Preprint

Extending the weighed-food reference meal set to restaurant-served meals: protocol and pilot data

DAI-PRE-2026-02

Sofia Rivera, MS, RD; Helena Weiss, PhD, MPH, RD
Published March 27, 2026

This is a preprint

This article is a preprint and has not undergone external peer review. The Dietary Assessment Initiative releases preprints to invite methodological critique prior to or alongside formal publication. This preprint is released alongside the Restaurant Pilot Meal Set (N=32) dataset descriptor as a companion publication.

1. Background

Weighed-food reference sets built in a metabolic kitchen are the gold standard for validating the energy estimates of dietary assessment applications. Every ingredient is weighed in grams before plating, each ingredient is cross-referenced against a controlled food composition database (in our case, USDA FoodData Central Foundation Foods), and the per-meal kcal ground truth follows deterministically. The Initiative’s Weighed-Meal Reference Set v1.0 (mini-180) is built this way.

This design has a known external-validity limit: many of the meals that users of dietary tracking applications log every day are not prepared in a metabolic kitchen. They are served in restaurants, ordered from delivery platforms, or assembled at cafeterias. When a validation study’s accuracy figures are derived only from home-style metabolic-kitchen meals, those figures may not transfer to the restaurant-meal case.

This preprint describes the protocol, decision rules, and pilot data for extending the reference set to restaurant-served meals. We take as a premise that a perfect gold-standard kcal value is not achievable for restaurant meals in the way it is for metabolic-kitchen meals; the best we can do is produce a reference value accompanied by an explicit uncertainty interval, and to report that uncertainty transparently alongside any downstream accuracy comparison.

2. Methods

2.1 Two-path protocol

We define two paths for obtaining a reference kcal value for a restaurant-served meal.

Path A — Published-nutrition reference (with verification). For restaurants (typically chains) that publish per-menu-item nutrition information, the published kcal value is adopted as the reference. To guard against portion drift, each served meal is dismantled and reweighed in a nearby prep area under controlled conditions, and if the dismantled served weight deviates from the published serving weight by more than 10%, the meal is flagged and the reference kcal is scaled by the weight ratio.

Path B — Recipe reconstruction. For restaurants (typically independents) that do not publish nutrition information, the meal is dismantled and reweighed in the prep area; a research dietitian reconstructs the recipe from the menu description, visible ingredients, and the dismantled weights; kcal is computed against USDA FDC Foundation Foods; a conservative uncertainty half-width is assigned based on the number of reconstruction assumptions required.

2.2 Pilot

Thirty-two meals from 14 restaurants (9 chain, 5 independent) in one metropolitan area were sampled during February 2026. Eighteen meals were eligible for Path A; 14 were routed to Path B.

3. Results

3.1 Path A outcomes

For the 18 Path A meals, dismantle-and-weigh verification found portion-size discrepancies exceeding plus/minus 10% in 6 of 18 (33.3%). In four cases the served portion was larger than published; in two cases it was smaller. Scaling the published kcal by the weight ratio produced a final reference value with an implicit uncertainty half-width of approximately plus/minus 4.0% (driven by residual composition uncertainty within the scaled portion).

3.2 Path B outcomes

For the 14 Path B meals, the reconstruction procedure produced a kcal estimate with an assigned uncertainty half-width of approximately plus/minus 12.4% on average (range plus/minus 7.2% to plus/minus 18.9%). The largest uncertainty values were associated with meals where the cooking-fat component could not be directly observed (e.g., deep-fried items).

3.3 Comparability to metabolic-kitchen reference

The per-meal uncertainty in the Restaurant Pilot Meal Set is substantially higher than in the metabolic-kitchen reference (where per-meal uncertainty is approximately plus/minus 2-3%, driven mainly by FDC entry matching). Any downstream application-accuracy analysis conducted against the restaurant extension will therefore need to propagate this additional uncertainty rather than treat the reference as fixed.

4. Discussion

Extending the reference set to restaurant-served meals is feasible but requires explicit acknowledgement that the reference is an estimate with uncertainty, not a gold-standard truth. The headline finding of the pilot — that a third of chain-restaurant portions deviated from published portion size by more than 10% — is consistent with previously reported restaurant portion variability and argues against uncritical adoption of chain publications as a reference.

The protocol above is, we believe, the minimum that can defensibly be called a reference for restaurant-served meals. We welcome methodological critique before scaling to the full N=200 extension planned for 2026-Q4.

Limitations: the pilot was conducted in one metropolitan area, which may not generalise; recipe reconstruction in Path B is inherently dependent on the reconstructing dietitian; dismantling in the prep area depended on restaurant cooperation and may have introduced selection bias.

References

1. Arroyo P, Beckman F. Restaurant portion variability: a systematic review. Public Health Nutr. 2022;25(11):3100-3112.
2. Bryant C, Dolan K. Verifying chain-restaurant nutrition disclosures: methods and findings. J Acad Nutr Diet. 2023;123(8):1120-1128.
3. Chaudhuri S, Ellis P. Recipe reconstruction for food composition analysis. Food Chem. 2021;362:130198.
4. Duvall N, Fiorella G. External validity in dietary assessment validation. J Nutr. 2024;154(3):612-620.
5. Gardner M, Hollister B. Uncertainty propagation in nutritional ground truth. Br J Nutr. 2023;130(5):801-810.
6. Henderson A, Iannucci T. Dismantle-and-weigh protocols: a practitioner’s guide. Nutr Methods. 2023;16(4):210-219.
7. Juárez E, Kopecky V. Restaurant menu kcal disclosures and served portions. Am J Prev Med. 2024;66(2):238-247.
8. Linder G, Matthews O. Extending weighed-food reference sets to out-of-home eating. Nutrients. 2024;16(12):1844.
9. Nader F, Okoye B. Deep-fried item uncertainty in restaurant recipe reconstruction. Int J Food Sci Nutr. 2024;75(7):701-712.
10. Pillai D, Quiroga L. Methodological transparency in restaurant meal validation studies. Appetite. 2023;184:106581.
11. Rawlings T, Schmidt E. Chain vs independent: portion variance patterns. Public Health Nutr. 2024;27(3):481-490.