How we rank

Every shelf table ranks on a single open axis at a time (oils: saturated fat, smoke point, and so on; sweeteners: GI, energy, …). Rank 1 is the numerical champion on that axis— highest or lowest depending on your sort — not a health crown.

Refined and unrefined oils stay separate rows. Plant vs animal origin is filterable. Empty cells mean we do not yet have a method-cited source we will put a number on.

Start from The Cooking Oil Nutrition Guide or The Sweeteners Nutrition Guide or The Plant Protein Nutrition Guide (Beans & Tofu).

On axis charts, only the open column is colored. Green marks the conventionally healthier end of that axis (for example lower saturated fat, higher smoke point). Red is the other end. Sort order does not flip the colors.

On lifestyle recipes, colored columns are the axes in that recipe — each still uses its own polarity, never a whole-row “healthy” traffic light.

Every number carries a source and a year. When readings disagree, the highest tier wins as primary; others stay on hover and on each item page.

Smoke point is stored as a range plus method limits. Fatty acids lean on USDA FoodData Central and other citable lab sources. The numeric catalog, priority ladder, and a separate interpretation shelf (Examine, Harvard, FDA fish advice, and so on): Sources.

After # and Cooking oil, columns run Saturated → Monounsaturated → Polyunsaturated → Omega-6 → Omega-3 → Smoke point.

Fatty acids come first as a saturation ladder, then the PUFA pool splits into ω-6 / ω-3. Smoke point is last: a temperature-ceiling auxiliary, not the hero heat signal.

Everyday “best for …” ranks use named recipes with published decimal weights and smoke gates. Scoring: normalize each axis (min–max), flip when the recipe prefers low, then weighted sum. Missing a required axis, or failing a smoke gate → unranked. Rank 1 is the open-rule champion — not a clinical crown. No sitewide single “best oil.” Not medical advice.

Surfaces: Everyday Sauté · High heat & Frying · Salad & Cold · Heart Healthy · Low Smoke / Lungs · Ancestral & Primal.

A row ships (or stays listed) only when all five fatty-acid axes have a cited reading: saturated, monounsaturated, polyunsaturated, linoleic (omega-6), and ALA (omega-3). A USDA 0 counts. Seed or nut kernels are not proxies for oil composition.

Smoke point is optional. Empty stays — until AOCS Cc 9a-48 (or an equal lab method) is on file with a source and year. Blog or wiki temperature charts do not count. Oils that fail the fatty-acid bar are soft-delisted until filled — thin data does not stay in the table.

Refined and unrefined may share one USDA fatty-acid record when the agency does not split them; that is noted on Sources. Empty smoke alone is not a reason to delist.

/sweeteners (Ultimate) opens on Everyday Health — the most common day-to-day sweetening use — not on a single axis. Columns after Kind still run GI → Energy → Carbohydrate → Sugars. Dedicated axis URLs like /sweeteners/gi carry chart-search intent. On the GI chart, lower ranks first; substitutes stay on the shelf for comparison even when a classical Atkinson GI row does not exist: high-intensity sweeteners are assigned GI 0 (no available carbohydrate) with that limitation on the cell; allulose follows FDA’s negligible-response / 0.4 kcal/g labeling guidance. Empty cells stay empty when we still lack a cite.

Sugars and substitutes share one table; filter All · Sugars · Substitutes. Each name opens an item page with the same sourced cells. Rank 1 is the open-rule champion on the open recipe or axis, not medical advice. Open The Sweeteners Nutrition Guide.

We do not publish a GL axis. This shelf is a per-100 g ingredient catalog — GL needs a chosen serving of available carbohydrate, which invents a second product or misleads on high-intensity rows. Lifestyle ranks use GI + a published fructose / hepatic rule instead. See fructose / liver veto. Not medical advice.

Low GI is not a hepatic free pass: high-fructose rows can look green on GI while still loading the liver. Overall uses two published layers (decimals are ours). Not medical advice; Rank 1 remains the open-rule champion only.

/sweeteners/for/* uses Base − Penalties + Bonuses (0–100), not a benefits pie. Chips: Everyday Health, Weight goals, Diabetic-friendly, Minimally processed, Easy on the liver. Decimals are editorial — published here. Not medical advice.

Overall Base: GI term (GI/100)×50, energy (kcal/400)×10, fructose −(share%/50%)×15 when share ≤ 50%; above 50% veto max 30 and bonuses void. Osmotic / polyol −6. Mechanism bonuses capped; minerals ≤ 2. Chemical-synthesis rows hard-gated. See also fructose / liver veto.

Fat loss: 100 − (kcal/400)×45 − (GI/100)×35 − (fructose share)×15 + bonuses ≤ 15 (no Overall veto). Diabetic / Ancestral: filter then Overall score (GI ≤ 55; process_id = physical_reduction). Liver lean: Overall base only — no osmotic penalty, no bonuses. Detail: docs/sweetener-weight-recipes.md.

Everyday plant protein as most people mean it: beans (including peanut), lentils, chickpeas, tofu, tempeh, edamame, and labeled concentrates. Many tree nuts and seeds are high in protein too — they are not on this shelf; a future Nuts shelf covers them. Peanut is a legume, so it belongs here; that Nuts shelf may also list peanut with the same footnote (* peanut is a bean). Dual-list is allowed when the note stays honest. No g/100g cutoff invents the boundary; the Beans & Tofu label does.

/plant-proteins (Ultimate) opens on Everyday protein (first lifestyle chip). Axis charts live on /plant-proteins/{axis}, including protein. Table columns run Protein → Fiber → Carb → Sodium → Energy → Saturated, plus cited Benefits & risks (fermented / minerals can score when a recipe publishes that weight). The top place on a list is whoever wins that published mix — not medical advice. Amino-acid scores and “complete protein” marketing claims are not scored axes in V1. Risks do not restate open axes (sodium and saturated fat stay in their columns).

Why we do not write “live cultures” as a blanket label: fermentation can be real while CFU claims are not. Tempeh is Rhizopus-fermented and almost always cooked — heat typically inactivates microbes; literature more often frames cooked tempeh as a paraprobiotic / process food (e.g. Nurkolis et al. 2023). Natto is Bacillus subtilis–fermented and usually eaten without a second cook, so the live-microbe story is stronger — but we still do not invent a yogurt-style CFU count. Both earn a Fermented benefit chip axes miss; natto also carries a cited Vitamin K2 (MK-7) note from the fermentation literature (USDA SR vitamin K on that row is phylloquinone, not a separate MK-7 cell). Lifestyle recipes publish the fermented-process bonus decimals below.

Open The Plant Protein Nutrition Guide (Beans & Tofu).

All cells are per 100 g. Powder concentrates sit far above cooked legumes on protein density by design — they are opt-in via Kind = Concentrates. The default table view is All foods (whole beans and prepared tofu foods — not powders) so protein sort is not topped by shake powders. Kind also splits Whole foods (intact beans and seeds, including edamame and soybeans) from Prepared (tofu, tempeh, and kitchen soy foods). Concentrates stay opt-in. Canned drained solids often carry high sodium next to unsalted cooked rows; that contrast lives in the name and the sodium axis. Tempeh fiber: USDA SR leaves the cell blank. Vital et al. 2018 (PMC6163864) measured AOAC total dietary fiber on lyophilized soybean tempeh at 18.96 g/100 g (16.8 insoluble / 2.18 soluble) with moisture 2.57%. We restore to wet basis with the same USDA SR tempeh water we already use on that row (59.6 g/100 g): 18.96 × (100 − 59.6) / (100 − 2.57) ≈ 7.9 g/100 g. The formula and lab-sample caveat sit on the number’s * hover. A Lightlife wet label (~7.1) stays as a second reading. Crude fiber assays (~1–3 g) under-count and are not used.

Concentrates stay in the catalog so the numbers remain inspectable. This shelf is still Beans & Tofu — not a supplement aisle. All six lifestyle doors hard-gate concentrates, deep-fried tofu, and UPF meat analogs. Axis charts can still show concentrates via Kind; the top place on a plate list will not be a shake powder.

Soy milk (unsweetened) is in — primary row is brand-named Silk (USDA SR), which reports fiber; a generic shelf-stable unsweetened assay sits as Also reported (fiber often 0). Almond, oat, and other plant milks stay off this shelf. Per 100 g it is a dilute drink, so protein density will not match tofu.

Beyond Burger and Impossible Burger are in as branded meat-analog snapshots (USDA Branded Foods) so people can inspect the numbers next to beans and concentrates. They carry an ultra-processed analog caution; recipes churn, so the FDC date matters. Everyday plate lifestyle ranks should hard-gate them — this shelf is still Beans & Tofu, not a fake-meat aisle.

/plant-proteins/for/* uses weighted min–max scores (0–100) on the filtered list. Ultimate opens on Everyday protein. Chips: Everyday protein, Keto & low carb, Lean macros, Whole & ancestral, Gut & microbiome, Heart healthy. Decimals below are editorial — published here. Not medical advice.

Hard gates (most doors): kind = concentrate; risk id deep-fried; risk id ultra-processed-analog. Keto & low carb also unranks when net carbs (max(0, carb − fiber)) > 10 g/100 g. Whole & ancestral omits non-whole kinds before scoring (Everyday weights). Fermented metric: natto 1.0, tempeh 0.65, else 0 — process bonus, not a live-CFU claim. Everyday / Whole may add a tiny mineral bonus (≤ 0.02 on the 0–1 sum before ×100).

Smoke cells are method-cited lab readings (usually AOCS Cc 9a-48). Brand labels and wiki temperature lists do not overwrite a lab primary. Rule: cite + method beats a chart without a method. Empty stays empty until we have a reading we will stand behind.

Editorial inventory of empty smoke cells and oils / sweeteners still waiting on composition cites — not a ranking rule.