Baseline OS

Paste an ingredient list (or copy from a packet in your pantry). The module normalizes whitespace and slashes, then runs longest-keyword-first substring matching with a word-boundary special case for vegetable oil (so plural "vegetable oils (palm)" is correctly skipped). Try the examples below.

A soft cap, not a flat penalty — how the rule actually scores

Seed oils are culturally controversial, but current science (Johns Hopkins 2025, Memorial Sloan Kettering, recent meta-analyses) finds them mixed-to-neutral at normal dietary doses. We respect the avoidant consumer's preference without overstating the science — so the rule is a ceiling, not a deduction:

• ▸
  One seed oil present → Processing Score is capped at 80 (Clean max). A product already scoring 65 isn't moved.
• ▸
  Two or more seed oils present → cap at 55 (Moderate max). The "blend of canola + soybean + safflower" formulation pattern.
• ✓
  Detection drives the filter tag too — anyone avoiding seed oils can sort/filter the catalog by it.
• ▸
  Genuinely evidence-backed fat concerns — trans fats, oxidation, hydrogenation — are handled on separate axes with their own evidence floors. Membership ≠ harm.

This keeps us honest: we serve the preference without overstating the science, and we don't punish a tin of tuna in sunflower oil as if it were a hot dog.

What counts as a seed oil — the "Hateful 8" + 1

61 keyword entries in the production set, organized around Tucker Goodrich's 8 canonical industrial seed oils plus refined peanut as a 9th and "vegetable oil" as the US-label catch-all. Each canonical oil includes every label spelling we observed on a 445k-product USDA corpus scan, plus international spellings as insurance for non-US labels.

A "partially hydrogenated soybean oil" hits this set and separately fires the trans-fat axis — membership and evidence are kept on independent tracks.

Three separate questions, three independent tags

Fats get classified on three orthogonal axes — an oil can carry more than one. The UI composes them on read (e.g. "Avoid Concerning Fats" = seed oil OR palm OR tropical).

What's not a seed oil — the deliberate carve-outs

Common mix-ups. These are intentionally excluded with their rationale logged in code, so we never re-litigate them every time someone asks "but what about ___ oil?":

Validation — 100% recall on ~119,000 real products

Measured on two disjoint corpora using an independent ground-truth regex (broader than our keyword set, so it catches anything we'd miss). Truth: label-declared Hateful 8.

FPs are ground-truth-regex artifacts on parenthetical lists where the broader regex flags inside a sublist that detect_seed_oils correctly skips — not real over-flags. PR: peak-health/baseline-data-pipeline#211

All — oils we've reviewed from the Open Food Facts taxonomy. Search a name to see exactly how it's tagged — or whether it's deliberately not flagged.

Where it's still rough (we'd rather you hear it from us)

• ▸
  It's an editorial line, not a hard fact. Open Food Facts has no "seed oil" flag — different sources draw the boundary differently. This is Peak Health's curated call (the "Hateful 8" cluster), maintained by hand.
• ▸
  It rides on the ingredient list. If a label hides the oil behind "vegetable oil" with no detail, we tag the generic — but we can't always tell which seed oil it is.
• ▸
  High-oleic cultivars are still flagged by label. A high-oleic sunflower is fatty-acid-closer to olive than to a standard RBD sunflower, but the label usually still says "sunflower oil." We tag conservatively and let the avoidant consumer decide. Lab-spec carve-out is a future call.
• ▸
  The taxonomy drifts. Open Food Facts is volunteer-edited; new oil entries appear. We re-audit annually and the importer warns on any oil we haven't yet classified.
• ▸
  The science may move. Today we treat seed oils as a soft cap, not a flat penalty. If the evidence base shifts harder either direction, the cap level (80 / 55) is what we'd revisit.