Cross-Reference Lineage

M6 asks whether dictionary-internal cross-reference graphs preserve lineage signals. The answer is deliberately not binary: AP and AP90 behave like the same dictionary across editions, while MW and PWG share a real cross-reference core without wholesale graph inheritance.

Trust Block

• Evidence: data/lexico/xref_lineage.json, data/lexico/xref_shared_edges.csv, data/lexico/xref_edges.csv, data/lexico/xref_hub_review.json, scripts/lexico/m3_xrefs.py, scripts/lexico/m6_xref_lineage.py, scripts/build-xref-lineage.mjs, and scripts/build-xref-hub-review.mjs.
• Limitations: overlap is a floor; messy targets fail to match, and a shared edge is a lineage signal rather than proof of direct copying by itself.
• Validation: generated by npm run build-xref-lineage and npm run build-xref-hub-review; source data by python scripts/lexico/validate_lexico.py; page by npm run build.
• Owner repo: csl-atlas.
• Next use: treat the chart as structural evidence, then use the companion hub review labels before making a lineage claim.

Shared Cross-Reference Edges

Inheritance-Rate Ceiling

AP x AP90 is the positive control: the same Apte dictionary across editions keeps about 85% of its comparable cross-reference edges. MW x PWG is much lower on both sides, so the finding is a shared core, not wholesale inheritance.

Pair Rows

MW-PWG Shared Edge Sample

How To Read It

• The comparison is restricted to lemmas both dictionaries cross-reference.
• A high AP/AP90 rate is a positive control for edition-level continuity.
• MW/PWG overlap is real but partial; most cross-reference expansion remains independent.
• Tiny shared-source pairs are shown in the table but should not drive lineage claims.

Generated by npm run build-xref-lineage. See docs/MICROSTRUCTURE_XREF_LINEAGE.md. CC-BY-SA-4.0.