[AImenes]

The 1.x code was a loose folder meant to be copied into site-packages and had several correctness problems: the entry point crashed on a missing import, the default upperlimit=None raised TypeError, role inclusions were skipped when the first argument was unbound, inverse_of was translated to P⁻ ⊑ P, "?_" written twice became a shared variable, constants broke unification, and duplicate rewritings were only detected for identical atom order.

Version 2 is a pip-installable perfectref package:

• perfectref.model: immutable terms, atoms and conjunctive queries kept in the paper's τ normal form, with canonical keys for deduplication
• perfectref.tbox: DL-Lite positive inclusions indexed by right-hand side, plus a plain-text DL syntax parser
• perfectref.owl: owlready2 extraction covering subclass/equivalence, some/min/exactly/value restrictions, intersections, unions on the left, general class axioms, domain/range, sub/inverse/equivalent/symmetric properties and imports; axioms outside DL-Lite are listed in tbox.skipped
• perfectref.algorithm: work-list PerfectRef with gr, mgu and reduce as in the paper, head specialisation on unification, and a max_rewritings cap
• perfectref.parser / formatting: Datalog-style query grammar with long variable names, constants and IRIs; text, JSON and SPARQL output
• perfectref.legacy: the 1.x API (get_entailed_queries, Query, QueryBody, AtomConcept, AtomRole, Variable, ...) kept working on top of the core, with cached ontology loading and picklable results
• CLI (perfectref rewrite|tbox), pyproject, GitHub Actions, ruff config
• tests: Example 1 of the paper reproduced exactly, OWL translation, parser, legacy API and CLI (55 tests)

Test ontologies move to tests/data (plus the paper's university TBox as OWL) and the Pizza ontology to examples/; redundant copies, generated output and committed bytecode are removed.