Markdown for LLMs

PROLEG

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

# PROLEG

**In short:** PROLEG decides the unproven fact against whoever bore the
burden; Arxo keeps it undecided and says what is missing. This page — one
of three separate Prolog-family studies alongside Logical English and
Blawx — is for readers who reason about judicial proof and want burden,
negation, goal order, and completion treated separately.

## What PROLEG is for

PROLEG implements Japan's presumed-ultimate-fact theory in Prolog: the
burden of persuasion is distributed per condition in advance, exceptions
nest with side-switching, unclaimed defenses are ignored, and the fact
base mirrors procedure — allegations, evidence, plausibility inputs,
admissions. Its strengths are the dispute-shaped trace (who failed to prove
what), proven equivalence with answer-set semantics via published
translations, coverage beyond civil law, and a one-file engine over a
standard Prolog.

## Where it meets Arxo

Burden of proof, presumptions, and required facts are the shared ground:
how each system treats the unproven fact. The studied scenario uses
published programs and recorded traces — lease termination, an
offer-and-capacity chain, a minor buyer's rescission with duress — as
recorded expectations for checking, not as engine-run reports. Running the
third-party engine copy is a deferred, separate step.

## Key differences

- **Two-valued by construction.** A burden-side unproven fact counts as
  false and reasoning proceeds. Arxo holds four-valued support: unknown,
  closed-world absence, missing evidence, and explicit false each need
  their own producer. The collapse is the design, not a bug — but it must
  never be quoted as "proven negation".
- **Burden is pre-distributed.** Each condition knows its side, and nested
  exceptions switch sides down the chain. Arxo composes unless-clauses,
  priorities, presumptions, and a judgment channel instead. Both answer
  "who loses on silence" — with different machinery.
- **Goal order is the interpreter.** Each body proves by its side, then
  each exception by the opposite side; an unproven exception means the
  defense failed. Arxo instead refuses via requires-judgment where the
  norm is open. Order of goals is load-bearing in both — and different.
- **Completion counts extensions.** Conflicting same-conclusion rules can
  yield zero or two extensions; Arxo shows the visible conflict without
  forcing a sign. Count, don't force.
- **Answer selection differs in kind.** Sign plus dispute trace versus an
  evaluation document with a proof graph; the modular variant's explicit
  negation is a three-rule construction with groundness conditions,
  mapping to unless-plus-negative-producer rather than to a single
  operator.

## A concrete scenario

Twelve published cases with traced sources plus three edits: a statute
reform that left one article stable and rewrote its neighbor, a classic
versus modular engine comparison, and byte-identical engine revisions with
a dataset-composition change. The first experiment checks records against
the model without running a foreign engine; dates in the material have no
established semantics, so date-core cases stay non-comparable.

## Choosing and combining

Choose PROLEG when the reasoning is judicial procedure itself —
pre-distributed burdens, exception chains, dispute traces — on a small
install. Look to Arxo when the unproven must stay visibly undecided, when
evaluative terms need named organs, or when editions and provenance carry
the review. Slide figures about rule counts and exam coverage stay quoted
as claimed until their report confirms them; plausibility inputs are
judge decisions, not inferred conclusions.

## Evidence and open questions

- Sources checked: September 2026 (papers with byte sizes and hashes, the
  modular-variant paper, the third-party engine copy — explicitly
  third-party, never an official release — all re-verified).
- Studied profile: the papers plus the third-party engine copy; no
  versioned release, repository, or license was found — a negative search
  result, recorded as such.
- Basis: confirmed by documentation plus a prepared protocol; comparative
  run not performed.
- Open: engine execution, date semantics, and the run itself.

## Sources and reproducible materials

- PROLEG papers (Satoh et al.):
  [jurisin2010-ksatoh.pdf](http://research.nii.ac.jp/~ksatoh/juris-informatics-papers/jurisin2010-ksatoh.pdf)
- Modular variant (CEUR-3257):
  [paper1.pdf](http://ceurspt.wikidata.dbis.rwth-aachen.de/Vol-3257/paper1.pdf)
- Third-party engine copy (attribution: not an official release):
  [github.com/liviorobaldo/compliancecheckers](https://github.com/liviorobaldo/compliancecheckers)