RFR-009 Metadata and Provenance Coverage Measurement

Summary

The research-integrity validator now measures canonical metadata adoption by artifact class and resolves non-empty declared related, related_documents, supersedes, and superseded_by targets by stable identifier or repository path. Coverage and unresolved relationships are diagnostic rather than blocking until an incremental migration policy is approved.

Measured baseline

The authored Markdown scope excludes Git internals, dependencies, generated publisher/build output, archives, intake material, and copied calibration outcomes. Within that declared scope:

• 131 authored Markdown artifacts were measured.
• 48 artifacts have front matter (36.6%).
• 47 artifacts expose a stable id or legacy identifier (35.9%).
• 45 declare status (34.6%) and 44 declare title in front matter (33.8%).
• No artifact currently declares canonical abstract, document_type, evidence_level, canonical, or reading_time_minutes metadata.
• Research has the highest meaningful identifier adoption: 42 of 62 artifacts (67.7%). Content has 0 of 17; documentation has 1 of 25.
• Five non-empty relationships are declared and all five resolve to repository paths. They use the legacy related_documents spelling.
• No active related, supersedes, or superseded_by edge is currently declared; the two documents containing supersession fields use null or empty values.

The machine-readable baseline is data/validation/research-integrity-inventory.json schema version 1.1.

Implementation and calibration

tools/repository/validate_research_integrity.py now:

• excludes dependency and generated-output trees from authored-document checks;
• reports required-field counts and percentages overall and by top-level artifact class;
• accepts identifier as a measured legacy alias for id;
• inventories declared relationship edges;
• resolves relationship targets by stable ID or repository-relative path; and
• reports unresolved targets without failing the repository audit.

Seven tests pass. The three added probes verify per-class coverage measurement, ID/path/unresolved relationship classification, and exclusion of generated and dependency trees. Existing duplicate-ID, broken-link, invalid-frontier-edge, and passing controls remain green.

Interpretation

The repository has a reliable structural validator but not a metadata-complete knowledge graph substrate. Enforcing the full metadata standard now would reject most authored artifacts and create a broad migration disconnected from measured retrieval value. The correct next step is a bounded adoption policy for current canonical research and content, followed by supersession-edge calibration when real edges exist.

Remaining uncertainty

• Which artifact classes should be required to adopt the complete canonical schema first?
• Does stable metadata improve real handoff/retrieval performance enough to justify migration cost?
• Should related_documents remain a supported alias or be migrated to canonical related?
• How should path relationships behave across renames compared with stable-ID relationships?
• Heading fragments, prose paths, evidence/hypothesis references, and reciprocal supersession consistency remain outside the enforced scope.

Decision

RFR-009 remains in Validation. Measurement and declared-relationship diagnostics are complete, but metadata adoption and real supersession coverage are too low to claim a complete provenance substrate.