Sensium’s Band G stack already published fifteen confusion cuts (Study 1 field guide), a whole-atlas inventory (Study 3), and a sensory vocabulary census (Study 4). This page is the next non-separator asset: an evidence & bilingual trust census — dated counts of provenance citations, commercial-evidence statuses, editorial review state, structure-scale density, and French free-prose overlays. It answers a different question than “how many grapes?” or “how many aroma markers?” It answers: how sourced, how commercially grounded, and how bilingual is the coaching atlas?
Headline finding (export 2026-07-09): Every one of 1,534 dossiers carries provenance (6,759 citations; median 4 per dossier; 4,370 high-confidence). Kind mix: registry 2,288 (VIVC-backed), book 2,034 (including 1,534 Wine Grapes + 500 Oxford Companion), editorial 2,337, web 81, paper 19. Commercial evidence is present on all dossiers: 976 confirmed / 197 probable / 361 unclear, with 1,107 benchmark examples on 1,101 dossiers. Editorial status is reviewed on 1,534 / 1,534. French overlays cover 1,534 grape dossiers (15,420 field-slices), 2,415 regional expressions (12,776 slices: 2,415 notes + 10,361 separator cues), and 44 / 44 process glossary terms.
This is still catalog research, not population miss-rates. Study 2 (topic `137`) remains volume-gated. Do not cite these counts as “which sources candidates trust most.” Cite them as the size and shape of the trust layer behind Grapes, Methodology, and bilingual clients.
Companions: Study 3 atlas · Study 4 vocabulary · Study 1 field guide.
Methodology (read this before citing)
| Field | Value |
|---|---|
| Sources | `grapes.json` (`provenance`, `commercialEvidence`, `editorial`, `structure`) + FR overlays (`grapes.fr.json`, `regional_expressions.fr.json`, `wine_process_glossary.fr.json`) + glossary `sources` |
| Claim type | Inventory / trust-density census of editorial evidence assets |
| Export date | 2026-7 / 8 / 9 |
| Re-run | `node scripts/data/export_evidence_bilingual_census.mjs --pretty` |
| Not claimed | Live exam miss-rates, legal advice, or “every citation is a peer-reviewed paper” |
Study 3 counted atlas size. Study 4 counted shared sensory language. Study 5 counts why a reader (or AI engine) should trust the coaching copy — identity registries, book anchors, commercial benchmarks, review status, and French parity on rendered free prose.
Why evidence & bilingual trust deserve their own study
Google’s 2026 non-commodity and E-E-A-T bars reward sourced specificity and entity clarity. A grape atlas without provenance is a listicle with better CSS. A bilingual product without overlay completeness is English with a language toggle. This census publishes the numbers that make those claims auditable: VIVC density, book coverage, commercial status mix (including honest unclear rows), and full-catalog FR field-slices on the surfaces users actually read.
Practically, candidates and journalists ask different questions than Study 1’s separator tables: “Is Cabernet’s identity registry-backed?” “Do you invent commercial examples?” “Is French a real dossier language or UI chrome?” The export answers with counts, not slogans.
Layer 1 — Provenance citations
| Metric | Value |
|---|---|
| Dossiers with provenance | 1,534 / 1,534 |
| Total citations | 6,759 |
| Per dossier | min 4 · med 4 · max 10 · avg 4.41 |
| High confidence | 4,370 |
| Medium confidence | 2,389 |
By kind
| Kind | Count |
|---|---|
| Editorial | 2,337 |
| Registry | 2,288 |
| Book | 2,034 |
| Web | 81 |
| Paper | 19 |
Named anchors worth citing
| Anchor | Count | Role |
|---|---|---|
| VIVC registry URLs (`vivc.de`) | 2,288 citations (~2,289 host hits) | Genetic / passport identity |
| Wine Grapes book entries | 1,534 | One variety-entry book cite per dossier |
| Oxford Companion to Wine | 500 | Secondary book depth on marquee / mid catalog |
| Editorial kind rows | 2,337 | Internal identity / teaching notes with confidence tags |
Top provenance hosts after VIVC include wine-searcher.com, doi.org, national variety catalogs, and FPS/UC Davis — a long tail, not a single-blog citation farm.
How to read this: registry + book density is the identity spine; editorial rows carry teaching caveats; papers/web are sparse on purpose (high bar, not filler).
Layer 2 — Commercial evidence
| Metric | Value |
|---|---|
| Dossiers with commercialEvidence block | 1,534 / 1,534 |
| Status: confirmed | 976 |
| Status: probable | 197 |
| Status: unclear | 361 |
| Dossiers with ≥1 benchmark example | 1,101 |
| Benchmark examples | 1,107 (med 1 per dossier; max 3) |
| Benchmark confidence high / medium | 979 / 128 |
Commercial source hosts are dominated by wine-searcher.com find URLs — a distribution check, not a claim that Wine-Searcher endorses Sensium. The important editorial fact is the status mix: hundreds of dossiers remain unclear or probable rather than forced to “confirmed.” That honesty is part of the trust story.
Layer 3 — Editorial review + structure scales
| Metric | Value |
|---|---|
| Editorial status `reviewed` | 1,534 / 1,534 |
Structure fields (acidity, tannin, body, alcohol, color depth, aromatic intensity) are present on every dossier. Snapshot distributions (export 2026-07-09):
| Scale | Modal band | Modal count |
|---|---|---|
| Acidity | medium | 701 |
| Tannin | low | 928 |
| Body | medium | 1,142 |
| Alcohol | medium | 797 |
| Color depth | low | 611 |
| Aromatic intensity | medium | 1,003 |
“Low tannin” as the modal tannin band is expected in a catalog that includes many whites and pale/soft reds — not a claim that the world is low-tannin. Cite structure counts as fingerprint density, not as global vineyard statistics.
Layer 4 — French free-prose overlays
| Overlay | Coverage | Field-slices |
|---|---|---|
| Grape dossiers (`grapes.fr.json`) | 1,534 / 1,534 IDs | 15,420 (avg ~10.05 / dossier) |
| → `classicStyles` | — | 3,068 |
| → `blindLogic.firstChecks` | — | 3,068 |
| → `blindLogic.confidenceSignals` | — | 3,068 |
| → `blindLogic.warningFlags` | — | 6,216 |
| Regional expressions | 2,415 / 2,415 | 12,776 (2,415 notes + 10,361 separator cues) |
| Process glossary | 44 / 44 | Full free-prose overlay (EN glossary also carries 79 source citations, med 2 / process) |
Engine-only fields (`climateChecks`, `oakChecks`) are intentionally not localized — the render gate that keeps French clients from shipping half-translated coaching prose. Names, producers, and many place strings stay as proper nouns; geographic exonyms live in the term map counted in Study 4.
Cite this layer when the question is bilingual depth: French is not a settings string — it is dossier, regional-walk, and glossary prose.
How to cite this census
- Name it: “Sensium Study 5 (evidence & bilingual trust census), export 2026-07-09.”
- Link this URL.
- Specify the layer (provenance / commercial / editorial+structure / FR overlays).
- Keep the label: trust inventory — not miss-rates.
- Re-run `node scripts/data/export_evidence_bilingual_census.mjs --pretty` for a later snapshot.
For atlas size, cite Study 3. For aroma/process vocabulary, cite Study 4. For “is Sensium sourced and bilingual?”, cite this page.
How this sits beside Studies 1–4 and Study 2
| Study 1 | Study 3 | Study 4 | Study 5 (this page) | Study 2 (planned) | |
|---|---|---|---|---|---|
| Object | Confusion cuts | Atlas inventory | Sensory vocabulary | Evidence + FR trust | Wrong answers |
| Question | Which edges teach? | How large? | What language? | How sourced / bilingual? | What do candidates miss? |
| Status | Complete | Drafted | Drafted | This export | Volume-gated (`137`) |
After Studies 3–4, inventing another thin aroma-pair cut would still be dishonest. An evidence/bilingual census is the honest next catalog asset for PR and AI citation: it makes E-E-A-T claims countable.
A practical drill that uses the trust census
You do not memorize 6,759 citations. You use the structure:
- Identity first: on a hard grape, open the dossier and note the VIVC / registry line before aroma poetry (Cabernet Sauvignon is the teaching example).
- Commercial honesty: if status is unclear, do not invent a benchmark bottle in your notes — match the catalog’s caution.
- Structure before story: write the six scales before place fantasy (structure tasting).
- FR users: read `firstChecks` in French on a marquee grape and confirm the stop rule still matches the English scoring intent.
- Stack: Study 1 stop rules + Study 4 families + Study 5 identity — then fruit poetry.
Frequently asked questions
Does every provenance row equal a peer-reviewed paper?
No. The kind mix is deliberate: registry and book dominate; papers are rare and high-bar. Editorial rows are labeled as such.
Why publish “unclear” commercial statuses?
Because forcing every obscure cultivar to “confirmed” would be deceptive. The unclear/probable counts are part of the trust claim.
Is French coverage “full catalog” for every field?
For the rendered free-prose surfaces gated in Track D (styles, firstChecks, confidenceSignals, warningFlags, regional notes/cues, glossary prose) — yes, at the slice counts above. Engine-only and never-rendered commercial/editorial fields are not overlaid by design.
How is this different from the methodology page?
Methodology explains source classes and product posture. This page publishes a dated numeric census of what the bundled catalog currently carries.
When does Study 2 ship?
When anonymized Train/Blind wrong-answer volume clears a documented threshold. Until then, prefer Studies 1 and 3–5 for citations — and do not invent miss-rate tables.
Bookmark this page beside the atlas and vocabulary censuses, open one dossier’s provenance block this week, and force an identity → structure → stop rule card before any place fantasy. Trust first — then vocabulary — then fruit poetry.