Beginner perfumer · 84
One field looks like a versatility ranking for materials, and seven of its top hundred have no suppliers at all
· 11 min read
The database has a field nobody uses called potential_uses, recording which accord types a material can go into — 7,807 records and 72,233 fragrance uses, from herbal, woody and floral to fougère, chypre and oriental. It looks like exactly what a beginner wants: an objective versatility ranking. Cross it against supplier counts and it breaks. Of the top hundred by accord breadth, 38 have fewer than ten suppliers and 7 have none; of the top five hundred, 33 have none. Its Spearman correlation with supplier count is only 0.365. The sharpest counterexample is patchouli oil: measured across 954 formulas it spans all fourteen types, while this field gives it 77 accords — behind a zdravetz absolute with zero suppliers.
About 5 minutes.
There is a field in this database I had never used: potential_uses.
It records not blending partners but accord types — which kinds of composition a material can go into.
It is not small
- 7,807 records have content (18.5% of the database), 2,616 of them with ten or more suppliers
- Entries split into FR (fragrance) 72,233, FL (flavour) 11,132, FL/FR 4,492
- Uses per material: median 3, mean 12.4, maximum 326
And the accord names are the familiar ones:
| Accord | Occurrences |
|---|---|
| Herbal | 1,260 |
| Woody | 1,186 |
| Floral | 1,170 |
| Balsam | 887 |
| Amber | 798 |
| Citrus | 797 |
| Fern (fougère) | 752 |
| Green | 734 |
| Spice | 730 |
| Oriental | 698 |
| … | |
| Chypre | 506 |
This looks like precisely the list a beginner wants: how many accord types a material can enter.
The ranking
Deduplicating the fragrance (FR) uses and sorting:
| Material | Accords | Suppliers |
|---|---|---|
| Linalool | 269 | 135 |
| Coranol (Firmenich) | 265 | 15 |
| Laevo-linalool | 262 | 25 |
| Ethyl linalool | 247 | 26 |
| Freesia heptanol | 246 | 35 |
| Zdravetz absolute | 201 | 0 |
| Citrus ocimenol acetate | 186 | 1 |
| Leerall | 182 | 40 |
| Citronellyl senecioate | 174 | 5 |
| beta-Sesquiphellandrene | 169 | 1 |
| Vanillin | 166 | 193 |
Linalool at the top makes sense — it is the most widely occurring molecule in this database.
But the sixth entry has zero suppliers.
Measuring how common that is
There are 4,618 materials with FR uses (median 5, mean 16.5, 90th percentile 55, 99th percentile 116).
Breaking down the top of the ranking:
| Fewer than 10 suppliers | Zero suppliers | |
|---|---|---|
| Top 100 by accords | 38 | 7 |
| Top 500 by accords | 196 | 33 |
More than a third of the top hundred are materials you can barely buy.
And the overall relationship is weak: Spearman correlation between accord count and supplier count is 0.365.
That is not zero — the field carries real signal. But it is not a proxy for how much a material is used.
The sharpest counterexample
In Workhorse materials can be measured I did another version of this, sorting 954 published formulas into fourteen types by title and counting how many types each material spans.
Patchouli oil came out spanning all fourteen — the most versatile of the set.
And this field gives patchouli oil 77 accords — far behind coranol (265 accords, 15 suppliers) and zdravetz absolute (201, zero).
Two ways of measuring versatility, two different answers.
Why
I cannot confirm how this field was built, but the shape of the data suggests something.
The values in potential_uses are links to accord pages. So a material's "accord count" is really how many accord pages list it.
That number reflects how thoroughly the compiler cross-referenced, not how much the industry uses it. An obscure but thoroughly documented material outranks a common but sparsely documented one.
That explains how a material with no suppliers reaches the top ten.
So what is it good for
- Use it as an ideas list for directions, not a ranking. A material listed under fougère and chypre is a direction worth trying.
- Running it backwards is more reliable: to build a chypre, look at which materials carry the 506 chypre tags. That is a candidate list, not a ranking.
- To judge how much something is used, use supplier counts or a formula corpus. Both have biases (supplier count traps, corpus bias), but at least they measure use.
- A high score with few suppliers is itself informative — it usually marks a well-documented material that the market barely stocks.
What this doesn't establish
- I do not know how
potential_useswas generated. "Cross-references from accord pages" is inferred from the data's shape, not a verified mechanism. - Spearman 0.365 is a correlation, not a cause. Supplier count is not a perfect proxy for use either.
- I counted only FR and FL/FR tags, deduplicated. The 11,132 flavour tags are untouched.
- Accord names are not standardised (herbal and fern sit at different levels), so comparing 269 against 77 is imprecise in itself.
- The patchouli comparison spans two entirely different sources — fourteen types from 954 formulas versus this field's accord pages. Their granularity differs, so "77 versus 14" is not one yardstick.
- I assessed the correctness of no individual use tag.