The Standard.

01 · The record is failing the subject

Outdated before it is published, and built on the wrong people.

Most academic research on patients using AI is outdated before it is published. Peer review runs on years. The tools patients use change in months.

The methods compound the lag. Surveys of attitudes stand in for observation of practice. Small, clinic-bound samples stand in for a global movement. Studies draw conclusions about models replaced twice since the data was collected. The result is a literature that describes a world that no longer exists, cited as though it does, and used to set policy for patients it never actually observed.

The defect runs deeper than speed. Consider the type specimen: a large randomized trial concluding that giving people AI does not improve their medical decisions — where the participants were crowdsourced workers paid a small sum to role-play someone else's symptoms from a script. No lived experience of the condition. No personal history. No fear, no stakes, no real question needing a real answer. From this, conclusions about patients. A person role-playing a headache for pocket change and a person awake at 2 a.m. afraid they are having a stroke are not interchangeable research subjects. A literature built on the second tells you nothing reliable about the first.

02 · What we do instead

Observe the real thing, at the speed it moves, with methods anyone can inspect.

Everything starts from primary sources: the code patients publish, the appeals they win, the data they collect, the papers themselves, read line by line. We observe daily rather than retrospectively. And every analysis runs through named frameworks and instruments with published criteria:

• CAIHL (Critical AI Health Literacy), from Hugo Campos and Liz Salmi's National Academy of Medicine commentary, asks the first question of every analysis: who does this AI serve? It separates institutional AI from patient-directed AI and judges every system by whether it expands or constrains patient agency.
• CLAIM (Contextual Literacy for AI in Medicine) defines the competencies a patient needs to judge AI output: grounding answers in personal context, interrogating rather than accepting, integrating across sources, activating judgment before action, and transferring those habits to the next question.
• REAL-PAIR rates the ecological authenticity of patient-AI research across five dimensions. It answers one question: did this study of patients using AI involve patients using AI?
• ASSAY extracts and grades the claims of any article, separating what a paper demonstrates from what it asserts.
• CLAIM-CSN audits the citation network behind a claim, detecting manufactured authority: bias, amplification, diversion, and the cascades that turn one weak finding into apparent consensus.
• PEER-REV evaluates papers for integrity, statistical validity, and patient utility, calibrated to what kind of claim a paper makes.

Every claim carries its date and its sources. When the field moves and something here goes stale, you can watch it go stale. When we get something wrong, the correction appears where the error appeared, dated, with the original preserved. The verdicts live in the Casebook, on the record.

03 · What this is not

Not anti-science. The opposite.

This is not anti-science. It is the opposite. It is patient-directed science, because patients know best what it is to be a patient and what needs to be studied, without the filter of professional bias. The patients this site serves depend on rigorous evidence more than anyone, because they are navigating without the safety net of regular care. They cannot afford a literature of vibes, whether the vibes run against AI or for it.

It is also not a claim that everything here is right. It is a claim that everything here is checkable. That is the difference between authority and evidence, and it is the entire point.

04 · The commitment

Better science, done in the open.

The reality of patients using AI deserves analysis as rigorous as anything in a journal and faster than any journal can move. When the published science is poor, the corrective is better science done in the open. Not lower standards with better intentions. Better work, on the record, where anyone can audit it.

That is the standard. Hold us to it.