Classical rules comparator audit v1

Status: reviewed calibration audit; no public comparator implementation yet
Date: 20 September 2026
Issue: #296
Profiles: bmja@1.0, western-tm@0.1, outside-the-box@0.1, buzzard-2000@0.1

Purpose

Use the four executable Classical-family profiles to establish a durable comparison vocabulary before MCR introduces a different scoring grammar.

This audit answers four questions:

Which dimensions are useful enough to become stable comparator rows?
Which values can come directly from executable profile/runtime truth and which still need source-backed reference data?
Can the model distinguish exact sameness from provisional implementation reuse, parameter changes, binding changes, policy changes, broader/narrower rules and genuinely profile-specific behaviour?
Which dimensions are useful later for Which rules am I playing?

This is a model/data audit, not a request to build the public UI and not a new rules-research pass.

Evidence discipline

A comparator cell must never collapse these two statements:

The current executable profiles behave the same here.

and

The underlying rules sources establish that these profiles are the same here.

That distinction is especially important for western-tm@0.1. Its Companion special-hand catalogue is source-certified, but ordinary rules that currently reuse BMJA-compatible behaviour remain provisional pending the intended ordinary-rules source in #121.

The useful cell-level evidence states are therefore:

source-verified — the intended source/profile evidence establishes the rule;
verified-club — the named club's own source establishes the rule;
source-provisional — executable/current value exists but source equivalence is not yet established;
source-unknown — current evidence does not establish the rule;
runtime-executable — the current sealed profile executes the value/strategy;
runtime-partial — some related behaviour is executable but the whole source rule is not represented;
runtime-not-modelled — the rule may exist at the real table, but the current companion does not model it.

Important absence rule

absent, unknown, and not modelled are not synonyms.

A future data contract must be able to distinguish at least:

rule status:     present | absent | unknown | not-applicable
runtime support: executable | partial | reference-only/not-modelled

For example, the current BMJA executable profile having no automatic Goulash hand-mode strategy does not prove that Goulash is absent from British source material.

Executable family baseline

The sealed current profiles already prove a substantial shared family core:

Runtime dimension	BMJA	T&M Western	Outside the Box	Buzzard 2000
Family	`family.classical-western`	same	same	same
Grammar	`classical-points-doubles`	same	same	same
Players	4	4	4	4
Seat model	E/S/W/N	same	same	same
Tile preset	`tiles.flowers-144`	same	same	same
Ordinary hand shape	four sets + pair	same	same	same
Scorer	current Classical scorer	same	same	same
Profile binding	BMJA	T&M	OTB	Buzzard
Profile scoring policy	BMJA	T&M	OTB	Buzzard
Default table limit	1,000	1,000	1,000	600
Ordinary progression strategy	Classical East cycle	same	same	same
Game-end strategy	Classical East cycle	same	same	same
Hand mode	normal only	normal only	normal ↔ Goulash	normal only
Settlement strategy	Classical pairwise	Classical pairwise	OTB incidents over Classical core	Buzzard incidents over Classical core
Validation	current Classical	current Classical	current Classical	Buzzard Classical variant

The runtime therefore supports a real family comparison rather than four unrelated implementations. It also already demonstrates that shared family/grammar does not imply identical profile behaviour.

Reviewed comparator dimensions

The following is the recommended v1 durable row catalogue. It deliberately prefers dimensions a player can understand over internal implementation seams.

A. Identity, table and tiles

Dimension ID	Public label	BMJA	T&M Western	Outside the Box	Buzzard 2000	Evidence / relationship note	Diagnostic value
`profile.identity`	Rules profile	BMJA 1.0	T&M Western 0.1	Outside the Box 0.1	Buzzard 2000 0.1	Exact version must remain visible in provenance	low
`profile.authority-status`	Rules/source status	published project profile	provisional ordinary profile	named club profile	named historical profile; source-complete for companion scope	Do not flatten these authority types	medium
`table.player-count`	Players	4	4	4	4	Runtime exact; T&M source relationship remains provisional	low within Classical; high cross-family
`table.seat-model`	Seats	East/South/West/North	same runtime	same	same	Common executable Classical core	low within Classical
`tiles.physical-set`	Main tile set	144-tile Flowers/Seasons preset	same runtime; ordinary source provisional	same ordinary runtime	same runtime	Runtime exact; public source wording must preserve T&M caveat	low within Classical; high cross-family
`tiles.bonus-tiles`	Flowers & Seasons	used	used in current runtime; source provisional	used	used	Basic presence shared; scoring treatment is not necessarily identical	medium
`tiles.substitute-special`	Jokers, blanks or substitutes	no substitute mechanism in current executable profile	no substitute mechanism in current executable profile; source unknown	4 blanks in Goulash	no substitute mechanism in companion evidence	OTB is the clear executable differentiator. Do not call BMJA/T&M source-absent from runtime silence	high

B. Hand structure and play

Dimension ID	Public label	BMJA	T&M Western	Outside the Box	Buzzard 2000	Evidence / relationship note	Diagnostic value
`hand.structural-size`	Normal hand size	13 before completion / 14 winning, Kongs add physical copies	same current Classical model	same ordinary model	same Classical model	BMJA source/project verified; T&M ordinary source provisional	low within Classical
`hand.normal-winning-shape`	Normal winning shape	four sets + pair	same runtime; source provisional	British/Classical four sets + pair	four sets + pair	Shared structural core	low within Classical; high cross-family
`play.ordinary-chow-policy`	Chows in an ordinary hand	maximum 1	current runtime maximum 1; source unverified	current ordinary runtime maximum 1; exact club-source ordinary cap not separately established	multiple Chows permitted; BMJA one-Chow validation deliberately removed	narrower/broader proof: Buzzard is broader than BMJA on this dimension	high
`play.exposure-model`	Exposed vs concealed	materially affects scoring/specials	Companion bindings explicitly use exposure; ordinary baseline provisional	materially affects ordinary/special scoring	materially affects scoring	Shared concept, profile-specific policies	medium
`play.rob-kong`	Robbing a Kong affects scoring	yes	current runtime compatible; source baseline provisional	yes in profile policy	yes	Same broad concept, source confidence differs	medium
`hand.irregular-special-model`	Special/irregular hands	curated BMJA patterns/events	very large Companion catalogue	BMJA + Western-derived + OTB-local catalogue	ten source-defined limit hands plus Classical ordinary rules	Same broad dimension, strongly different bindings/catalogues	high

C. Ready / declaration concepts

Dimension ID	Public label	BMJA	T&M Western	Outside the Box	Buzzard 2000	Evidence / relationship note	Diagnostic value
`ready.model`	Ready / waiting concepts	special-hand fishing; Original Call is separate	Companion gives winning/fishing treatments; ordinary context still provisional	winning/fishing values in club special-hand catalogue	Calling exists; Calling player may declare Standing Hand, which adds +100 if completed	related analogue, not equivalence. Standing Hand must not be mapped to BMJA Original Call	high
`ready.fishing-score`	Non-winning fishing score	supported for many special hands	source-certified in Companion catalogue	explicit winning/fishing pairs	not the BMJA/T&M fishing-value model; separate Calling/Standing semantics	Do not force one universal `ready` boolean	high

D. Scoring

Dimension ID	Public label	BMJA	T&M Western	Outside the Box	Buzzard 2000	Evidence / relationship note	Diagnostic value
`scoring.grammar`	Scoring system	points + doubles	points + doubles runtime	points + doubles	points + doubles	Exact executable family sameness; important future cross-family row	low within Classical; very high cross-family
`scoring.ordinary-base`	Basic Pung/Kong/pair/bonus points	BMJA Classical table	current runtime compatible; ordinary source unverified	source-verified club table matches BMJA core	source table directly matches Classical core	BMJA↔OTB and BMJA↔Buzzard are strong exact/shared evidence; T&M is `same-provisional`	low-medium
`scoring.winner-base-bonus`	Basic Mahjong / self-draw bonuses	+20 / +2 live-wall	current runtime compatible; source provisional	+20 / +2	+20 / +2	Shared executable core; source status differs	low
`scoring.multiplier-bonus-policy`	Doubles and extra bonuses	BMJA doubles including own/round Wind, Dragons and profile rules	current profile + Companion bindings; ordinary source provisional	shared Classical doubles plus OTB-local additions/removals	shared Classical doubles plus significant ADD/AMEND/REMOVE set, including Standing +100 and several +10/+2 bonuses	Same grammar does not mean same policy	high
`scoring.limit-cap`	Normal limit / table limit	normally 1,000	runtime default 1,000; source baseline unresolved and Companion includes values above 1,000	1,000 configured ordinary limit	table-configured; 600 source-derived default/example	Clear parameter variation; T&M value remains implementation not source proof	high
`scoring.limit-composition`	How limits combine with bonus scoring	BMJA fixed-special/bonus-tile rules	profile-specific Companion treatments	fixed special may retain independently calculated Flower/Season side score above limit (e.g. source-backed 1,008 case)	source-defined agreed limit hands; limit amount configurable	policy variation, not just different cap number	medium
`scoring.special-catalogue`	Named special-hand / pattern catalogue	narrow BMJA catalogue (18 inventoried bindings plus calculated/event rules)	84 unique source hands / 85 bindings	33 unique named club hands	10 named source limit hands	Major differentiator; counts are profile-local, not universal concepts	very high
`scoring.special-binding`	Values/exposure for the same structural pattern	BMJA-local	T&M-local	often reuses structure but overrides value/exposure	Buzzard-local limit bindings	Canonical structure can be shared while name/value/exposure/membership differ	very high
`scoring.nonwinner-special-result`	Exceptional non-winner special score	no equivalent executable special-result seam	no equivalent current profile seam	no equivalent current profile seam	incomplete Four-Wind / Three-Dragon achievements can score limit against other losers	Buzzard-specific architecture/result semantics	high
`scoring.decomposition-policy`	How alternate hand interpretations are handled	Classical scorer chooses lawful interpretation under profile bindings	same engine/profile bindings	same engine/profile bindings	same engine with Buzzard validation/policy	Keep as durable cross-family dimension; not highly differentiating within Classical	low within Classical; high cross-family

E. Settlement and incidents

Dimension ID	Public label	BMJA	T&M Western	Outside the Box	Buzzard 2000	Evidence / relationship note	Diagnostic value
`settlement.winner-payments`	Who pays the winner?	each loser pays winner	same runtime; source ordinary settlement unresolved	same Classical core	same Classical core unless liability overrides	Shared executable core; T&M source status remains provisional	medium
`settlement.loser-to-loser`	Do non-winners settle with each other?	yes, pairwise score differences	yes in current runtime; source verification still required	yes, verified-club	yes, source-backed	Excellent example of `same-provisional` for T&M	high
`settlement.dealer-multiplier`	East/dealer payment multiplier	East pays/receives double	same runtime; source provisional	same, verified-club	same, source-backed	Shared core; cross-family discriminator later	medium
`settlement.liability`	Liability / one player pays for others	current profile has no dedicated incident strategy; do not infer source absence	current profile has no dedicated incident strategy; source unknown	Cannon with `No choice!` exception and other club incidents	dangerous-discard full-payment liability for source-defined dangerous hands	OTB and Buzzard are related analogues with materially different policy, not one universal cannon rule	very high
`settlement.false-mahjong`	False Mahjong consequence	not modelled as a dedicated current-profile incident here; source procedure exists	not modelled; source unknown	club-specific consequence depends on exposure	Buzzard: fully exposed invalid Mahjong pays double limit to each other player; withdrawal possible if not fully exposed	Same label can hide materially different semantics	high
`settlement.incorrect-hand`	Wrong tile-count consequence	current profile does not use a dedicated profile incident strategy	source/runtime profile-specific treatment unresolved	OTB-specific too-few/too-many consequences	Buzzard-specific too-few/too-many settlement consequences	Shared incident concept, profile-specific policy	high

F. Draws, dealer and progression

Dimension ID	Public label	BMJA	T&M Western	Outside the Box	Buzzard 2000	Evidence / relationship note	Diagnostic value
`round.draw-settlement`	What happens financially on a draw?	no transfers in current product interpretation	same current runtime; source provisional	no score/settlement	dead hand: no scoring	Shared current result, evidence strength differs	medium
`round.draw-follow-up`	What happens after a draw?	East retained; current executable hand mode remains normal	same runtime; source provisional	East retained and next hand is Goulash	East retained; normal mode	OTB's automatic Goulash transition is a strong unique discriminator	very high
`progression.dealer-retention`	When does East stay East?	after East win or draw	same runtime; source provisional	same current progression	source-backed same	Shared Classical progression	medium
`progression.seat-rotation`	When do seats rotate?	non-East win	same runtime; source provisional	same current progression	source-backed same	Shared Classical progression	low-medium
`progression.prevailing-wind`	Prevailing Wind progression	East → South → West → North over cycles	same runtime; source provisional	same current progression	same, source-backed	Shared current strategy	medium
`progression.game-end`	Full-game end	after North prevailing-Wind cycle in current product	same current runtime; source ordinary rules provisional	current profile uses same product strategy	source describes complete game as four rounds	Same executable strategy; source confidence differs	medium

Relationship vocabulary proved by the four profiles

The proposed #296 relationship vocabulary is not theoretical; the current family supplies real examples for every important class except a source-proven hard absent case, which should remain deliberately rare.

Relationship	Classical proof case
`exact-same`	BMJA ↔ OTB ordinary intrinsic Pung/Kong values; BMJA ↔ Buzzard ordinary base score table; BMJA ↔ Buzzard ordinary settlement shape
`same-provisional`	BMJA ↔ T&M current ordinary settlement/progression/validation behaviour pending #121 source verification
`parameter-variation`	configured/default table limit 1,000 vs Buzzard 600
`binding-variation`	same canonical hand structure with different T&M/OTB/BMJA value/exposure bindings; e.g. Three Great Scholars / Four Blessings families
`policy-variation`	OTB vs Buzzard liability, false-Mahjong and incorrect-hand consequences
`narrower` / `broader`	BMJA maximum-one-Chow ordinary validation vs Buzzard's broader ordinary Chow allowance
`related-analogue`	OTB cannon and Buzzard dangerous-discard liability; BMJA/T&M/OTB fishing concepts versus Buzzard Calling/Standing concepts
`unique`	OTB executable draw→Goulash hand mode; Buzzard Standing Hand bonus and exceptional non-winner limit result
`absent`	Require direct source evidence before using this label; runtime omission alone is insufficient
`unknown`	T&M ordinary settlement/source rules pending #121; exact OTB ordinary Chow-source cap where not directly established

This is enough evidence to retain the relationship vocabulary.

Strongest player-visible differentiators

These are the most promising inputs for a later diagnostic because a normal player could plausibly observe or recognise them at a table.

What happens after a drawn hand? OTB's executable draw→Goulash transition is highly distinctive.
Are ordinary hands restricted to one Chow? Buzzard deliberately removes the BMJA one-Chow restriction.
Do you use Calling / Standing Hand and a +100 Standing bonus? Strong Buzzard signal.
Which special-hand catalogue looks familiar? T&M's large Companion catalogue, OTB's 33-hand club selection, BMJA's narrower catalogue and Buzzard's ten limit hands differ substantially.
What table limit do you normally agree? 600 is a Buzzard source example/default while the current British/OTB profiles use 1,000, but this must be treated as a configurable clue rather than proof.
Do you use Cannon / dangerous-discard liability, and what exactly happens? Similar vocabulary can separate OTB and Buzzard only if the consequence is asked, not merely whether “cannon” exists.
Do non-winners settle score differences with each other? Useful across future families; within current Classical it is mostly shared, with T&M source status still provisional.

Diagnostic questions should prefer observable consequences over asking users to name their ruleset or know specialist terminology.

Encyclopaedia link map

The audit supports the #251 join without requiring one page per row.

Dimension family	Likely canonical concept / section
player count / seats / East	Players, seats and Winds
physical tiles / Flowers / blanks	Mahjong tile sets; Flowers and Seasons; substitute tiles
hand size / normal shape / irregular shapes	Mahjong hand structure
Chows / claims / exposure	Chows and claiming; exposed and concealed sets
ready / fishing / Calling / Standing	Ready hands, Fishing, Calling, Tenpai and declarations
grammar / ordinary points / doubles	Mahjong scoring systems; Classical points and doubles
limits	Limits, caps and maximum scores
special catalogue / bindings	Named hands and profile treatments
settlement	Mahjong settlement / who pays whom
liability / penalties	Liability and penalties
draws / Goulash	Drawn hands / wash-outs / Goulash
East retention / Winds / game end	Dealer/East and round progression

A row may link to the broad concept and an individual cell may deep-link to that profile's treatment/anchor.

Data-contract consequences before UI work

The audit exposes several requirements that should be settled before implementing the public comparison table.

1. Values need provenance and support state, not just display text

A profile cell should conceptually carry:

dimensionId
profileRef
structuredRuleValue
humanDisplay
ruleStatus
runtimeSupport
source/provenance reference
qualification/caveat
conceptRef + optional anchor
diagnostic metadata

Do not use shared strings as semantic identity.

2. Relationship is not always derivable from equal values

1000 === 1000 can prove a numeric match, but not source equivalence. Reviewed relationship/provenance may therefore need to accompany derived comparison output where semantics matter.

3. Some dimensions are scalar; others are structured

Examples:

player count: scalar;
table limit: numeric/configurable value;
tile set: composition object;
ready model: typed concept set / policy;
liability: incident + trigger + consequence policy;
catalogue: profile binding collection.

Do not reduce every dimension to string | number merely because the first UI can render text.

4. Difference-only mode must keep uncertainty visible

Two cells with the same current runtime value are not safely collapsible if one is source-verified and the other source-provisional. Differences only must therefore be able to retain a row because evidence status differs even when display values match.

5. Runtime truth and reference truth should join, not duplicate

Where a sealed profile already owns a value such as player count, grammar, table-limit default or strategy identity, comparator data should reference/derive from that truth. Source-only/procedural dimensions may live in a reviewed reference layer. The comparator must not copy scoring arithmetic into presentation data.

MCR falsification gate

Do not build a large public content engine from this Classical-only model yet.

After MCR A/B establishes executable pattern-accumulator scoring and table semantics, attempt to add MCR as another column using the same model.

The model passes if MCR can express, without comparator-specific hacks:

four-player/table/tile facts;
ordinary winning structure;
pattern-accumulator scoring grammar;
eight-point minimum qualification;
interaction/non-combination policy;
Flowers after qualification;
different settlement shape;
no Classical loser-to-loser/East-multiplier assumptions where not applicable;
MCR dealer/progression/game-end semantics;
source/version/provenance status.

MCR may legitimately add a small number of new durable dimensions such as qualification order or pattern-interaction policy. That is extension of the catalogue, not failure. The failure condition is needing MCR-only ad-hoc fields because the profile/dimension/concept model cannot represent another grammar cleanly.

Audit conclusion

The Classical family is sufficiently diverse to define the comparator foundation now.

The durable model is:

verified runtime/source truth
        ↓
stable comparison dimension
        ↓
profile-specific value + evidence/support state
        ↓
reviewed relationship / differences view
        ↓
Encyclopaedia concept + profile treatment
        ↓
high-discrimination dimensions feed the diagnostic

The four-profile audit supports moving on to MCR rather than adding another Classical profile first. The next useful comparison work is not a large UI build: it is to keep this audit as the v1 model, let MCR falsify it, then implement the mature reusable data/UI layer against at least two scoring grammars.