Methodology & evidence standards
Every Commerce Lab publication states what its data can support, and stops there. This page defines the labels, the anonymization policy, and the correction policy that govern all of them.
The four evidence labels
Each publication discloses its primary evidence label near the top, adjacent to its headline claim, so an extracted quote carries its own qualification.
Based on publicly available marketplace information.
Based on private account or campaign data BGIQ accessed directly.
Produced through a deliberately designed experiment with a stated hypothesis, measurement plan and comparison.
Aggregated across multiple businesses or datasets, with sample, denominator and selection method disclosed.
Nothing is labeled “verified” without naming who verified it and how. Observational data is never described as an experiment, and a single case is never generalized into a benchmark.
What every publication must contain
Measurement periods, comparison design, inclusions and exclusions are disclosed in a metadata panel at the top of each study. Raw counts and denominators accompany percentages wherever confidentiality allows. Measured findings, BGIQ interpretation, and acknowledged unknowns are kept in separate, labeled sections. Confounders that could explain a result are named in the body — not in a footnote — and when the data cannot support a conclusion, the conclusion is narrowed or changed.
Anonymization and permission
Studies using account data under BGIQ management are published only with the account holder’s permission. Before publication, brand names, product identifiers, product types, exact revenue and unit figures, calendar dates, ingredient details and pack formats are removed. Sensitive absolute figures are indexed to a pre-period baseline of 100; counts are reported raw only where they do not identify the account. Underlying data is confidential and not available for redistribution.
Correction policy
Findings are dated snapshots and studies are not updated as the underlying accounts continue to trade. If an error is identified in a published figure or claim, a correction is issued on the study page itself, marked and dated.