Home / Methodology

Methodology

How a flavour, an ingredient and a molecule become one record, which signals we watch, and how a trend score is computed. Written for readers, not for statisticians.

One identity, many names

Each flavour, ingredient or molecule is an entity with a stable slug. Every external identifier we know for it (a CAS number, a FEMA number, a Wikidata item, a PubChem CID, an EU FL number) is stored as a cross-reference pointing at that entity. When a new source arrives we match on those identifiers first and on names and synonyms second, so "limonene", "D-limonene", CAS 5989-27-5 and FEMA 2633 all resolve to the same page.

Limonene entity · /molecules/limonene CAS 5989-27-5FEMA 2633PubChem 440917EU FL 01.045Wikidata Q278809
The identity graph: one entity, any number of cross-references (xrefs).

The signals we track

Five kinds of weekly signal are attached to a flavour. Each is stored as a time series and kept separate so you can see which one moves.

WikipediaattentionPressmentionsGDELTnews volumePapersresearchProductslaunches0.350.200.150.100.20
Default weights of the momentum blend. Signals that have no data for a flavour are dropped and the remaining weights are re-normalised.
  • Wikipedia attention: weekly article views, the broadest and most stable public-interest proxy.
  • Press mentions: headlines in trade publications that name the flavour.
  • GDELT news volume: share of global news articles mentioning the flavour.
  • Papers: new research articles indexed with the flavour or its key molecules.
  • Products: launches linked to the flavour in Open Food Facts.

Momentum

Momentum answers one question: is attention rising right now compared with the recent past? For each signal we compare the mean of the last 4 weeks with the mean of the 48 weeks before it, then blend the per-signal ratios with the weights above.

prior 48 weeks · baselinelast 4 weeks−52 wnow

momentum = Σsignal ws × ( mean(last 4 weeks) − mean(prior 48 weeks) ) / mean(prior 48 weeks)

A value of +0.40 reads as "attention is 40% above its yearly baseline". Flavours with fewer than 12 weeks of history get no momentum rather than a noisy one.

Lifecycle stages

Momentum says how fast; the lifecycle stage says how far along. It combines absolute attention with the momentum trend.

EmergingLow attention, strongly positive momentum. First menu and press appearances.
GrowingRising attention, launches accelerating across categories.
MainstreamHigh attention, flat momentum. Present in mass-market products.
MatureHigh or stable attention, negative momentum. A classic, not a trend.

Limits

Descriptions, accords and some links between flavours and molecules are produced with the help of AI models and editorial review. AI and model outputs can be wrong. Signals measure attention, not sales. Regulatory identifiers are reproduced for reference only: verify regulatory status against the official text before using it in a product decision.