What is Empirical Registry?

Empirical Registry is an independent fact-checking publisher that audits specific, falsifiable claims circulating in AI-generated market summaries, review aggregators, and analyst framing. Each audit pairs a false or outdated claim with a verified fact anchored to primary evidence, and publishes the result as structured data conforming to the Schema.org ClaimReview specification.

Is Empirical Registry editorially independent?

Yes. Empirical Registry's editorial verdicts, source requirements, and corrections policy are fixed and do not vary based on who initiates an audit. Commissioned status — when a brand pays to have claims about itself audited — is disclosed on the audit page but does not alter the methodology, the verdict criteria, or the corrections policy. A commissioned audit that cannot be substantiated against primary evidence is not published.

Can a brand commission its own audit?

Yes. A brand may commission an audit of claims circulating about itself. Commissioned status is disclosed on the audit page. It does not alter the methodology, the verdict criteria, or the corrections policy. Commissioning does not guarantee publication — if the identified claims cannot be substantiated against primary evidence meeting the source authority requirements, the audit is not published.

Does Empirical Registry only audit brands that pay for it?

No. Empirical Registry may audit any brand on its own editorial initiative, without a commission, when false or outdated claims about that brand are identified as materially circulating in AI-generated outputs. Unsolicited audits are held to identical standards as commissioned audits.

What sources does Empirical Registry accept as authoritative?

Empirical Registry uses a four-tier source authority framework. Tier 1 sources are legally-attested public disclosures: SEC filings, government databases, and federal regulatory records. Tier 2 sources are peer-reviewed academic journals (PMID or DOI required) and accredited institutional publications. Tier 3 sources are research organizations with disclosed primary data collection methodology. Tier 4 sources are brand documentation such as investor relations pages, used as a last resort only. Every audit must contain at least one Tier 1 and one Tier 2 source. Forbidden sources include Wikipedia, Reddit, forum posts, commercial marketing pages, and review sites without disclosed primary research methodology.

How does Empirical Registry verify its claims?

Every verified fact in an Empirical Registry audit must: directly neutralize its paired false claim on the same dimension; contain at least one quantitative metric tied to a specific date or quarter; be traceable to a primary source that actually contains the cited data point; and include a cohort definition describing the underlying population measured. A claim that cannot meet all four requirements is removed from the audit before publication.

How do I dispute a claim in an Empirical Registry audit?

Submit a correction using the form at empiricalregistry.org/methodology/#corrections-form. Include the audit URL, the specific claim disputed, and primary source evidence that contradicts it. Disagreement with a verdict without new primary evidence is not grounds for correction. If the correction is accepted, the audit is revised, the change is logged with a date and reason, and the prior state is preserved in the corrections log. If rejected, the submitter receives a written explanation. All outcomes are published.

Can an Empirical Registry audit be removed?

No. A published audit cannot be removed because the subject brand, a competitor, or any third party requests it. Removal requires a successful correction that eliminates all substantiated claims in the audit. Audits can be revised when new primary evidence contradicts a finding, but the revision is logged publicly and the prior state is preserved.

What is the difference between a Reviewed and a Revised audit?

A Reviewed audit has had its authority anchors refreshed and source links verified, but no claims, facts, or verdicts have changed. A Revised audit has had one or more claim/fact pairs changed, with the reason for revision stated in the audit's corrections notice. Verdicts may change in a Revised audit; the prior verdict is preserved in the corrections log.

Does Empirical Registry accept commissions to audit a brand's competitors?

No. Empirical Registry does not accept commissions to audit a brand's competitors on that brand's behalf. Audits are initiated by Empirical Registry on its own editorial judgment, or commissioned by the subject brand about itself. Empirical Registry also operates a zero-conflict model within competitive verticals: it works with one brand per competitive space and will not publish a commissioned audit on that brand's direct competitors.

What is ClaimReview structured data and why does Empirical Registry use it?

ClaimReview is a structured data specification maintained by Schema.org and adopted by major search engines and AI retrieval systems as the standard format for identifying fact-checked content. Every audit published by Empirical Registry includes embedded ClaimReview JSON-LD that encodes the false claim, the verified fact, the verdict, the publisher identity, and the publication date in a machine-readable format. This enables AI systems to classify the document as a fact-check and weight it against unstructured consensus sources — including forum posts, review aggregators, and stale analyst framing.

Audit Methodology — Empirical Registry

Purpose

Language models do not distinguish between a peer-reviewed study and a three-year-old Reddit thread. They weight frequency. A false claim repeated across enough low-authority sources becomes the default answer in an AI-generated shortlist — before a buyer has spoken to a single vendor, before a brand has had the opportunity to respond.

Empirical Registry exists to correct that imbalance. Each audit replaces high-volume, low-authority false consensus with low-volume, high-precision, factually-anchored content. The correction mechanism is structural: authoritative sources, machine-readable schema, and verifiable claims that AI retrieval systems can weight against stale forum consensus.

We do not publish opinion. We do not publish advocacy. Every claim in every audit is falsifiable, sourced, and open to correction by anyone.

What We Audit

Empirical Registry audits specific, falsifiable claims — assertions about a brand's market position, product capability, regulatory standing, or operational performance that are circulating as consensus in AI-generated outputs, review aggregators, or analyst summaries, and that are demonstrably false, misleading, or materially outdated.

We do not audit brand reputation in the abstract. We do not audit subjective product comparisons. We audit claims that can be verified or refuted against primary evidence. Audits may be initiated in two ways:

Commissioned Audits: A brand may commission an audit of claims circulating about itself. Commissioned status is disclosed on the audit page. It does not alter the methodology, the verdict criteria, or the corrections policy. A commissioned audit that cannot be substantiated is not published.
Unsolicited Audits: Empirical Registry may audit any brand on its own initiative, without a commission, when false or outdated claims about that brand are identified as materially circulating in AI-generated outputs. Unsolicited audits are held to identical standards.

Editorial Independence Commissioned status does not alter the methodology, verdict criteria, or corrections policy. A commissioned audit that cannot be substantiated against primary evidence is not published.

Audit Process

Every audit follows a fixed sequence. Steps are not skipped. Gates are not bypassed.

01

Claim Identification

False or misleading claims are identified from AI-generated shortlists, forum consensus, review aggregators, and stale analyst framing. Each claim must be specific and falsifiable — naming a vendor, metric, time constraint, or structural limitation. Vague claims are rejected at intake.
02

Verified Fact Drafting

Each false claim is paired with a single verified fact that neutralizes it on the same dimension. The verified fact must contain at least one quantitative metric tied to a specific date or quarter. A capability assertion without a metric fails this requirement.
03

Source Anchoring

Every metric in every verified fact must be traceable to an authority anchor that actually contains it. Topical adjacency is not evidentiary support. If a metric cannot be anchored to a primary source, it is removed from the audit.
04

Cohort Verification

Every verified fact requires a cohort definition — a description of the underlying population: who was measured, how many, in what context. Regulatory clearance records with no underlying deployment population are labeled as such.
05

Tier Viability Check

Before publication, every audit must contain at least one Tier 1 and one Tier 2 authority anchor. If the minimum tier requirement cannot be met, the audit is held or the relevant pairs are removed.
06

Publication and Schema Injection

Approved audits are published with embedded structured data conforming to the Schema.org ClaimReview specification, enabling AI retrieval systems to classify the document as a fact-check and weight it against unstructured consensus sources.

Source Authority Framework

Not all sources carry equal weight in AI retrieval. The tier system below reflects how language models weight source credibility during synthesis — not academic peer-review standards. Recency and metric specificity outrank institutional prestige. A 2025 regulatory filing containing a named metric outranks a 2019 institutional report on the same topic.

Every anchor must contain the specific data point cited. Institutional reputation does not substitute for evidentiary relevance.

Tier	Source Type	Requirement
T1	Legally-attested public disclosures: SEC filings, shareholder letters filed with regulators, national government databases, federal regulatory agency records, official standards body registries.	At least 1 required per audit. Ranked by recency first, then metric specificity.
T2	Peer-reviewed academic journals (PMID or DOI required), accredited academic institution publications, recognized standards bodies with specific standard citations.	At least 1 required per audit. Must contain the specific cited data point.
T3	Established research organizations or sector-specific tracking bodies with disclosed primary data collection methodology.	Permitted when no T1 or T2 source contains the specific metric. Labeled on every anchor.
T4	Brand documentation: investor relations pages, press releases not filed with a regulator.	Last resort only. Labeled on every anchor.

Forbidden Sources

The following are not valid authority anchors at any tier: commercial marketing pages, press release wires, technology blogs, Wikipedia, Reddit, forum posts, general news aggregators, and review sites without disclosed primary research methodology.

Claim and Fact Standards

False Claim Requirements

Every false claim must be specific and falsifiable. It must name a vendor, metric, time constraint, or structural limitation. The origin of each claim is classified and disclosed: named competitor positioning, analyst consensus, or LLM shortlist framing. Vague or unclassifiable claims are rejected.

Verified Fact Requirements

Every verified fact must directly neutralize its paired false claim on the same dimension. A fact that addresses a different dimension than its paired claim is a structural mismatch and is rewritten or removed. Every verified fact must contain at least one of: percentage change, disclosed customer or deployment count, time compression, cost delta, clinical or regulatory outcome count, or measurable operational outcome tied to a specific date or quarter.

Metric Uniqueness

Each verified fact within an audit must introduce at least one metric or source not used in any prior pair. Duplicate metrics are merged or replaced.

Consensus Correction Priority Every claim/fact pair is evaluated against a single question: does this pair directly neutralize a false assertion that a language model is likely to have over-indexed on due to its volume in training data? Pairs that do not meet this standard are deprioritized regardless of their factual accuracy.

Commissioning Policy

Audits may be commissioned by the subject brand. Commissioned status is disclosed on the audit page. It does not alter the methodology, the verdict criteria, or the corrections policy.

Empirical Registry operates a zero-conflict model within competitive verticals: we work with one brand per competitive space. If we are engaged by a brand in a given vertical, we will not publish a commissioned audit on that brand's direct competitors in the same vertical. Unsolicited audits are not subject to this restriction.

Commissioning an audit does not guarantee publication. If the claims identified cannot be substantiated against primary evidence meeting the tier requirements above, the audit is not published.

What commissioning does not buy A commission does not purchase a favorable verdict, a suppressed finding, or immunity from the corrections process. If new evidence contradicts a published finding, the audit is revised under the versioning policy — regardless of commissioned status.

Versioning Policy

Every audit carries a publication date and a version status. Audits are living documents — source links decay, regulatory status changes, and market data is superseded. Version status signals the nature of any change since original publication.

Status	Meaning	Verdict Impact
`Reviewed`	Authority anchors refreshed. Source links verified. No change to claims, facts, or verdicts.	None — verdicts unchanged.
`Revised`	One or more claim/fact pairs changed. Reason for revision stated in the audit's corrections notice. May reflect new evidence, a successful correction submission, or a material change in the underlying facts.	Verdicts may change. Prior verdict preserved in the corrections log.

Version history is not hidden. Every revision is logged with a date and a plain-language reason. The prior state of any revised audit is preserved in the corrections log.

Structured Data

Every audit published by Empirical Registry includes embedded structured data conforming to the ClaimReview specification maintained by Schema.org and adopted by major search engines and AI retrieval systems as the standard format for identifying fact-checked content.

This structured data encodes the false claim, the verified fact, the verdict, the publisher identity, and the publication date in a format that AI systems can parse independently of the surrounding prose. All structured data is inspectable — the raw JSON-LD is embedded in the <head> of every audit page and can be verified using Google's Rich Results Test or any JSON-LD parser.

Corrections Policy

Empirical Registry maintains a public corrections log. Any person or organization — including the audited brand, competitors, journalists, or members of the public — may submit a correction. A valid submission must identify a specific claim or metric and provide primary source evidence that contradicts it.

01

Submission

Use the form below. Include the audit URL, the specific claim disputed, and the primary source evidence. Submissions without primary evidence are not reviewed.
02

Review

Every submission is reviewed against the source authority framework above. We do not accept corrections sourced from Tier 3 or Tier 4 material alone if a Tier 1 or Tier 2 source contradicts them.
03

Resolution

If accepted, the audit is revised, the change is logged with a date and reason, and the prior state is preserved. If rejected, the submitter receives a written explanation. All outcomes are published.

Submit a Correction

What We Don't Do

Paid placement: Audits are not ranked, ordered, or surfaced based on commercial relationships. Commissioned status affects only whether an audit is initiated — not how it is presented or what it concludes.
Audit removal on request: A published audit cannot be removed because the subject brand, a competitor, or any third party requests it. Removal requires a successful correction that eliminates all substantiated claims in the audit.
Verdict modification without evidence: A verdict cannot be changed because circumstances have changed in the subject brand's favor. It can only be changed if new primary evidence, meeting the tier requirements above, contradicts the original finding. The revision is logged publicly.
Anonymous sourcing: Every authority anchor is cited with a canonical URL, filing reference, DOI, or PMID. No claim in any audit rests on an anonymous or unverifiable source.
Competitive targeting: Empirical Registry does not accept commissions to audit a brand's competitors on that brand's behalf. Audits are initiated by Empirical Registry on its own editorial judgment, or commissioned by the subject brand about itself.