Beginner perfumer · 85
Rescuing the field that failed last time: add one specificity condition and it becomes a usable accord palette
· 12 min read
Last time I showed that potential_uses does not work as a versatility ranking — seven of its top hundred have no suppliers. But the field is not junk; I was reading it the wrong way round. Run it backwards: the chypre tag covers 506 materials and fern 752, with 57% of each having ten or more suppliers. The second condition is specificity — linalool is tagged with 269 accords, so its chypre tag carries no information, while cistus twig/leaf absolute is tagged with seven, one of which is chypre, and that tag carries a lot. Filtered that way the lists read like real palettes: cistus absolute, isobutyl quinoline and geranium absolute for chypre; lavandin absolute, hyssop oil and birch tar oil for fougère.
About 4 minutes.
Last time I tested potential_uses as a versatility ranking and showed it cannot be used that way — of the top hundred by accord count, 38 have fewer than ten suppliers and 7 have none.
But that conclusion only rules out one use. This article rescues the field.
Run it backwards
The field's reliable use is the reverse direction: not "how many accords can this material enter" but "which materials can enter this accord".
| Accord | Materials tagged | With 10+ suppliers |
|---|---|---|
| Chypre | 506 | 290 (57%) |
| Fougère (fern) | 752 | 429 (57%) |
More than half the candidates for each accord are buyable. That alone is more useful than the ranking.
But sorting by supplier count is still wrong. The two lists have almost the same top:
| Chypre top six | Fougère top six |
|---|---|
| Patchouli oil 148 | Geraniol 198 |
| Linalool 135 | Vanillin 193 |
| Raspberry ketone 112 | Patchouli oil 148 |
| Basil oil 110 | Linalool 135 |
| Nootkatone 104 | Citral 135 |
| Hedione 102 | Citronellol 119 |
Patchouli, linalool, raspberry ketone and basil oil appear on both. That is less likely to mean they are the shared key to both accords than that they appear on every list.
Add specificity
Last time established the cause: a material's "accord count" is really how many accord pages list it.
That flaw converts directly into a filter.
- Linalool is tagged with 269 accords. Its chypre tag carries almost no information — it is tagged with everything.
- Cistus twig/leaf absolute is tagged with seven, one of which is chypre. That tag carries far more.
So the condition becomes two: enough suppliers (buyable) and few enough total accord tags (specific).
The filtered lists
Chypre (20+ suppliers, fewest total accord tags):
| Material | Suppliers | Total accords | Odour type |
|---|---|---|---|
| Rose absolute (Rosa centifolia) | 34 | 3 | floral |
| Geranium absolute | 22 | 6 | green |
| Dihydroeugenol | 23 | 7 | spicy |
| Petitgrain mandarin oil | 26 | 7 | floral |
| Cistus twig/leaf absolute | 51 | 7 | woody |
| Hop oil | 21 | 8 | herbal |
| Agarwood oil | 26 | 8 | woody |
| Curled parsley leaf oil | 26 | 9 | herbal |
| Valerian root extract | 30 | 9 | fruity |
| Isobutyl quinoline | 22 | 10 | animal |
Fougère (same conditions):
| Material | Suppliers | Total accords | Odour type |
|---|---|---|---|
| Green cognac oil | 23 | 2 | green |
| White cognac oil | 32 | 3 | herbal |
| Copaiba balsam | 25 | 4 | woody |
| 3-Ethyl pyridine | 26 | 4 | tobacco |
| Opoponax oil | 25 | 5 | balsamic |
| Petitgrain bigarade oil | 26 | 6 | floral |
| Spikenard oil | 26 | 6 | woody |
| Lavandin absolute | 37 | 6 | herbal |
| Hyssop oil | 52 | 6 | herbal |
| Birch tar oil | 30 | 7 | smoky |
These read like palettes rather than rankings.
Cistus (labdanum) and isobutyl quinoline are chypre backbone; lavandin, hyssop and birch tar point at fougère. And none of them was picked by me — two conditions produced the lists.
Why the filter works
Because it separates "thoroughly documented" from "belongs to this accord".
A material appearing on many accord pages means the compiler cross-referenced it thoroughly. A material appearing on only a few, one of which is chypre, more likely reflects that it really belongs there.
This is a general move: when a field's noise comes from documentation completeness, using the total tag count as a denominator divides that noise out. The lift I computed for formula co-occurrence in I bought thirty materials and got stuck is another version of the same idea.
What to do
- To build an accord, pull its tag and apply both conditions: enough suppliers to buy, few enough total tags to be specific.
- Tune the tag threshold yourself. I used ten or fewer to keep the list readable; twenty brings in another tier.
- Do not use the ranking direction. Last time showed it is broken.
- What comes out is a candidate list, not a formula. You still have to smell, test and know the doses.
- The materials topping both lists are still worth noticing — patchouli is top three in both, and measured on other data it spans all fourteen formula types, so that may genuinely be its nature.
What this doesn't establish
- "Few accord tags means specific" is a proxy, not a definition. A material may simply be sparsely documented.
- The thresholds of 20 suppliers and 10 tags are mine, and different numbers give different lists.
- I did not verify that any of these materials is used in commercial chypres or fougères. This is an analysis of database tags, not perfume history.
- I do not know how this field defines "chypre" or "fern". They are accord page names, and accord boundaries are contested anyway.
- Last time's caveat still holds: the field reflects the compiler's indexing, and different accord pages may be documented to different depths.
- I handled only FR and FL/FR tags. The 11,132 flavour tags remain untouched.