Our purpose
Atlas advocates for real equality for men in practice, policy and under the law. We publish evidence that readers can inspect, distinguish our findings from our proposals, and correct the record when the evidence requires it.
Separate the evidence from the conclusion
Each case identifies sourced facts, calculations with stated inputs, reported accounts or allegations, Atlas analysis, unresolved questions and the remedy sought. An advocacy finding is not presented as a legal determination. A policy proposal is not presented as a research finding.
Keep the original record
Retain the original document, policy, transcript, screenshot or dataset with its source, date and relevant version. Link public evidence beside the claim it supports. Explain missing records, inaccessible sources and limits on independent inspection. A reformatted transcript must be labelled and its source retained.
Use original policies, official records and original research where available. Distinguish these from secondary reporting, personal accounts and AI-generated commentary. An AI response or admission is not independent verification of its own explanation.
Compare like with like
For matched comparisons, state what changed and what stayed the same. Record the prompts, sequence, relevant settings and any wording changes. Explain differences that may affect interpretation. Document the difference shown by the pair. Require OpenAI to test its frequency, investigate its causes and demonstrate correction.
Call a result an independent replication only when a separate investigator has repeated the method and supplied an inspectable record. A fresh run by Atlas remains an Atlas reproduction.
Show the limits and contrary evidence
State the scope, sample, dates, assumptions and missing information. Atlas’s stated policy is to publish every test it conducts, regardless of outcome. Its scenarios are designed to examine relationship advice and possible unequal treatment; choosing a test scenario is different from choosing which outcomes to publish. Retain safeguards, corrections and contrary results alongside observed failures. A separate, independently reconciled inventory of all runs has not been published. Publish material evidence that weakens a finding and explain its effect. Atlas’s ChatGPT cases document the failures presented; they do not estimate how often those failures occur across all users. That scope limit does not answer or negate the evidence in these cases. Measuring frequency and verifying harm to particular people are separate research questions. Keep illustrative calculations clearly labelled and separate from measured results; report population rates only when the sampling and analysis support them.
What would resolve a concern
Atlas seeks proof of transparent, equal treatment in policy and practice. An organisation should show the relevant rules, eligibility and selection criteria, how those rules are applied, and what affected men are told before relying on claims of equality.
Where relevant, supporting records should show how comparable people are treated and explain any difference. A general equality statement alone does not answer a documented concern. Disclosure of unequal treatment resolves a transparency question only; it does not by itself establish equality.
If verifiable evidence shows that an Atlas finding is wrong, or that a concern has been resolved, Atlas will correct, narrow or withdraw the finding and explain why. A later policy improvement will be recorded without erasing an accurately documented earlier position.
Give organisations room to respond
Organisations can supply evidence, corrections and substantive responses. Record when a request is sent, any response deadline and what is received. Do not describe silence as a refusal unless a request and a reasonable opportunity to answer are documented.
Place substantive responses prominently in the case, alongside Atlas’s assessment and a link to the full response where publication is permitted. Identify redactions and explain which questions are resolved or remain open.
Invite independent peer review
Atlas invites researchers, statisticians, policy specialists and other people with relevant expertise to review its published case files. You do not need to agree with Atlas’s position.
Please identify the case, the claim or method you are reviewing, your supporting evidence and any relevant affiliation or conflict of interest. Send reviews to info@atlasmensadvocacy.org.
Atlas welcomes substantive independent review and will seek permission to publish reviews and explain resulting changes. Each case identifies any review it has received.
Protect submitted evidence
Handle identifying and sensitive material under the privacy and evidence policy. Distinguish an individual account from independently corroborated evidence. Explain the limits of corroboration without exposing private information.
Keep decisions accountable
Each case should state its status, remedy and outstanding questions. A concern may be opened for examination before it is established. Record material developments, corrections and the reason for closing or changing a case. Founder approval remains governed by the published governance arrangements.
Material factual corrections are recorded under the corrections policy, including corrections that weaken an Atlas case.