Method

How the compendium is put together.

Where the pages come from

Every page here is generated from one set of data: the node tree, the bibliography, and the per-variant claims. Each fact is written once, so the pages can't disagree with each other.

Sources first, analysis after

The site carries two kinds of material and never mixes them. Sources are the documents: what each one is, where it can be read, and when it was checked. AI-generated material is what an AI model wrote from them: the confidence and evidence verdicts, the cross-comparison and its reading, the findings, and the summaries of sources added while cross-checking.

On a variant page the sources come first, on a solid gold rail. The analysis follows, below them, on a dashed grey-blue one, and every figure in it is footnoted back to a numbered source above. Other pages mark the same line the same way: the references are sources, the findings are analysis, and wherever the two meet on one page the analysis carries the badge. Where analysis and source disagree, the source is right.

Only what was read

A claim enters the data only if the source was fetched and read. Search-engine summaries don't count. A page that couldn't be opened is marked unreachable. Where a source contradicts itself, both readings are recorded. One entry wasn't read at source, and its page says so.

Confidence and evidence

Confidence is how good a variant's sources are. Evidence is whether they agree. A variant can have one good source, or several weak ones. Both are the analysis's verdicts, not properties of the sources, and are shown with the rest of it.

How sources are cited

Every variant page opens with its own numbered sources, ahead of any analysis, with the full citation, a link to the source itself, and a note on what it covers. Numbering is local to each page, so [1] on one variant is not [1] on another; each footnote names its entry in the bibliography, which indexes every source across the whole compendium. Each footnote also lists the other variants citing the same source, so you can see every variant that uses it.

A source is marked canonical where the chart drew the variant from it, supporting where the chart cites it alongside, and cross-check where it was checked afterwards to test the chart's claims.

Artwork

The tiles on the tool pages are FluffyStuff’s riichi tile set, in the public domain under CC0. The site’s icon is the scroll from Twemoji, © Twitter, Inc. and other contributors, licensed under CC BY 4.0.

What isn't included

Copyrighted rulesets: the NMJL and AMJA annual cards for any year, Mah Jongg Made Easy, and any variant's in-copyright rulebook. The rules are noted as facts, like tile counts, but the text of those works isn't copied.

Provenance vocabulary

For a collected file, in the ledger table:

| Value | Meaning | | --- | --- | | primary | The ruleset itself, from the body that issued it | | facsimile | A scan or verbatim reproduction of a primary source | | secondary | Someone's description or translation of a ruleset | | informant | An account of how a game is played, from a player, with no published rules behind it | | derived | Assembled from other sources, including anything reconstructed here | | unknown | Origin not established |

For a variant folder, in the README front matter and the root glossary:

| Value | Meaning | | --- | --- | | documented | A published ruleset or rule description exists and is cited | | catalogued | Listed in a variant catalogue with tile counts and hand sizes, but no rules text has been verified | | single-informant | The whole entry rests on one report from one person | | disputed | The descent shown is contested; the chart drew this link dashed | | placeholder | No source establishes the descent at all; the parent link is a guess | | lost | No rules text survives |

These describe the sources, not the game. placeholder means nobody has established where a game came from, not that the game is doubtful.