FlavorDB / FlavorDB2
~25,595 flavor molecules; 2,254 linked to ~936 ingredients across 34 categories. This library uses a curated subset, not a live or licensed extract. Commercial density requires clearance.
02 · Pairing
Molecular overlap plus what cooks actually pair.
Search an ingredient, inspect the chemistry, and rank it against what you already own. Classic bridges, unexpected chemistry, and the same scores Match uses.
Data version 2026.08.19-curated-v7·Last reviewed 2026-08-19·Curated library — not a live or licensed FlavorDB / FooDB extract.
FooDB
~70k+ food constituents, nutrients, PubChem links. CSV / JSON / MySQL dumps. Curated subset only — not a live pull. Same license boundary as FlavorDB for commercial density.
Supporting sets
Flavornet, BitterDB, SuperSweet, Fenaroli, ChemTastesDB, odor thresholds, and GC-MS literature. Used to ground key odorants in each profile.
Recipe co-occurrence
OpenRecipes, public-domain cookbooks, and CC collections for NPMI-style statistics. Cultural and regional bias remain — Western and classic bar corpora are still over-represented relative to Levantine, Indian, West African, Mexican, and East Asian practice.
Explorer
Why this pairs
spirit
mixer
0.13Molecular Jaccard
0.90Recipe co-occurrence
0.51Composite
Why Cooks pair these often; shared volatiles in this curated set are still thin.
Synergy gate +0.08 only when Molecular Jaccard > 0.25 and Recipe co-occurrence > 0.5. Never invents a pair from one signal alone.
Bonus on this pair: not applied
Shared compounds
vanillin· Vanilla, sweet oak
A: richB: moderate
Oak, vanilla, baking spice. Fortified botanical wine.
Best neighbors for Bourbon
Method
- Score = hybrid of (A) IDF-weighted molecular Jaccard and (B) recipe co-occurrence, plus a synergy bonus when both are high.
- (A) Molecular: IDF-weighted Jaccard on shared volatiles — rare compounds count more than ubiquitous ones (hexanal, ethanol).
- (B) Co-occurrence: NPMI-style statistics over open recipe collections and classic pairs (not proprietary menus).
- Default blend is 50/50. Sparse profiles lean 35/65 toward co-occurrence so ranking stays useful while chemistry density is incomplete.
- Synergy bonus: +0.08 only when IDF molecular > 0.25 AND Recipe co-occurrence > 0.5. Never invents chemistry from one side alone.
- Coverage labels (rich / moderate / sparse) mark how dense each profile is.
- Sources, formula, data version, and limits are always visible: matrix effects, cooking transformations, sparsity, cultural bias, license constraints.
- Any generative suggestion must be constrained by these scores so unsupported pairs cannot be invented.
- Match flavorScore uses the same pair table but blends 35% molecular / 65% co-occurrence on required on-hand pairs, so ranking tracks the denser corpus.
- Unexpected bridges: IDF molecular ≥ 0.25 with co-occurrence still below 0.40 — chemistry without (yet) a recipe habit.