Research publication standards

IndexFair Research exists to make IndexFair the originating source of facts about operational behaviour in regulated digital commerce — not a blog, not content marketing. Three properties come before anything else: we publish findings, never other people's data; every published number is independently obtainable by a competent third party; and we publish our own limits before our own results. This page is the public version of the internal standard every study is checked against — the strongest asset the programme has, made visible.

IndexFair never publishes a finding that only one source could have produced. That single sentence is the standard. Operationally it is two tests, and a published claim must pass both.

Corroboration. Every published quantitative claim is supported, within a stated tolerance, by at least two independent provenance families — a different rights-holder AND a different adjacency group (e.g. Apple App Store and Google Play count as one family, not two). A claim that survives on one family only is suppressed or widened (a coarser bucket, a longer window, a higher level of aggregation) until it corroborates — never published “with a caveat”.

Independent obtainability. Could a competent third party with lawful access obtain this finding? If the only route to a number is having scraped one particular source, the number belongs to that source's data, and no amount of aggregation changes that. This is a resilience and evidential property, not a legal defence: no single platform can shut down a published finding by cutting us off, and no single platform can credibly claim a finding is theirs.

Two narrow exemptions exist. First-party data — measurements IndexFair produces itself — is exempt from the corroboration test because we are the rights-holder, but must instead pass an instrument-disclosure test: the measurement method is published in enough detail that a third party could build the same instrument and replicate the result. Statutory registers published by a regulator are exempt for the same reason — the register is the population, not one source's private data — but must pass a fidelity test: the published figure is reproducible from the named public release, with the release URL, the retrieval date and a file checksum stated.

Every study is exactly one class. The class fixes what data, if any, can accompany the finding — decided before a word of the study is written, never after.

Class D (derived from an encumbered third-party review corpus) may publish numbers, method, limitations and charts — never the underlying corpus, never a per-review row, never a dataset DOI. A file-less report DOI is permitted, because a persistent identifier on a findings document that redistributes no data re-utilises nothing. Classes R, M and V may release the derived dataset itself, under a stated open licence, because the underlying rights are clean: a statutory register, our own first-party instrument, or our own score history. Class O (opinion / explainer) carries no data release and is never called “research” — that content lives at /insights.

Every study is checked against this list before it goes live:

• Every headline number traces to a query on a frozen, re-runnable data snapshot — never a live table that could answer differently tomorrow with no explanation.
• The Two-Source Rule verdict (or the applicable exemption) is recorded per headline claim.
• No published bucket falls below the minimum group size that protects against re-identifying an individual source or complainant; no raw source-attributed volume is published.
• Denominators are stated everywhere; an attempt count is never presented as a record count.
• The limitations section is written before the findings section.
• Nothing characterises a named entity without a right-of-reply window having run first.
• No forward-looking probability is published about a named brand.
• A study page on indexfair.com is the canonical, on-site home of its own finding and may state a value with its basis. Anything that leaves the site — a chart-pack image, a social post, a press quote, or any asset built to be embedded on a third-party page — carries an unlabelled band, never a value, no matter who serves it or requests it. The test is whether the asset can be placed on another page, not what we intended when we built it.
• No marketing lexicon (superlatives, “best”, “leading”) anywhere in the piece.
• The class-appropriate distribution ladder is followed; a class-D study carries no data dump and no dataset DOI.
• The corrections policy and a named contact appear on the page.

Every study is reviewed against this checklist and signed off by the founder personally before it publishes — see /founders for the standing editorial-governance disclosure that applies to every publication on this site, research included.

This standard governs everything published under /research and any downloadable data at /datasets. No study exists under this standard yet — this page describes the rule a first study must pass, published in advance of that study rather than after it, so the rule cannot be read as written to fit a result.

Corrections policy: /policies/corrections. Press & media: /press. Methodology: /methodology.