Markdown for LLMs

LegalRuleML

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

# LegalRuleML

**In short:** LegalRuleML is a standard for exchanging norms, not a program
that runs them — nothing executes a LegalRuleML document by itself. This
page compares representation and transfer: what survives a move between
Arxo and the standard, what needs extensions, and what is lost. It is for
readers who interchange normative models between tools.

## What LegalRuleML is for

The OASIS LegalRuleML Core standard encodes norms in XML for interchange:
defeasible rules with strict, defeasible, and defeater strengths plus
override relations, deontic modalities with violation and reparation
structure, temporal axes, and rich metadata — authority, jurisdiction,
sources, references, and alternative readings. Its strengths are
engine independence (one model, many reasoners), two provably equivalent
serializations with schema validation, and source isomorphism that keeps
the link from rule to article.

## Where it meets Arxo

The comparable object is the markup, never runtime speed. Two directions
are studied: writing a small Arxo deontic model out to the compact
serialization and back, and reading published standard examples into Arxo.
The chosen inference mechanism for any future answer comparison is named
explicitly — a defeasible reasoner reached through a published transform —
and it is currently blocked, not used. No answer equality is claimed
without it.

## Key differences

- **Three checks, not one.** Structural preservation (schema validity),
  semantic preservation (a construct-by-construct mapping table with manual
  review), and answer equality (needs the external reasoner) are distinct
  claims. A transfer can pass the first while the other two stay open.
- **Round-trip by its own importer proves little.** Reading back what you
  wrote shows structural reversibility at most. It says nothing about
  extensions, about recovery from service data, or about preservation of
  inference results — the page keeps those four claims separate.
- **Defeasibility maps architecturally, not provably.** Override and rule
  strength correspond to Arxo's priorities and defeater behavior on paper,
  but conclusion equality with the external reasoner is unconfirmed until
  that reasoner runs.
- **Negation needs a pre-run decision.** The standard's two negative forms
  meet Arxo's explicit negation on one side and absence handling on the
  other; unresolved inputs get an explicit unresolved annotation, never a
  guess. Element order and comments are format freedom, not defects.
- **Numbers, proofs, and tests live outside the format.** Arithmetic,
  explanations, and test shapes are Arxo-side concerns. Comparing them
  against the standard is out of purpose.

## A concrete scenario

The prepared experiment round-trips a minimal deontic model — Arxo to
compact LegalRuleML to Arxo — classifying every construct as preserved,
extended, or lost, crossed with encoded, approximated, or residue. Twelve
cases come from the standard's own examples; four edits cover a
specification zero-delta regression, comment and order moves, an
override-as-qualification open reading, and temporal-cell swaps. Open
readings stay open with both variants on record.

## Choosing and combining

Choose LegalRuleML when norms must travel between tools, when source
metadata and alternative readings ride with the rules, or when a standard
serialization is the deliverable. Look to Arxo when the model must execute
with dated editions, exact arithmetic, and replayable derivations. Combined,
the standard carries the model across tool boundaries while Arxo executes
and explains it — with the mapping table, not the schema alone, as the
trust anchor.

## Evidence and open questions

- Sources checked: September 2026 (the standard full text, both
  serializations of all published examples with matching hashes,
  schemas, transforms, the reasoner-transform paper, and a live
  integration rig).
- Studied profile: OASIS LegalRuleML Core 1.0, compact serialization
  through the stock normalizer; example files are pin-only under a limited
  license and were not copied.
- Basis: confirmed by documentation plus a prepared protocol; comparative
  run not performed, no transfer rig built. The Arxo mini-model passes its
  static check; its scenarios are written, not executed.
- Open: answer equality pending the external reasoner (unreachable at
  study time), plus four open construct readings listed above.

## Sources and reproducible materials

- The standard:
  [OASIS LegalRuleML technical committee](https://www.oasis-open.org/committees/tc_home.php?wg_abbrev=legalruleml)
- Reasoner transform paper (Lam, Hashmi):
  [arxiv.org/abs/1711.06128](https://arxiv.org/abs/1711.06128)