Beginner perfumer #114: the formula says 'ylang ylang oil', the database has 22 records, which do you buy?

About a 5 minute read.

The short version

• The corpus holds 16 spellings of ylang across 152 formulas; 113 write the unqualified ylang ylang oil
• Exactly 2 name a grade
• The database holds 22 ylang records, four of them serious candidates
• Grade I: substantivity 319 hours, 10 words in the odour field
• Grades II and III: 2 words each (floral; ylang), and no strength rank at all
• This is not a name mismatch. It is the corpus naming things at a coarser grain than the database
• Scale: 39 natural-material names appear 15+ times, with a median of 21 candidates each. lemon oil has 98

The naming problem stopped moving here

I spent four rounds on names: the alias field is broken, strip the manufacturer parenthetical, the aldehyde C-numbers need a lookup, and twelve hand aliases take it to 82%.

Of the 3,150 records still unmatched, 795 are natural materials. Last round I said that is not a naming problem. This article opens it up.

Twenty-two ylangs

ylang ylang oil appears 113 times in the corpus. Searching the database for that string hits 22 records. Here are the four candidates that have suppliers:

Database record	Suppliers	CAS	Substantivity	Strength rank	Odour field
ylang ylang flower oil	93	8006-81-3	140 h	2	floral; petal; neroli; lily; spicy (5 words)
ylang ylang flower oil I	42	68606-83-7	319 h	2	floral; petal; jasmin; neroli; lily of the valley; spicy; balsamic; earthy; woody; winey (10 words)
ylang ylang flower oil II	30	68606-83-7	(empty)	(empty)	floral; ylang (2 words)
ylang ylang flower oil III	45	68606-83-7	(empty)	(empty)	floral; ylang (2 words)

Three things are worth stopping on.

One. Grade I's substantivity is 2.3 times the unqualified record's, 319 hours against 140. Buy III thinking you are following the formula and it will not hold for as long as you expected.

Two. The odour fields differ fivefold in word count. Grade I has ten, II and III have two each. That is not the compiler being lazy. Ylang is fractionally distilled: grade I is the earliest fraction, where the light esters concentrate, and later fractions get progressively simpler. The poverty of the field reflects the poverty of the material.

(Grade III's evaluation says: "Sweeter and lighter than the extra quality, more spicy carnation. Aquatic sea effect and definitely more woody guaiac like." It refers to an extra grade, and the database has no record for one.)

Three. The CAS numbers differ. The unqualified record is 8006-81-3; all three grades are 68606-83-7. So matching by CAS does not line them up either.

Two formulas out of 152 say which grade

Sixteen spellings across 152 formulas:

Corpus writes	Formulas
ylang ylang oil	113
ylang synth	18
ylang ylang oil replacer	6
ylang oil extra	2
ylang ylang flower essence	2
ylang ylang flower oil i	1
ylang ylang flower oil iii	1
nine other spellings, one each	9

113 against 2. People writing formulas almost never write the grade.

Something else: ylang synth (18), ylang ylang oil replacer (6), plus scattered ylang (synthetic) and ylang ylang oil imitation entries add up to about 24 formulas choosing not to use the natural at all, or 16% of the 152.

(Merging every spelling, the dose is a median of 2.50%, quartiles 1.46–4.90%, max 24.8%.)

How widespread this is

I pulled every natural material appearing 15+ times whose name contains oil, absolute or extract, and counted how many database records share its stem:

Corpus formulas	Database candidates	With suppliers	Name
66	98	88	lemon oil
52	79	76	orange oil
23	58	43	jasmin absolute
48	45	30	geranium oil bourbon
33	23	19	mandarin oil
132	21	19	patchouli oil
113	16	12	ylang ylang oil
60	12	9	sandalwood oil
40	9	8	clove bud oil

Thirty-nine names in all, with a median of 21 candidates.

A formula that says lemon oil points at 98 database records. That is not something you can look up.

Merging here is a choice, not a correction

The merges in the last four rounds were corrections: dipg and dipropylene glycol are the same material, and merging moves the numbers closer to the truth.

This is different. Collapsing ylang's four grades into one throws away the grade, and the grade is worth the difference between 319 hours and 140.

A correction points two names back at one thing. Merging grades treats four different things as one.

Lebanov and colleagues did the quantitative version of this in J Chromatogr A in 2020, classifying ylang oils systematically by distillation time, origin and chemical composition using a multivariate approach (PMID 31959459). The grade differences are measurable, not marketing.

So what do you buy

When a formula says ylang ylang oil and nothing more:

1. Look at the dose. Above about 3% the author is using it to carry floral volume, and the unqualified complete oil (93 suppliers, the most widely sold record) is a safe default.
2. Below 1% it is usually there for the spicy top and the fruity esters, which is closer to grade I.
3. III is the cheapest and the flattest (two words in its odour field). It suits functional products, not a place where you need it to carry the theme.
4. Ask the supplier for a GC trace, or at least for the grade, before you buy. The price and quality spread on this material is wide.

And if you are just practising, buy one bottle of the complete oil. Do not start with four grades.

What this doesn't establish

• "Grade I is the earliest fraction" is a general principle, not something this database says. The database gave four records and their field differences. The distillation-time-to-composition relationship comes from Lebanov and colleagues.
• A short odour field does not necessarily mean a simple material. It could just mean nobody filled those two records in. I used the empty strength ranks on II and III as supporting evidence, but that is still a data-completeness argument.
• The "about 24 synthetic replacers" figure sums every spelling that looks synthetic, and I did not verify each one individually.
• Candidate counts come from stem matching. The 98 for lemon oil includes lemon oil italy, lemon oil terpenes and so on. They are candidates, but some are obviously not what the author meant. This is an upper bound, not the real ambiguity count.
• I have not compared the actual smell of any two grades. This rests on fields and one paper.
• The buying advice is inferred from fields and doses, not from trading experience. I have not bought these four grades.