Wong practice-table default
A claim about what the current game or companion does. Source code and regression tests are decisive for this scope.
Rules Atlas · How claims are reviewed
Methodology, sources and corrections
This project maps consequential Singapore Mahjong variants. It does not turn one household, club, competition or software implementation into a supposed national rulebook.
Maintained by Wong Hong Wei · Methodology reviewed · Rules registry version 2026.08.26
Method in one sentence
Name the scope, state the exact Wong practice-table result, compare documented alternatives without merging them, attach sources and a review date, then verify implementation claims with deterministic tests.
Four labels keep different evidence separate
Named external ruleset
A rule published by a competition, club, established reference or software implementation. Its publisher and limitation remain attached.
House or club variant
A convention reported by a real group but not yet independently documented. It is never promoted to a broad claim without review.
Proposed choice
A useful option not yet validated as an established Singapore convention. It is labeled plainly or kept out of curated profiles.
Review workflow
- Define the question narrowly. Bao trigger, duration and payment are separate facts.
- Record the implemented default. Inspect the final live code path and its tests, not superseded prose alone.
- Compare named sources. Capture publisher, type, scope, access date and limitation.
- Explain material effect. Show when a choice changes legality, Tai, payer, winner count or claim availability.
- Publish a worked example and fixture. The result must be reproducible rather than chosen by AI at runtime.
- Date and correct. Corrections update the registry, tests and affected pages together.
Public machine-readable registry
API index and compatibility policy
Stable entry point, media types, versioning rules and the current plus legacy fingerprint formats.
Sources
Publisher, source class, coverage, access date and limitation.
Rules
Stable IDs, dimensions, current default, material effect, source references and fixture IDs.
Wong practice profile
The current companion choices plus their expanded meaning and limitations.
Conformance examples
Public facts and expected outcomes backed by the larger deterministic suite.
Executable Ping Hu knowledge
The source-generated canonical record for SGMJ-PINGHU-001, including five hands tested directly against the playable scorer.
Profile choices schema
JSON Schema for every consequential scorer choice, including accepted values and normalization bounds.
Companion backup schema
The complete portable file shape for profiles, player names, ledgers, completed tables and drill results, with privacy and runtime-invariant notes.
Research submission contract
The pilot index for evidence-backed table profiles, expert-labelled AI positions and observed product-use findings.
Research submission schema
A privacy-safe JSON contract with explicit provenance, unknowns, contradictions, legal information boundaries and decision packets.
Research submission template
An empty, valid starting file. Its records are deliberately blank and are not evidence.
AI benchmark prediction template
Pair model choices and visible-information references with expert-reviewed positions, then score them with node scripts/score-mahjong-ai-benchmark.mjs SUBMISSION.json PREDICTIONS.json. The scorer keeps holdout results separate from tuning records and reports descriptive reviewer-pair agreement. For the real pilot, use at least three blinded independent reviewers per position. Anonymous IDs and automated checks cannot prove that reviewers were independent or that the model was blind to holdout labels.
Blinded expert-study protocol
Freeze stable position hashes, issue label-free packets with opaque item IDs, validate private reviewer responses and assemble only pseudonymous reviews into the public submission. Hashes reduce accidental mixing but cannot prove reviewer identity, independence, qualification or real-world blinding.
Versioning promise: additive fields may appear under v1 and consumers should ignore fields they do not recognize. Breaking document or meaning changes require a new API path. Material rule changes require a version change, updated fixtures and a dated explanation; private shared profiles remain non-indexed.
Publication and review status
| Layer | Current status | Boundary |
|---|---|---|
| English editorial copy | Repository editorial pass completed on 25 August 2026 | This automated and maintainer quality-control pass is not independent certification or named human language review. |
| Playable-table behaviour | Published fixtures and implementation claims are checked by deterministic repository tests. | Passing tests proves the stated software result, not that the rule is nationally prevalent or strategically best. |
| External ruleset comparison | Named public references are linked with their own scope and limitations. | The current registry is not a representative survey of Singapore tables. |
| Independent Mahjong expert review | Pending | No instructor, club or competition official is presented as an endorser. |
| Chinese terminology and publication | Pending fluent human review | Chinese characters on English pages are reference aids, not complete Chinese-language publication. |
| Field evidence | Pilot contract published; observations and expert labels not yet collected. | No prevalence, usability or AI-quality conclusion is inferred from an empty template. |
Automated and agent review is useful quality control, but it is not independent human expert sign-off. A future reviewer should be named only with permission, a declared role and a dated review scope.
Current limitations
- The first registry covers four consequential dispute topics; it is not a complete national survey.
- The compact companion bundles several Ping Hu, Bao and passed-claim dimensions. The registry exposes their expanded meaning without pretending each is independently selectable.
- No independent instructor or club is yet presented as a verified reviewer.
- English is the reviewed publication language. Full Chinese rule pages require human language review.
Corrections and contributions
Report a factual error, submit a documented source or describe how a real Singapore table plays differently. Include the exact rule, when it changes a result and whether the source is public. Material site corrections receive stable records in the public correction history.