1. Introduction
Volumes 1 and 2 of this report series described the neural mechanisms linking odour to emotion and memory at the systems level: the anatomical pathways, the roles of the amygdala, OFC and hippocampus, and the certainty of applied claims. The remaining question is the one the fragrance industry actually needs answered: what do we know when we get down to individual molecules?
This question matters because all compounding takes place at the molecular level. A perfumer does not choose “a pleasant smell”; they choose linalool at what concentration, paired with linalyl acetate in what ratio. If a reliable map from molecule to emotional effect existed, fragrance design would move from art to engineering. The industry’s commercial literature generally assumes that such a map already exists.
This report tests that assumption. Three questions:
- Q1 — Coverage. For how many molecules do we actually have a complete data chain from human receptor to measured emotional effect?
- Q2 — Reliability. Where data exist, how good are they, and how many popular claims are in fact misattributed to their source?
- Q3 — Levers. If the molecule is not the main control variable, which variables are, and how well are they quantified?
A note on scope. The question of “steering emotion with odour” has two very different readings. The first is deliberate design of an experience that the user knows about and chooses — a bottle of perfume, a shop space, a care product. The second is acting on people who do not know they are being acted upon. This report serves the first reading. The data below also show that the second reading has no technical basis: effects below the threshold of awareness are all small, unstable across individuals, and disappear when the experimental design controls for expectation.
2. Methods
2.1 Design
Four independent collection streams, run in parallel, sharing an 18-field data schema (Appendix D):
- Stream R — human odorant receptors with identified ligands, and genetic variants that alter perception.
- Stream N — human neuroimaging and electrophysiology data, extracted per chemical named in the Methods section of the original paper.
- Stream A — psychophysiological and emotional trials in humans, with effect sizes.
- Stream S — structure–odour–valence relationships, concentration, mixture interactions, adaptation, thresholds, cultural differences.
2.2 Four mandatory rules
- Human data only for the neural and emotional tiers. Animal data are used only for the mechanism tier, and the species must be stated. This is the rule that eliminates the most claims.
- Concentration and mode of presentation must be recorded. One molecule at two concentrations is treated as two data rows.
- Preparations must be classified under four labels: PURE pure molecule · DIL molecule diluted in an inert solvent · MIX essential oil or extract · ORAL oral preparation. Lavender oil contains over a hundred compounds; all of its results fall under the MIX label and may not be attributed to linalool.
- Conflicts of interest must be recorded, especially funding from fragrance houses.
2.3 Data merging
The four databases were merged by normalised name plus a synonym table, separating compound-class entries and mixtures from single molecules. When two streams described the same compound, list fields were combined and text fields kept both versions. An entry was counted as having data at a given tier only when its source was an actual study, not a cross-reference to another stream.
2.4 Limitations of the method
This is not a systematic review with a registered protocol. Molecule selection started from a predefined list plus expansion as further compounds were found, so there is a bias towards heavily studied compounds and commercially common ones. Many key sources sit behind paywalls or bot verification and only their abstracts were accessible; every such case is flagged. There was no independent double extraction by two people.
Scientific reference material, not medical advice and not promotional material. All assessments of evidence quality are the compilers’ own. Figures are redrawn from published data.