Markdown for LLMs

Source hierarchy

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

# Source hierarchy

## Intention

I want to compare source rank while keeping the rules defeasible.

## Incorrect form and why it stays silent

```law title="Incorrect form"
rule Higher strict { for p: Person; when eligible(p); then not allowed(p); }

// Объявление entity RuleApplication отсутствует.
for high: RuleApplication;
```

Strictness is a logical category. Legal rank is given by data and policy; a missing rank is not a highest or zero rank. RuleApplication is a local type here, not a built-in: it must be declared. Meta-predicates rule_of and conflicts, by contrast, are supplied by the engine; a local twin would shadow them.

## Correct form

```law
language "law.core" version "0.2";
package recipes.b.r07 version "0.1.0";
namespace "urn:recipe:b-rules:07";

entity Person;
relation eligible(p: Person);
relation special(p: Person);
relation allowed(p: Person);
entity RuleApplication;
relation rank(rule_name: Text, level: Integer);
rule Lower defeasible { for p: Person; when eligible(p); then allowed(p); }
rule Higher defeasible { for p: Person; when eligible(p); then not allowed(p); }
priority policy Hierarchy {
    for high: RuleApplication; for low: RuleApplication;
    when conflicts(high, low)
        and exists hr in (collect hr: Text where rule_of(high, hr))
            where exists lr in (collect lr: Text where rule_of(low, lr))
                where exists hk in (collect hk: Integer where rank(hr, hk))
                    where exists lk in (collect lk: Integer where rank(lr, lk)) where hk < lk;
    prefer high over low; reason lex_superior;
}
```

## Frozen execution scene

| Facts and choice | Question | Answer |
|---|---|---|
| higher level | `truth(allowed(entity_ref("urn:recipe:b-rules:07:p")))` | `FALSE_ONLY` / `COMPUTED` |
| rank not given | `truth(allowed(entity_ref("urn:recipe:b-rules:07:p")))` | `BOTH` / `COMPUTED` |
| policy removed | `truth(allowed(entity_ref("urn:recipe:b-rules:07:p")))` | `BOTH` / `COMPUTED` |
| swapped ranks | `truth(allowed(entity_ref("urn:recipe:b-rules:07:p")))` | `TRUE_ONLY` / `COMPUTED` |

```law
test "senior level" {
    given {
        context {
            legal_time @2026-09-13;
            decision_time @2026-09-13T09:00:00+05:00;
            knowledge_time @2026-09-13T09:00:00+05:00;
            timezone "Asia/Almaty";
        }
        assert eligible(entity_ref("urn:recipe:b-rules:07:p")); assert rank(Higher, 1); assert rank(Lower, 2);
    }
    evaluate truth(allowed(entity_ref("urn:recipe:b-rules:07:p")));
    expect truth_status == FALSE_ONLY;
    expect evaluation_status == COMPUTED;
}
```

```law
test "rank not given" {
    given {
        context {
            legal_time @2026-09-13;
            decision_time @2026-09-13T09:00:00+05:00;
            knowledge_time @2026-09-13T09:00:00+05:00;
            timezone "Asia/Almaty";
        }
        assert eligible(entity_ref("urn:recipe:b-rules:07:p"));
    }
    evaluate truth(allowed(entity_ref("urn:recipe:b-rules:07:p")));
    expect truth_status == BOTH;
    expect evaluation_status == COMPUTED;
}
```

```law
test "ranking policy removed" {
    given {
        context {
            legal_time @2026-09-13;
            decision_time @2026-09-13T09:00:00+05:00;
            knowledge_time @2026-09-13T09:00:00+05:00;
            timezone "Asia/Almaty";
        }
        assert eligible(entity_ref("urn:recipe:b-rules:07:p")); assert rank(Higher, 1); assert rank(Lower, 2);
    }
    evaluate truth(allowed(entity_ref("urn:recipe:b-rules:07:p")));
    expect truth_status == BOTH;
    expect evaluation_status == COMPUTED;
}
```

```law
test "ranks swapped" {
    given {
        context {
            legal_time @2026-09-13;
            decision_time @2026-09-13T09:00:00+05:00;
            knowledge_time @2026-09-13T09:00:00+05:00;
            timezone "Asia/Almaty";
        }
        assert eligible(entity_ref("urn:recipe:b-rules:07:p")); assert rank(Higher, 2); assert rank(Lower, 1);
    }
    evaluate truth(allowed(entity_ref("urn:recipe:b-rules:07:p")));
    expect truth_status == TRUE_ONLY;
    expect evaluation_status == COMPUTED;
}
```

## Counterfactual

Mutation: `entity RuleApplication;` → ``; rejection LDC-E2101. Scenes that delete a fragment are marked explicitly in the table.

The same swapped ranks: replace `rule Higher defeasible` with `rule Higher strict`. The result is FALSE_ONLY instead of TRUE_ONLY. This is a rank-model error: strict support cannot be lifted, so this is not an evaluator defect.

This extra scene is executed by the adapter on the mutated program, with both evaluators and a byte comparison:

```law
test "strict higher source refuses" {
    given {
        context {
            legal_time @2026-09-13;
            decision_time @2026-09-13T09:00:00+05:00;
            knowledge_time @2026-09-13T09:00:00+05:00;
            timezone "Asia/Almaty";
        }
        assert eligible(entity_ref("urn:recipe:b-rules:07:p")); assert rank(Higher, 2); assert rank(Lower, 1);
    }
    evaluate truth(allowed(entity_ref("urn:recipe:b-rules:07:p")));
    expect truth_status == FALSE_ONLY;
    expect evaluation_status == COMPUTED;
}
```

```python
>>> import runpy
>>> checks = runpy.run_path("docs/recipes/b-rules/resources/check.py")
>>> checks["check_strength"](https://github.com/arxohq/law/blob/master/docs/recipes/b-rules/7)
'Б7: FALSE_ONLY; lawc = lawref; ошибка выбора силы подтверждена'
```

## Boundary

The rank is instructional: a smaller number means a higher level. Linking a rule to a source in a real package must have provenance. Here rule → rank is set directly for a minimal witness. The extra counterfactual below replaces Higher with strict and shows that rank can no longer override it.