Markdown for LLMs

Facts and evidence: corpus forms

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

# Facts and evidence: corpus forms

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

## Law

### 1. Mandatory Social Health Insurance (Kazakhstan) — block fact with the term text as label (exemplary)

Package `kz.corpus.osms` (Mandatory Social Health Insurance, Kazakhstan), definitions:

```law
assert defined_term(Term1) {
    label ru-KZ official "Государственная корпорация "Правительство для граждан" (далее – …)";
}
```

Exemplary: the act’s vocabulary as facts — the term as the value (`Term1`),
the official text as the label, not a second fact. Block form (3 969 in the
corpus) where `id` or `label` is needed; one-line form (10 939) for bare
reference facts. Form choice by content, not by habit.

### 2. National Bank exchange rates — one-line handbook facts (exemplary)

Package `kz.national_bank.fx` (National Bank of Kazakhstan, exchange rates) — rates as package
facts: part of the program, visible in every case against the package.
Exemplary for release-changing data: a new release is a new fact file,
rules untouched. A package fact is not “a default the case overrides”:
there is no overriding, supports add up.

### 3. Codes — facts with `id` for article anchors (exemplary)

Packages `kz.corpus.appc` (Administrative Procedure Code of Kazakhstan) and
`kz.corpus.osms` (Mandatory Social Health Insurance, Kazakhstan):
2 402 facts with `id` in 7 packages. Exemplary: an explicit `id` where the
fact is referenced from outside the file — proof-graph premises address it
by StableId, and a case is hashed whole, so the case hash must
not change on renumbering.

## Science

### 4. Kazakh grammar — fact-group export (exemplary)

Package `kz.grammar.lexicon` (Kazakh language lexicon):

```law
pub facts KirispeMysaldary {
    label kk-KZ official "кіріспе очерк сөздері: …";
    ...
    assert kz.grammar.phon::songy_buyn_undestigi(entity_ref("urn:kz:grammar:clir:lexicon#Qala"), kz.grammar.phon::Zhuan);
}
```

Exemplary: a fact vocabulary as one export unit; the consumer
includes it with the line `facts package::Name`, the neighbour’s rules do
not execute. The facts use the neighbour’s qualified predicates:
a group may assert in another’s vocabulary, but only
facts. `pub facts` total — 40 in 7 packages; group
inclusions — 5 in one.

## Test-surface measurements

### 5. A test fact is almost always complete

In test files — 36 117 facts, almost every one with `id` (25 379) and
`origin` (24 346, of which 24 357 are `case_input` — sic); `valid`,
`observed`, `recorded` never occur there. Conclusion for the
author: times are a property of package facts (dated decisions, windows),
not of test submissions; a test needing `valid` is a smell: the window
belongs in the package, the test only moves the law date (as in
the `research.facts.valid_window` teaching package).

### 6. Negative act facts are rare

`assert not p(…)` — 5 occurrences in 2 corpus packages; in tests — 239.
`valid [a, b)` on a fact — 2 occurrences in 1 package (a dated organ
decision): a `valid` window belongs to package facts, not to test
submissions (test files carry no times — see above).
Rating: not a mistake but a use profile — the act asserts, the case
disputes. A negative package fact (`assert not in_liquidation…` in
the polarity teaching package) is a lawful but rare form: an organ’s recorded
finding must stand behind it, not “nothing heard to the contrary yet”.

## Teaching packages

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

Package `research.facts.polarity` — positive and negative package facts side
by side: `check OK`, `law test` 2/2 (Established with the case fact /
Not established, not refuted with an empty case). Package `research.facts.valid_window` — a fact
with a `valid` window: `check OK`, `law test` 2/2 (inside the window
Established, outside Not established, not refuted on the same fact).