Sensium’s Band G research is not one ranking page. It is a stack of fifteen catalog cuts — Studies 1a through 1o — each answering a different question about the same `confusionPairs` graph. This field guide is the citation hub: what each cut is, what it is not, how to drill them in order, and how to cite them without inventing population miss-rates.
Headline: Studies 1a–1o are complete as catalog-graph research. Study 2 (topic `137` — anonymized Train/Blind miss-rates) remains volume-gated. Until Study 2 ships, do not call any Study 1 table “most-missed pairs.” Call them what they are: reproducible catalog separators, hubs, magnets, and stop rules.
Companions already live: year-in-review pairs hub · 1a graph · 1o dark-fruit. This page is the index and method, not a sixteenth aroma cut.
What Study 1 is (and is not)
| Study 1 (1a / 1b / 1c / 1d / 1e / 1f / 1g / 1h / 1i / 1j / 1k / 1l / 1m / 1n / 1o) | Study 2 (planned) | |
|---|---|---|
| Source | Bundled catalog `confusionPairs` + planting ranks | Anonymized Train/Blind wrong answers |
| Claim type | What the editorial graph already encodes | What candidates miss in aggregate |
| Reproducible? | Yes — `node scripts/data/export_*.mjs` | When n clears a documented threshold |
| Cite as | “Sensium catalog Study 1x, 2026-07-09” | Not yet — do not invent tables |
Google’s 2026 non-commodity bar still applies: dated method, re-run commands, clear “not telemetry” labels on every spoke.
The 15 cuts at a glance
| Study | Cut | Primary question | Spoke | Export |
|---|---|---|---|---|
| **1a** | Mutual × planting | Which pairs does the product elevate globally? | Confusion graph | `export_confusion_graph_ranking.mjs` |
| **1b** | Top-15 neighborhoods | What is the local four-neighbor map? | Neighborhoods | `export_confusion_neighborhoods.mjs` |
| **1c** | One-way exam hubs | Which syllabus grapes does mutual rank under-weight? | One-way hubs | `export_one_way_exam_hubs.mjs` |
| **1d** | Identity traps | Which edges stop at name / parent / synonym? | Identity traps | `export_identity_trap_pairs.mjs` |
| **1e** | Planting-asymmetric | Which mutuals pair a top-20 hub with a mid-catalog twin? | Asymmetric pairs | `export_planting_asymmetric_pairs.mjs` |
| **1f** | Confusion magnets | Which grapes have the highest inbound degree? | Magnets | `export_confusion_magnets.mjs` |
| **1g** | Color-frame | Which related names must stop at white / red / rosé? | Color-frame | `export_color_frame_pairs.mjs` |
| **1h** | Oak-led | Which edges are decided by oak stance? | Oak-led | `export_oak_led_pairs.mjs` |
| **1i** | Dual-structure | Which edges name tannin and acid together? | Dual-structure | `export_dual_structure_pairs.mjs` |
| **1j** | Pepper-smoke | Black pepper / olive / smoky savoriness? | Pepper-smoke | `export_pepper_smoke_pairs.mjs` |
| **1k** | Citrus-led | Green-citrus / saline citrus / citrus-herbal? | Citrus-led | `export_citrus_led_pairs.mjs` |
| **1l** | Floral-perfume | Rose / lychee / violet / perfume / rose-tar? | Floral-perfume | `export_floral_perfume_pairs.mjs` |
| **1m** | Savory-led | Gamey / red-pepper savory / savory herbs? | Savory-led | `export_savory_led_pairs.mjs` |
| **1n** | Textural-white | Wax / lanolin / pear-fennel / pear-almond? | Textural-white | `export_textural_white_pairs.mjs` |
| **1o** | Dark-fruit | Cassis / plum / inky / leafy / black-tea / brambly? | Dark-fruit | `export_dark_fruit_pairs.mjs` |
Teaching tables on the spokes are curated slices of each export (often ~10–15 rows). Re-run the script for the full filtered list.
Three layers of the stack
Layer A — Graph shape (1a–1f)
Start here if you need where the product points. Mutual planting weight (1a), local neighborhoods (1b), one-way syllabus hubs (1c), identity traps (1d), famous×mid-catalog mutuals (1e), and inbound magnets (1f) answer structure of the graph, not yet wording of the separator.
Use Layer A when building a syllabus week, a Compare shortlist, or a press citation about “what Sensium elevates.”
Layer B — Stop-rule families (1g–1i)
Color-frame (1g), oak-led (1h), and dual-structure (1i) are decision-type cuts: the first separator is a frame, an oak stance, or a two-pillar structure fork. Drill these before free-floating aroma poetry.
Layer C — Aroma / texture vocabulary (1j–1o)
Pepper-smoke, citrus-led, floral-perfume, savory-led, textural-white, and dark-fruit are vocabulary cuts. Each excludes the prior families so the teaching lists stay disjoint. After 1o, residual first-separator aroma cuts under the same exclusion stack are thin — that is why this field guide exists instead of a sixteenth thin aroma post.
How to cite Study 1 (classroom / press / AI)
- Name the cut: “Sensium Study 1j (pepper-smoke pairs), catalog export 2026-07-09.”
- Link the spoke URL (table above).
- Keep the label: catalog-graph research, not population miss-rates.
- Point to the re-run command on the spoke (or this page’s export column).
- Do not merge cuts: Syrah pepper-smoke (1j) ≠ Carmenère red-pepper savory (1m) ≠ Cabernet cassis (1o).
For a single link-magnet checklist of pairs (not research cuts), use the year-in-review hub.
Four-week drill stack (exam-first)
| Week | Focus | Cuts | Habit |
|---|---|---|---|
| 1 | Graph + identity | 1a, 1b, 1c, 1d | Open Compare for top mutuals; write identity stop rules before aroma |
| 2 | Structure stop rules | 1e, 1f, 1g, 1h, 1i | Magnets + oak + dual-structure; force tannin/acid/oak lines |
| 3 | Red aroma lanes | 1j, 1m, 1o | Pepper-smoke ≠ savory ≠ dark-fruit — three sheets |
| 4 | White + perfume | 1k, 1l, 1n | Citrus ≠ perfume ≠ wax/pear texture |
Service note: attach each week to real glasses (home flights; exam shopping). Structure after the stop rule (structure tasting).
A useful exam habit: keep a twelve-question stop-rule card built from 1d/1g–1o — identity? color-frame? oak? dual-structure? pepper-smoke? citrus? floral? savory? textural-white? dark-fruit? (plus planting-asymmetric and magnet checks from 1e/1f when the pair is famous×obscure). Candidates who skip the card invent generic spice and place fantasy the catalog already forbids.
Magnet grapes worth memorizing
Across Layer C, a few cultivars appear as contrast poles because the catalog uses their lines repeatedly:
| Magnet | Study | Stance to write first |
|---|---|---|
| Syrah | 1j | Black pepper / olive / smoky savoriness |
| Sauvignon Blanc | 1k | Pungent green-citrus |
| Gewürztraminer / Cinsault / Nebbiolo | 1l | Lychee-rose / perfumed red-fruit / rose-tar |
| Carmenère / Mourvèdre | 1m | Red-pepper savory / gamey savory |
| Chenin Blanc / Sémillon / Macabeo | 1n | Wax / lanolin / apple-fennel |
| Cabernet Sauvignon | 1o | Cassis + firmer tannin + cedar length |
Drill the other grape first, then confirm on Compare.
Shared methodology (all spokes)
Unless a spoke says otherwise:
- Source: first `separators[0]` string on a deduped confusionPair edge (or graph metrics for 1a–1f).
- Catalog size: 1,534 dossiers at export date 2026-07-09.
- Filters: combined planting caps and exclusion stacks are documented per spoke.
- Not claimed: live exam miss-rates, chemistry papers, or “most confused in the world.”
If two spokes seem to overlap (e.g. Syrah near Carmenère), read the separator wording — that is the fork the research owns.
What we will publish next (Study 2)
When anonymized Train/Blind wrong-answer volume clears a documented n threshold, Study 2 will publish miss-rates beside Studies 1a–1o — not instead of them. Until then:
- Prefer this field guide + the spoke URLs for citations.
- Prefer publish-ops / GEO distribution over inventing a sixteenth thin aroma cut.
- Do not fabricate Study 2 tables.
Frequently asked questions
Is this the same as the year-in-review pairs list?
No. The year-in-review hub is a checklist of pairs. This page is a research-cut index (1a–1o) with citation rules and a drill stack.
Which cut should I open first?
If you are new: 1a then 1d then 1j and 1o (graph → identity → Syrah pepper → Cabernet fruit-shape). If you teach whites: 1k then 1n. If you teach Italian pale walls: 1c then 1l (Nebbiolo rose-tar).
Can I cite the whole stack in one footnote?
Yes — “Sensium Study 1 field guide (1a–1o), 2026-07-09” plus this URL. For a specific claim, cite the spoke that owns that separator family.
How do I reproduce any table?
Run the export in the table above from the repo root, e.g. `node scripts/data/export_pepper_smoke_pairs.mjs`.
Why no Study 1p aroma cut?
After 1o, residual first-separator aroma/texture edges under the cumulative exclusion stack are too thin for a durable teaching list. The honest next research asset is Study 2 when volume clears — or a non-separator Band G page like this field guide.
Bookmark this page, open 1a and 1o this week, and force one stop-rule question in Blind before any place fantasy. Stack first — then fruit poetry.