Consolidate consistency checks into the lint rule system

[JohnRDOrazio]

Summary

The consistency check (consistency_service.py) and the lint system (linter.py) have significant overlap. The lint system is more mature (progressive levels, per-project config, persistence in PostgreSQL, entity-type-aware results), while the consistency check is a lightweight all-or-nothing runner with ephemeral Redis caching.

Prerequisite: This work should not begin until PR #94 (per-project lint config) is merged, as it builds on the lint infrastructure introduced there.

Related discussion: #87 — Consolidate duplicate detection across lint, consistency, and duplicates tab

Current State

Overlap (3 fully redundant checks)

Consistency Rule	Equivalent Lint Rule	Action
cycle-detect	circular-hierarchy	Remove from consistency
missing-label	missing-label	Remove from consistency
missing-comment	missing-comment	Remove from consistency

Partial overlap (3 checks with different behavior)

Consistency Rule	Related Lint Rule	Difference
orphan-class	orphan-class	Consistency also checks for zero instances (stricter)
duplicate-label	duplicate-label	Consistency is exact-match + same-type only; lint is case-insensitive + all types
dangling-ref	undefined-parent	Consistency checks subClassOf + domain + range; lint only checks subClassOf

Unique to consistency (6 rules to migrate)

Rule	Description	Suggested Lint Level
unused-property	Property declared but never used as predicate	Level 4 (Quality)
orphan-individual	Individual's rdf:type not declared as owl:Class	Level 2 (Consistency)
empty-domain	ObjectProperty/DatatypeProperty has no rdfs:domain	Level 4 (Quality)
empty-range	ObjectProperty/DatatypeProperty has no rdfs:range	Level 4 (Quality)
deprecated-parent	Class subclasses an owl:deprecated class	Level 2 (Consistency)
multi-root	More than 5 root classes (design smell)	Level 4 (Quality)

Proposed Plan

Phase 1: Migrate unique rules to lint

• Add the 6 unique consistency rules as new lint rules
• Assign appropriate levels and scopes
• Include subject_type classification

Phase 2: Enhance partial-overlap rules

• Update lint orphan-class to optionally check for zero instances
• Expand lint undefined-parent → dangling-ref to also check domain/range references
• Decide on duplicate-label behavior (case-sensitivity, same-type constraint) — see discussion Consolidate duplicate detection across lint, consistency, and duplicates tab #87

Phase 3: Deprecate consistency service

• Remove the 3 fully redundant checks from consistency
• Once all rules are migrated, deprecate consistency_service.py
• Update the frontend Consistency tab to use lint results filtered by rule category
• Eventually remove the separate consistency endpoint

Benefits

• Single system for all ontology quality checks
• Per-project configuration applies to all checks
• Progressive levels include consistency checks
• Persistent results (PostgreSQL vs ephemeral Redis)
• Entity-type-aware navigation for all issues
• Simpler codebase to maintain

🤖 Generated with Claude Code

[JohnRDOrazio]

@damienriehl — kicking off scoping for this. The plan in the body splits cleanly into two PRs:

1. PR1 — add the 6 unique consistency rules to linter.py, reconcile semantics for the 3 partial-overlap rules, leave consistency_service.py in place so we can canary-validate that lint produces the same findings the old service does.
2. PR2 — remove consistency_service.py, drop the /quality/consistency-check routes + worker task, and remove the frontend Consistency tab entirely so every quality check lives under the Lint tab.

The Phase 2 reconciliation has three rules whose lint and consistency variants disagree. I'd like your input on which direction to take for two of them — duplicate-label we've already settled.

Rule	Lint today	Consistency today
orphan-class	no parent + no children	no parent + no children + no instances
undefined-parent ↔ dangling-ref	only checks subClassOf targets	also checks rdfs:domain, rdfs:range
duplicate-label	case-insensitive, classes only	exact match, same-type, all entity types

Settled — duplicate-label: keep case-insensitive matching, expand scope to all entity types with same-type constraint (the consistency rule's per-type behavior).

Open for your input:

1. orphan-class — adopt the stricter "no instances" check from consistency, or keep the current looser lint definition? My lean is to keep the current lint rule (no instance check). "Instances" here means individuals declared with rdf:type Class, and a class without individuals isn't necessarily orphan-worthy: TBox-only ontologies, abstract classes, and taxonomies that have been declared but not yet populated all legitimately have classes with no individuals. Flagging those would be a false-positive factory.

2. undefined-parent ↔ dangling-ref — expand the lint rule to cover rdfs:domain and rdfs:range references and rename it to dangling-ref, or keep undefined-parent and add a separate rule for property-axis dangling references? My lean is the rename + expand: the underlying problem ("you reference a URI that isn't defined") is the same regardless of which predicate flagged it, so one rule with a predicate detail beats two near-duplicates.

Once we settle these I'll write the design doc and link it back here. The wider rdflib-vs-SQL question raised in discussion #87 is orthogonal and I'm leaving it out of #99's scope.