Definitions before comparisons.

Data classification

Product-label data

Information checked directly on a package or reliable package image.

Third-party database data

Product information retrieved from an external database. It is not treated as direct label verification.

User-reported PerBiscuit data

A portion submitted for a specified product by a contributor.

Survey or research data

Results from a cited study or survey, kept within that source's population and design.

PerBiscuit calculation

A value derived from stated inputs and a published formula.

PerBiscuit assumption

An input chosen for analysis without observational support.

Illustrative example

A worked example used to explain a method, not to describe behaviour.

Portion rescaling

Inputs must use comparable units. The multiplier is the selected quantity divided by the label serving. Each nutrition value for the label serving is multiplied by the same number.

Rescaled value = label-serving nutrition × (selected portion ÷ label serving)

Display values may be rounded, but ranking and threshold calculations use unrounded values. If units cannot be converted reliably, PerBiscuit does not calculate a comparison.

Illustrative scenarios

Assumed quantities are labelled in the heading, table and interpretation. They are not combined with user reports and are not used to rate a product or company.

Pack-to-serving ranking

The current US ranking searches a declared set of food and drink terms in the US region of the FatSecret Platform API. Returned product IDs are deduplicated.

A record is eligible only when it has a product name, a named serving, a numeric serving amount and a parsed pack amount in the same dimension. The pack must be larger than the serving.

Pack-to-serving ratio = parsed pack amount ÷ label serving amount

Eligible records are sorted by the unrounded ratio, highest first. Data completeness and then product name break exact ties. The page reports search-term count, unique retrieved records, eligible records, displayed records and retrieval date.

The search is purposive and is not a census. The ratio does not represent a reported portion or intended use of the pack.

Serving Match Score

This proposed score is not published for any product yet. Let L be the label serving and M the median valid reported portion, converted to comparable units.

Serving Match Score = 100 × min(L ÷ M, M ÷ L)

Eligibility requires at least 30 valid reports from at least 20 distinct sessions across at least 30 days for the same product variant and market. Duplicate, automated, mismatched and non-convertible reports are excluded.

A 95% bootstrap confidence interval for the median will be translated into score limits and displayed. Products with overlapping score intervals will be shown as tied. No leaderboard is published until at least ten products qualify.

Provisional award

The proposed US award uses the median reported portion divided by the label serving. It applies the same per-product report threshold as the Serving Match Score and requires at least ten qualifying products.

A sole winner is declared only when the leading product's lower 95% interval bound is above the runner-up's upper bound. Otherwise the outcome is tied or inconclusive. No 2026 winner has been declared.

Sugar-teaspoon display

Where shown, PerBiscuit divides sugar grams by 4 to produce an illustrative teaspoon equivalent. This is a display assumption, not a measured spoon volume or dietary recommendation. The gram value remains the primary quantity.

Display bands

The homepage no longer labels nutrient amounts low, medium or high. It displays sourced or calculated quantities without implying a dietary threshold.

Regulatory explainers

Regulatory articles link important claims to current primary sources. PerBiscuit states the evidence-check date and distinguishes the rule from its interpretation. These explainers are not legal advice.

Version and review

Method version: 1.0. Published and checked July 14, 2026. Independent statistical review of the Serving Match Score and award method is pending. Next scheduled review: October 14, 2026.