Markdown for LLMs

Vocabulary: corpus forms

The source Markdown for this article. Copy it into your assistant or download it as a text file.

# Vocabulary: corpus forms

Breakdown of real fragments: package identifier, human act name, form rating
(exemplary / debatable / wrong form choice) and why.

## Standards

### 1. Scrum — public enumeration with a label (exemplary)

Package `scrum.guide` (Scrum Guide), team:

```law
pub enum DeveloperAccountability {
    label en official "what the Developers are always accountable for";
    CreatingSprintBacklog,
    AdheringToDefinitionOfDone,
    AdaptingPlanDaily,
    HoldingEachOtherAccountable
```

Exemplary: a closed list of accountabilities is exactly an `enum`, not four
boolean relations; `pub` exports the type and every member at once (there
is no partly-public enumeration). Same file: a second
enumeration `ProductOwnerAccountability` — neighbouring lists live side by
side without mixing (members of different enumerations are incomparable).

## Law

### 2. OFAC 50% rule — roles as subtypes (exemplary)

Package `us.ofac.fifty_percent` (OFAC 50% rule, US sanctions):

```law
entity BlockedPerson : Party { label en official "blocked person"; }
entity Entity : Party { label en official "entity"; }
```

Exemplary: a party role as a subtype, not a flag relation. A subtype
is the only subtyping: `BlockedPerson` fits a `Party` slot, the reverse
is `LDC-E2104`. Time- or case-dependent roles would be relations, not
subtypes — here the roles are permanent, the choice is right.

### 3. FNMA income rules — key with a comment on the honest conflict (exemplary)

Package `us.fnma.selling_income` (FNMA income rules, US), ontology:

```law
relation military_base_pay_monthly(src: IncomeSource, amount: Money) kind empirical {
    label en official "the borrower's monthly military base pay";
    key(src);
}
```

Further in the file — a comment: submitting both inputs at once is “a case
contradiction that honestly gives KEY_CONFLICT, not a silent pick”.
Exemplary twice over: both the key where the source has one sum,
and documenting the intent — the file reader knows the conflict is designed,
not accidental.

## Historical law

### 4. 1574 apparel statutes — named individuals as constants (exemplary)

Package `eng.corpus.statutes_of_apparel_1574` (Statutes of Apparel 1574, England):

```law
const Sovereign: Rank = entity_ref("urn:eng:corpus:clir:statutes-of-apparel-1574#Sovereign") { label en unofficial "rank: sovereign"; };
```

Exemplary: a rank gets a name only this way — `entity_ref` gives the value,
`const` gives the name; the label in the body names it
for the reader. A series of such constants further down the file — the act
names the estate vocabulary whole. Nine of the corpus’s ten constants are
named individuals, not numbers.

## Measurement and form boundaries

### 5. Kind distribution

`kind empirical` — 45 142 in 486 packages (case facts in almost every act);
`institutional` — 33 211 in 487; `derived` — 1 109 in 24; `key(…)` — 2 722
in 147; explicit `internal` — 0 packages (nobody writes the default).
Conclusion: an average act’s vocabulary is case empirics plus institutional
statuses; `derived` marks the inferred and is rare — rule heads more often
live in `institutional` without remarking, and that is the norm, not an
omission.

### 5a. Constants, subtypes, enumerations

`const Name: Type = entity_ref(…) { label … };` — 8 606 in 189 packages:
nine of the corpus’s ten constants are named individuals, not numbers.
`const …: Decimal` / `Integer` / `Natural` — 166 (thresholds and
coefficients); `const …: Money = 500.00 USD { … }` — 18; there are no
const-percents — a threshold is written as a literal in the guard.
`entity A : B { label … }` — 426 in 41 packages. `pub enum … { label …;
Member, … }` — 149 of 850 enumerations.

### 6. Judgment channel — vocabulary with an organ

`external judgment relation … { authority …; }` — 366 in 140 packages.
A vocabulary form, not a rule one: the relation answered by an organ
is declared in the vocabulary with an `authority`, and rules read it as an
ordinary literal. While no answer exists — Not established, not refuted with
`REQUIRES_JUDGMENT` regardless of strength.

## Teaching packages

### 7. This section’s examples (runs green)

Package `research.vocabulary.company_registry` — entities, a `const` individual,
`const Decimal`, `enum`, keys, three reader rules: `check OK`, `law test`
2/2. Package `research.vocabulary.subtype_key` — subtype and key: `check OK`,
`law test` 2/2, a two-region conflict gives Established + `KEY_CONFLICT`
(`expect issue(KEY_CONFLICT)` passes).