Architecture overview

This section is for the curious and for contributors: how owlcompare actually turns two files into a diff. This overview page will sketch the whole pipeline at a glance and link to the deeper per-topic pages. For the user-facing version of these ideas, see Understanding the output.

This page is being expanded

The outline below is in place; the full narrative and pipeline diagram are coming.

What this page will cover

• The pipeline, end to end — load → canonicalize → Layer 0 diff → Layer 1 slices → rename detection → severity refinement → render.
• The four-layer model — what each layer answers and why they're separated.
• Canonicalization — why both inputs are normalized before anything is compared (see Canonicalization).
• The internal model — frozen dataclasses (OntologySnapshot, Entity, Change) rather than raw rdflib objects, and why.
• The orchestrator — how the slices are sequenced and how subsumption keeps the raw triple noise out of the final report.
• Severity post-processing — the cross-cutting refinement pass that runs last.
• The component breakdown — how the codebase maps onto the spec-driven components.