Bibliography
Sources behind Database v0.1
Every carb, sugar, and glycemic-index figure in RecipeMend traces to a named source. This page lists them, and — just as importantly — lists what is not yet independently verified.
Database v0.0.1 (draft) — updated 2026-07-21
A correction we caught
Primary bibliography
- USDA FoodData Central (fdc.nal.usda.gov) — carbohydrate and sugar values per 100g for ingredients and sweeteners, primarily via mirror-site republications (myfooddata.com, nutritionvalue.org, recipal.com) rather than direct food-detail page fetches for most entries. One confirmed direct FDC ID citation: granulated white sugar, FDC 169655.
- University of Sydney Glycemic Index Database (glycemicindex.com) — individual fruit/sugar GI values, mostly via search-result compilations rather than single direct page fetches per food.
- Atkinson FS, Brand-Miller JC, Foster-Powell K, Buyken AE, Goletzke J (2021), "International tables of glycemic index and glycemic load values 2021," American Journal of Clinical Nutrition, doi: 10.1093/ajcn/nqab233 — citation/abstract level only; the full data table was not independently fetched this session.
- FDA.gov — Guidance for Industry: The Declaration of Allulose and Calories From Allulose on Nutrition and Supplement Facts Labels (final guidance, October 2020).
- Witkowski et al. (Hazen lab), Nature Medicine, Feb 2023, doi: 10.1038/s41591-023-02223-9 — erythritol/cardiovascular-event association, via institutional secondary coverage (Cleveland Clinic newsroom, NIH Research Matters), not the primary paper's full text.
- Monash University FODMAP program — sugar-alcohol serving-size and laxation guidance, via secondary-source synthesis rather than a direct Monash database query.
- Merck Veterinary Manual — "Xylitol Toxicosis in Dogs," cross-checked against Cornell, VCA, and Small Door Vet secondary summaries.
- Named manufacturer baking guidance (explicitly labeled as manufacturer guidance, never presented as independent science): Lakanto, Swerve, Now Foods / Icon Foods, and general Truvia-pattern literature — used for sweetener substitution ratios and bake-behavior notes.
- Romo-Romo et al., AJCN 2018, doi: 10.1093/ajcn/nqy152 — sucralose RCT finding on insulin sensitivity/glucose AUC.
What this database gets right, and how it was checked
An adversarial verification pass ran 2026-07-21: roughly 30 ingredient carb/sugar values were spot-checked directly against USDA FDC and its mirrors, prioritizing flours, sugars/syrups, dried fruit, chocolate, and every near-boundary "low-impact" entry; all 13 sweeteners were audited against their primary sources; all 21 swap rules were checked for correct ingredient-id resolution and for never recommending a swap into an equal-or-worse impact bucket. Eleven corrections were applied as a result — including one bucket-level fix (the cherries correction, detailed next).
The one bucket-level correction that mattered
Cherries were originally bucketed low-impact using a GI of 22 — but that figure belongs to sour/tart cherries. The sweet cherries this database actually describes have a measured GI of about 63±6 (International Tables of Glycemic Index 2021), which moves them to moderate. This was the one provably-wrong "low-impact" classification found in the file — the highest-stakes error class, since a false "safe" label is worse than a false "caution" label.
Corrections we caught before you did
Every one of these was in the draft, and every one is now fixed. Each reads in the same repair grammar as the tool: the wrong value struck out, the sourced value in its place.
Cherries — bucket-changing fix
Sweet cherries carry the sweet-cherry index, not the sour-cherry one — low-impact → moderate.
Vanilla extract — false-zero fix
Its carbohydrate is counted as sugar; the draft's 0g was outright wrong. Stays low-impact at teaspoon use.
Cream cheese — bound fix
The draft's lower bound excluded the true USDA value; bucket unchanged.
Unresolved entries — flagged, not fabricated
About nine entries in Database v0.1 rely on branded or secondary-only sourcing and don't have a clean USDA Standard Reference match. They keep their assigned bucket (which the verification pass confirmed is directionally safe either way) but their exact numbers should be treated as estimates, not precise figures:
- Date syrup and corn syrup — no USDA SR/Foundation reference food exists; values are branded/estimate-based.
- Cassava flour — branded-only; the true value may run higher (~85–88g carbs/100g) than currently listed.
- Potato starch — currently aligned to potato flour figures (~83g); pure potato starch runs closer to 95–100g on branded labels.
- Coconut milk (canned) — no generic USDA entry; branded/label-derived, and USDA's own generic figure shows ~0 sugars versus this database's 2.6–3.3g.
- Maltitol GI compilation values (powder ~29–36, syrup ~52) — from comparative consumer/manufacturer sources, not a primary clinical GI study. See the maltitol page.
- Molasses ("blackstrap") — no USDA blackstrap-specific entry exists; the generic USDA molasses figure (~74.7g sugars) is notably higher than true blackstrap, which is lower-sugar. Bucket (high-spike) is safe regardless of which figure is correct.
- Pastry, raw vs. baked basis — puff pastry and standard pie crust figures reflect the unbaked/frozen product; USDA baked-product values run higher. Both remain high-spike either way, but the exact number depends on which preparation stage a recipe means.
What's genuinely out of scope in v0.1
Database v0.1 covers 88 ingredients and 13 sweeteners — common dessert-baking staples, not an exhaustive food database. Branded products, regional ingredients, and less-common items (liquid stevia drops as a distinct entry, for example) are not yet included. When the tool can't match a pasted ingredient, it says so plainly rather than guessing or silently dropping the line — see How it works.