Every teaching accord drawn as a fingerprint of the records: descriptor families, volatility, structure. Compare any two. A picture of what is filed, never a prediction.
An interpretable fingerprint — not a predictive model Each teaching accord drawn as a fingerprint: the share of the accord on each descriptor family, its volatility and its structure, summed from the records. Nothing is learned or predicted.
The comparison view needs JavaScript, which is off or has not loaded. Every fingerprint below is drawn without it, with its vector as a table.
Compare two fingerprints
A and B on each family axis (share of each), and the difference. An interpretable fingerprint — not a predictive model.
Family axis
A
B
B minus A
How a fingerprint is made
Components and shares. A teaching accord’s components with their modeled teaching parts, made to sum to one. The teaching accords are the house’s, for learning: not a recipe, not for skin, never Radler’s own compositions.
Words. Each component’s words are the odour descriptors its record links. A material whose record files no words takes its constituents’ words, marked “via”. A component with no words keeps its share apart, as “no words on record”.
Families. A component’s share is split evenly over its words, and each word counts on its family axis (the descriptor record’s family, after the published classifications). Words filed as “a quality, not a family” (fresh, sweet, soapy and others) go to the qualities line.
Volatility and structure. The shares by volatility class, read as the Sillage Lab reads them (modeled); the count of components, molecules and materials; the regulation records the accord names (IFRA Standards are the industry’s own standards and are not law in the US; EU rules apply in the EU; each record names its jurisdiction); the historical fragrances it is linked to.
Resemblance. The cosine of two family vectors, from 0 (no family in common) to 1 (the same proportions). Read as close (0.8 or more), related (0.5), distant (0.2) or unrelated.
What it is not. An interpretable fingerprint — not a predictive model. It predicts nothing about how a perfume smells; it shows, in one picture, what the records file about an accord. Change a record and the fingerprint changes with it. Other approaches map molecules to odour from data at scale; this is not one of them.
The vector behind the wheel: the share on each family axis and the components that put it there. Modeled teaching shares; words as filed on the records.
The vector behind the wheel: the share on each family axis and the components that put it there. Modeled teaching shares; words as filed on the records.
The vector behind the wheel: the share on each family axis and the components that put it there. Modeled teaching shares; words as filed on the records.
The vector behind the wheel: the share on each family axis and the components that put it there. Modeled teaching shares; words as filed on the records.
The vector behind the wheel: the share on each family axis and the components that put it there. Modeled teaching shares; words as filed on the records.
The vector behind the wheel: the share on each family axis and the components that put it there. Modeled teaching shares; words as filed on the records.
The vector behind the wheel: the share on each family axis and the components that put it there. Modeled teaching shares; words as filed on the records.
The vector behind the wheel: the share on each family axis and the components that put it there. Modeled teaching shares; words as filed on the records.
Family axis
Share
From
citrus
0.11
Iso E Super (lemony)
aromatic
0
nothing on record
herbaceous/herbal
0
nothing on record
green
0
nothing on record
fruity
0
nothing on record
floral
0.22
Iso E Super (floral); Iso E Super (violet)
spicy
0
nothing on record
balsamic
0
nothing on record
amber
0.11
Iso E Super (ambergris)
wood
0.48
Iso E Super (woody); sandalwood oil (creamy); sandalwood oil (milky); vetiver oil (smoky); vetiver oil (rooty)
musk
0
nothing on record
moss/leather/animalic
0
nothing on record
aldehydic
0
nothing on record
qualities (not families)
0.08
sweet 0.08
no words on record
0.00
–
Iso E Super · 55 parts · woody, floral, ambergris, violet, lemony
A historical fragrance record holds no composition beyond what a filed source names, and no proportions. Its fingerprint is therefore short: the family it is classified in, its year and house, and any material a source names. Where a source names materials with words on record, they are compared above with equal weights. The teaching accord of its family is the house’s, offered for learning; it is not the fragrance’s composition.
The historical fragrance records: what the records hold. Names are trademarks of their owners; Radler Parfums is not affiliated with them and names them for history and education.
