Empirica Logic · Published Methodology
And why we publish the method rather than describe it.
Most medical content packages are assembled the way you'd expect: someone reads a research summary, writes around it, and ships. The problem isn't laziness. It's that errors propagate silently. A statistic gets miscopied from a digest, that digest feeds an article, the article feeds a social post, and six months later a clinician is quoting a number to a patient that doesn't appear anywhere in the study it's attributed to.
We've found that exact chain in circulation more than once. A widely shared figure claiming a 33% cancer-risk reduction, traced back, came from a preprint that had since been published with different numbers and in a different journal. A clot-risk statistic circulating as “420%” doesn't reconcile with any published analysis. These weren't invented maliciously. They were copied faithfully from something that was already wrong.
The method below exists to make that specific failure impossible. We publish it for the same reason we tell the clinicians who use our kits to publish their sources and name their own limitations: a process you can't inspect isn't a process, it's a claim.
Every kit is built on a source a reader can open for free — a public-domain government document, an open-access journal article, or a paper with free full text on PubMed Central. The direct link appears on the first page of every kit.
We never bundle the paper. Journal articles are copyrighted, and free-to-read is not free-to-redistribute. That applies to abstracts as well as full text. Every summary in every kit is original prose describing the findings.
Occasionally a source is important enough to include despite being paywalled. When that happens the kit says so explicitly and links to what is available, rather than quietly presenting a summary as though the reader could check it.
Each kit is generated from a single structured record holding the citation, the study design, every number the kit is permitted to state, what the source does not support, the position we think the evidence supports, and the limitations.
Every asset — the article, the patient handout, the social posts, the schema block — is a projection of that record. Nothing is written directly into an asset.
This sounds like an implementation detail. It is the whole safety argument. A figure is entered once and every asset is generated from it. Nothing can drift, because nothing was ever typed twice.
A clinician writing under their own license is entitled to interpret published evidence and reach conclusions mainstream consensus hasn't adopted. That is normal medicine. What they are not entitled to do is present an interpretation as an established finding, or turn either one into an advertising claim.
So every claim in a kit carries a marker answering one question: who owns this sentence?
| Marker | Who owns it | Example |
|---|---|---|
| Settled | Nobody — it's simply true | “Estrogen is the most effective treatment for hot flashes.” |
| Sourced | The study | “A 2019 review of 36 randomized trials found testosterone improved desire.” |
| Your Call | The clinician | “My conclusion is that we frightened women about the wrong molecule.” |
| Wrong Question | The clinician, disputing the study's conclusion | “That trial tested a different molecule in a much older population than the one it's cited about.” |
| Open Question | Nobody yet | “The evidence here genuinely conflicts.” |
Settled and Sourced claims can appear in advertising and structured data. Your Call and Wrong Question require the clinician's byline. Open Question is educational only.
It's the marker most often done badly. It means accepting a study's data and rejecting its conclusion, on the grounds that the study answered a different question than the one it is cited for — wrong molecule, wrong route, wrong population, wrong endpoint. That is a construct-validity critique, a standard and respected form of scientific argument.
It is not the same as saying a study was fraudulent or fabricated, and our kits never make that move. The rule we apply: the flaw has to be documented by someone other than us, ideally by the study's own authors or by a regulator. A premise critique that rests on assertion isn't a critique.
Two fields are mandatory in every record, and a kit cannot be generated without them.
We require these because content that only cites favorable evidence reads as marketing, and because a clinician who names the weakness in their own best study is more persuasive than one who doesn't. It is also simply more honest. If a kit has no counter-evidence entry, someone has to write down why. It cannot be left empty by default.
Kits aren't proofread into compliance. They're gated. Before a single asset is generated, the record is checked. Then every generated asset is checked again. The build fails — producing nothing — on any of:
Here is the gate rejecting deliberately bad drafts during development:
post 1: REJECTED - untraceable number "33%" post 2: REJECTED - prohibited phrase "the FDA says hormone therapy is safe" post 3: REJECTED - inference not flagged as requiring a byline post 4: KEPT
That first line is the point. The 33% figure is a real number in real circulation, attached to a real study that doesn't contain it. A human reviewer misses that. A reviewer who has read the same digest twenty times misses it reliably. The gate doesn't.
Some kit components — social post variants in particular — are drafted with AI assistance, because twenty posts written from templates read like twenty posts written from templates.
The model is given only the approved findings from the record and instructed that it may not introduce any figure outside that list. Then every post it produces is checked against the record anyway. Posts that fail are regenerated once and dropped if they fail again. Nothing reaches a kit unvalidated.
Source selection, claim verification, the assertion markers, the concessions and the clinical positioning are human work. We would rather say that plainly than imply a level of automation we don't have, or hide a level we do.
In the spirit of the concession we require in every kit.
It makes a kit traceable. Every claim can be followed to a source and checked. That is a meaningfully different and more modest guarantee than accuracy, and it is the one we can actually make.
The positions in our analyses are ours, clearly marked as ours, and reasonable clinicians disagree with some of them. The markers exist so you can tell which parts are findings and which are argument — including when the argument is ours.
The markers reflect editorial judgment about evidentiary strength, not a determination of what any particular advertising authority or licensing board will accept in any jurisdiction. Clinicians remain responsible for what they publish.
A kit is a research and communication tool. It isn't a protocol, a guideline, or a treatment recommendation.
A kit reflects the literature at the time it was written. We don’t monitor past kits for new evidence, and every kit links to its source so you can check it before republishing.