I looked up every material name in 954 formulas. 53.7% missed — and that is not a coverage problem

About a 5 minute read.

I keep answering questions with this database, and I keep running statistics on those 954 public formulas. But I had never asked a basic question: how many of the material names in those formulas can the database actually find?

The numbers first

The 954 formulas contain 2,994 distinct material names across 17,481 occurrences.

Looking each one up (against name and head_synonym, after stripping a trailing dilution marker like "10%"):

	Count	Share
Matched on the name field	1,186	39.6%
Matched on the synonym field	199	6.6%
Missed	1,609	53.7%

More than half miss. And those misses account for 6,936 occurrences — 39.7% of all occurrences.

That looks bad. It is not a coverage problem.

The top twenty misses

Occurrences	Name
445	dipg
165	hedione
131	bergamot oil bergaptene reduced
113	ylang ylang oil
104	iso e super
93	cinnamic alcohol
82	galaxolide 50 ipm
77	styrallyl acetate
76	gamma-undecalactone
60	cis-3-hexenol
52	orange oil
52	triplal
50	lyral
50	vertofix
49	phenylethyl alcohol
49	verdox
42	linalol
42	aldehyde c-14

Not one of these is a material the database actually lacks.

Six kinds of gap

I resolved the top twenty by hand (using only the name and synonym fields, never supplier text — that layer produces false positives, a trap I walked into in Four ways a material has no data):

1. Abbreviation.

Formula writes	Database name	Suppliers
dipg	dipropylene glycol	37

The most frequent name in the whole corpus is an abbreviation. 445 occurrences, 2.5% of everything, and it is dipropylene glycol — not even a material, a solvent.

2. Trade name absent from the name field.

Formula writes	Database name	Suppliers
hedione	methyl dihydrojasmonate	102
iso e super	patchouli ethanone	79
vertofix	methyl cedryl ketone	71
triplal	2,4-ivy carbaldehyde	54
verdox	green acetate	52
undecavertol	violet decenol	29

This database uses a chemical description as the primary name and puts the trade name elsewhere. Why Hedione does not come up got its own article.

3. The trade name is there, spelled differently.

Formula writes	Database name	Suppliers
lyral	leerall	40

This one is worth pausing on. Lyral and leerall are two trade spellings of the same material, and the database carries the second. String matching will never find it; a person sees it immediately.

4. Spelling variants.

Formula writes	Database name
linalol	linalool
phenylethyl alcohol	phenethyl alcohol
cinnamic alcohol	cinnamyl alcohol
styrallyl acetate	styralyl acetate

The last one differs by a single L.

5. The database name has an alias embedded in it.

Formula writes	Database name
gamma-undecalactone	gamma-undecalactone (aldehyde C-14 …)
aldehyde c-14	same record

One record, and neither spelling finds it — while the name contains both. Because the match is whole-string equality, not containment. That record has 119 suppliers; the two spellings together occur 118 times in the corpus.

6. Coarse natural name versus the database's specific one.

Formula writes	What the database has
ylang ylang oil	ylang ylang flower oil, flower oil I / II / III
orange oil	orange oil terpenes, blood orange oil italy…
bergamot oil bergaptene reduced	bergamot oil bergaptene reduced italy

The formula names a category; the database carries specific products. The last one differs by an origin word. (Different products sitting under one CAS is something I measured in Records that share a CAS.)

So what does "53.7% miss" mean

It is naming distance, not a gap in coverage.

Of the 1,609 misses, the top twenty I resolved by hand all have a corresponding record, most with a healthy supplier count (37 to 119). Materials genuinely absent from the database certainly exist — newly launched compounded bases, for one — but they are not the bulk of that number.

What this means for you: when a lookup fails, do not start by assuming the database lacks it. Run these six first.

A lookup checklist

1. Is it an abbreviation? dipg, IPM, DEP, TEC.
2. Is it a trade name? Look up the chemical name. Hedione, Iso E Super.
3. Is the trade name spelled the way you think? Lyral / leerall differ by a letter.
4. A spelling variant? linalol / linalool, -yl / -ic endings.
5. Try "contains" instead of "equals." The database name may have a parenthetical alias inside it.
6. For naturals, add origin and plant part and try again. "Neroli oil" becomes "neroli oil france."

If all six fail, then it is worth writing down — at that point it might genuinely be new.

What this doesn't establish

• The six kinds are induced from the top twenty. I did not classify the remaining 1,589 misses, so I do not know their relative proportions.
• The resolution was done by hand, using what I already knew about these trade names, not by an automated match.
• "Genuinely absent materials are not the bulk" is an inference. Proving it would require working through all 1,609 manually.
• The match used only name and head_synonym. The synonyms field holds truncated fragments (things like "3,7-") and cannot be matched against.
• The 954-formula corpus is itself biased — publicly shared formulas are not all formulas, and this batch skews toward commercial demonstration formulas.
• Dilution handling only stripped a trailing "N%". Forms like "galaxolide 50 ipm" were not normalised.