Heritage witness — an independent, non-Cologne confirmation

The Sanskrit Heritage Platform (INRIA) lexicon is an independent witness, entirely outside the Cologne Digital Sanskrit Dictionaries pipeline. The MW↔Heritage crosswalk (built from the mirror's own MW↔DICO alignment — no OCR or fuzzy matching) makes it joinable onto MW's headword set: this page shows which MW headwords Heritage independently confirms, useful evidence for the P-series papers' attestation arguments.

Of distinct MW headwords, Heritage covers (%) — resolved to an exact DICO entry anchor, covered but unresolved (MW's plain key drops DICO's homonym suffix).

Coverage by initial letter

Where Heritage coverage concentrates across the MW alphabet — each bar is one initial (SLP1), split anchored vs covered-without-anchor, against the total MW headword count for that initial.

Heritage-covered headwords

Every MW headword Heritage covers, with its match tier and a link into the source viewer at the first MW occurrence (headwords shown in IAST).

Chart Trust Block

• Claim: which MW headwords the Sanskrit Heritage Platform's lexicon independently lists, and at what match confidence (anchored vs covered-without-a-resolved-anchor).
• Evidence label: derived — deterministic key join on normalized SLP1; no classification or model.
• Source files: SanskritLexicography/HeadwordLists/mw_heritage_crosswalk.tsv ( rows, built read-only from the Heritage mirror's own MW↔DICO alignment by heritage_mw_crosswalk.py, owned by SanskritLexicography) joined onto MW's <k1> headwords from csl-orig/v02/mw/mw.txt; committed atlas packet src/data/heritage/heritage_witness.json with crosswalk-commit provenance in its .source.json sidecar.
• Generated by: npm run build-heritage-witness
• Validation: npm run validate-heritage-witness — totals arithmetic, per-initial sums, witnessed-row tier consistency, plus a full rebuild cross-check against the sibling crosswalk + live MW dict when the checkout is present.
• Known false positives: none at the join level — entry existence is read directly off the committed crosswalk's covered_flag/anchor columns, not inferred.
• Known false negatives: entry-existence only, not sense-level agreement — a match confirms Heritage lists the headword, not that the two dictionaries agree on meaning; MW records differing only by homonym suffix or accent collapse onto one normalized row, so a homonym-specific Heritage mismatch is invisible here; Heritage coverage reflects the mirror's snapshot, not a live INRIA query.
• Review status: machine-reviewed (deterministic validator; 20 rows of the underlying crosswalk hand-verified against the mirror's DICO/*.html and MW's raw HTML per the crosswalk's own sanity check).
• Owner repo: csl-atlas (rendering + join); the crosswalk itself is owned by SanskritLexicography.
• Next action: the parallel kosha ingest (H345) joins the same crosswalk onto kosha's frequency layer as a third coverage witness — once merged, a follow-on could add a "corpus-attested AND Heritage-covered" combined axis here.
• External dependencies: SanskritLexicography (crosswalk), csl-orig v02 (MW headwords).
• Boundary note: the atlas joins and renders; the DICO alignment itself, its homonym-fallback logic, and any Heritage-side corrections stay with SanskritLexicography.