The science

Why two people disagree about the same bottle

On average, two people differ functionally at more than 30% of their odorant receptor alleles (Mainland et al., 2013, Nature Neuroscience). Across more than 1,600 people tested, the probability of having some specific blind spot approaches one — which one is individual to you (Croy et al., 2015, Cortex).

That is not a curiosity. It is the reason a fragrance everyone raves about can smell like nothing on you, and why we warn you when a recommendation leans on one of these materials.

Known blind spots we screen for

Prevalence varies a lot by ancestry, so these are ranges rather than single numbers.

• Beta-ionone, violet, iris and ionone florals

  OR5A1 (rs6591536)

  Measured: 48.4% among people of East Asian ancestry, 38.1% admixed, 6.3% African ancestry, so the range is genuinely wide by population.

  Leans on violet and orris. A common OR5A1 variant makes beta-ionone barely register for a large minority of people, closer to one in two in some populations. If violet or powdery-iris scents have always smelled faint or empty to you, this one may too.

  Jaeger et al. 2013, Current Biology

• Androstenone and androstadienone, animalic musk

  OR7D4 (R88W + T133M)

  Measured: roughly half of adults perceive no odour at all when sniffing androstenone.

  Carries animalic character. About half of people cannot smell androstenone at all, and among those who can, a single receptor gene decides whether it reads sweet-floral or strongly urinous. This is the note most likely to make you and a friend disagree violently about the same bottle.

  Keller et al. 2007, Nature; Wysocki et al. 1989, PNAS

• Galaxolide, polycyclic "clean laundry" musk

  OR4D6 (rs1453542) · prevalence estimated from published allele frequencies

  Estimated from published allele frequencies: roughly 7.8% (European ancestry), 2.9% (East Asian), 0.3% (African).

  Built on clean white musk. Around one in thirteen people of European ancestry carry two copies of a variant that makes this musk class smell like nothing. If "clean musky" scents have always seemed to vanish on you within minutes, that is likely why.

  Li et al. 2022, PLoS Genetics

• Guaiacol, smoky, phenolic, tar-like

  OR10G4

  No population prevalence has been published — we do not invent one.

  Smoky phenolic notes are one of the few where a single gene measurably shifts both how strong something smells and whether people like it, genotype explains over 15% of the variation in perceived intensity. Strong reactions in both directions are normal here.

  Mainland et al. 2013, Nature Neuroscience

• cis-3-hexen-1-ol, cut grass, green leaf

  OR2J3 (rs28757581) · prevalence estimated from published allele frequencies

  Estimated from published allele frequencies: roughly 14% (African ancestry), around 1.4% (European and East Asian).

  The classic cut-grass green note has a known receptor variant. If green fragrances tend to read as flat or just "generic fresh" to you, you may be smelling a narrower version of this than the reviews describe.

  Li et al. 2022, PLoS Genetics

• Muscone and macrocyclic musks

  OR5AN1 (L289F)

  No population prevalence has been published — we do not invent one.

  Musk anosmia is material-specific, several unrelated musk molecules use different receptors, so smelling one kind tells you nothing about another. There is also evidence that people most sensitive to violet tend to be less sensitive to macrocyclic musks.

  Sato-Akuhara et al. 2023, Chemical Senses

How we measure similarity

We compare the whole scent profile as one vector rather than counting shared notes. Snitz et al. (2013, PLoS Computational Biology) tested exactly that choice: a model comparing components separately predicted perceived similarity poorly, while treating the mixture as a single vector reached r = 0.85. Ravia et al. (2020, Nature) then built a validated similarity measure as an angle over a feature vector, which is the form we use.

A consequence worth knowing: shared notes are neither necessary nor sufficient for two fragrances to smell alike. That is why we will sometimes recommend something whose note list looks nothing like what you love.

We are honest about the limit, too: that research validated the metric on molecules of known composition. Ours are human-assigned descriptions of commercial formulas that are trade secrets. The approach is reasonable; the transfer is an assumption.

Why our confidence numbers are modest

In the DREAM Olfaction Prediction Challenge (Keller et al., 2017, Science), predicting how pleasant a molecule smells reached r ≈ 0.71 for a population average but only r ≈ 0.41 for a specific individual. Nobody can currently predict one person’s taste with high accuracy. So we show a confidence band and say where the uncertainty comes from, instead of printing a reassuring percentage.

Why we say sample first

Sniffing in a shop adapts your nose within minutes, and that adaptation is odour-specific and can persist for weeks after long exposure (Dalton, 2000; Dalton & Wysocki, 1996). A blotter also evaporates differently from skin.

There is a useful asymmetry in first impressions: repeated exposure raises liking for odours you found neutral or mildly pleasant, but does notrescue ones you actively disliked (Delplanque et al., 2015). So “boring” is worth a second wearing. “I hated it” usually is not — and we treat those two signals differently.

Things we will not tell you

These sound scientific and are not supported, so they do not appear anywhere in this product:

• That your skin pH or “body chemistry” changes how a fragrance smells.
• That heat amplifies projection — the controlled evidence points the other way.
• That top, heart and base notes are three phases your nose experiences.
• That an extrait is proportionally stronger than an eau de toilette.
• That seasonal matching is a fact about perception rather than a convention.

Projection and longevity ratings are community consensus, not measurements. There is no agreed scientific standard for either, and we label them accordingly.