch changed its frame mid-history
Mainland regionals began on a Chinese Classical frame and later adopted HKOS's simplified scoring.
Entries affected: Hong Kong old style · Sources: [8] [7]
The mainland regional games are filed under Shanghai / Hong Kong new style, which is where they began: a Chinese Classical frame, with the tile set and hand structure that implies. What they score with arrived later and from elsewhere, after Hong Kong old style had dropped fu in the 1950s and simplified scoring to something a table could total in its head.
A node has one position, so the tree records the frame and loses the borrowing. Read the eleven mainland entries as games with two inheritances: an older skeleton, and a newer scoring layer that came back up the branch from Hong Kong.
Some scoring travelled backwards, from West to East
All Pairs and All Green flowed backwards from Western Mah Jong into most other modern rulesets, meaning the us branch fed the cn branch, which no tree layout shows.
Entries affected: Babcock rules, Western / British Empire style · Sources: [32]
A family tree can only draw divergence. This one shows the Western branch leaving the Chinese line in 1920 with Babcock's rules and never rejoining it, yet the wiki's own Western Mah Jong scoring page credits British play with All Green, Seven Pairs and the Jewel Dragons — hands that then entered rulesets across the Chinese branch.
No book is named for that attribution, so it is recorded here as the wiki's claim rather than as an established fact. The direction still matters for anyone using the tree to date a hand: a pattern found in a Chinese ruleset is not automatically older than the Western one beside it.
Eleven nodes stand in for more than a hundred games
The eleven mainland nodes stand in for a hundred-plus real variants, most undocumented in English.
Entries affected: Sichuan / Hunan / Tibetan, Sichuan bloody rules, Sichuan 72-tile game · Sources: [1] [7]
The mainland branch is the most compressed part of the tree. Its eleven entries are the ones with an English-language description attached; the real distribution of play is regional to a degree no tree of this size could represent, and most of those games have never been written up in English at all.
The effect is that the tree's shape is partly a map of English-language documentation rather than of play. A thin branch is not evidence that little is played there: Sichuan alone carries three entries here and far more variation than three in practice.
Three entries rest on one person's report
Several entries (Shenzhen, Tui Dao Hu, Fujian) rest on a single informant report.
Entries affected: Shenzhen style, Tui Dao Hu, Fujian style, Sichuan 72-tile game · Sources: [1] [35] [24]
Shenzhen style, Tui Dao Hu and Fujian style each come from a single informant, material of that kind having largely been argued out in the rec.games.mahjong archive before reaching the chart through Sloper's variant catalogue. Nothing independent confirms any of the three. The compendium tags a fourth the chart's own line does not name: the Sichuan 72-tile game, reported by one person.
They are kept rather than dropped, and tagged single-informant so the weakness travels with the entry instead of being lost in the tree around it. One report is a lead worth filing; it is not two sources agreeing, and nothing here quietly promotes it to one.
The Korean 104-tile game has no established parent
No source establishes how Korean 104-tile mahjong descends from anything; the link is a placeholder.
Entries affected: Korean traditional · Sources: [1] [7]
Korean traditional play uses 104 tiles and drops the bamboo suit entirely, which puts it structurally further from its neighbours than any other entry on the tree. The chart hangs it under Chinese classical and draws the link dashed: no source examined establishes that descent, or proposes another.
The entry is tagged placeholder for exactly that reason, and it is one of only two dashed links in the chart — the Nanyang group is the other. The honest reading is that its position is a filing decision, not a finding.
Filipino style sits on a branch its hand size argues against
Filipino style sits under Nanyang despite a 16-tile hand resembling Taiwanese. The distinguishing test is that Filipino play treats winds and dragons as flowers.
Entries affected: Filipino style, Nanyang variants · Sources: [1] [7]
Filipino style is played with a 16-tile hand, which is the structural signature of the Taiwanese fork, and yet the chart files it under the Nanyang group. What keeps it there is its treatment of honours: Filipino play handles winds and dragons as flowers, which Taiwanese play does not.
That leaves the entry two levels of doubt deep, because the Nanyang link above it is itself dashed and tagged disputed. A shared hand size is evidence of structural kinship, not proof of descent, and this is the clearest case in the tree of those two things coming apart.
Playing some of them
Six of the 62 rulesets have a scoring tool. Each deals a winning hand and scores it by a specific rulebook: riichi under the European Mahjong Association’s rules, modern Hong Kong under the Hong Kong Mahjong Association’s, Hong Kong old style by the “Clear Chapter” fan list, and Taiwanese in tai over seventeen tiles, and the two Vietnamese games in phán, with jokers that stand in for other tiles. Side by side, they show that the same fourteen tiles score differently from one ruleset to the next, in Taiwan they are not even fourteen, and in Vietnam some of them are not the tiles they appear to be.