A biomarker is not an outcome.
“Raises NAD+” or “improves an aging clock” is not “you live longer or healthier.” We say which one the evidence actually shows.
The method
How we check.
The verdict comes first, the evidence sits under it, and the limits get named out loud. Confident about the method; humble about what it can’t tell you.
The Caveat Standard
Five steps, every time.
- 01 · Source
Trace the claim to its primary source. No source, no verdict.
- 02 · Quote
Quote it exactly, what was said, not the paraphrase.
- 03 · Grade
Grade the evidence GRADE-style: A high → D very low.
- 04 · Conflicts
Follow the funding, who paid for the study, and the claimant.
- 05 · Caveat
Name the limit, and whether it's untested or disproven.
The verdict
Five verdicts. One stays apart.
Unsupported
No credible evidence supports the claim as stated.
Misleading
A kernel of truth, framed to imply far more than the evidence shows.
Mixed
Partly supported, partly not, it depends on the specifics.
Supported
The claim matches solid human evidence.
Unproven
Not shown, but not disproven. Plausible, and untested in humans. An honest grade, not a failing one.
Evidence grade
GRADE-style certainty.
A
High certainty
B
Moderate certainty
C
Low certainty
D
Very low certainty
The verdict says whether the claim matches the evidence. The grade says how far the evidence itself can be trusted, downgraded for small samples, short trials, surrogate endpoints, weak design, inconsistency, and funding conflicts.
Principles
Six we repeat.
Mechanism is not proof.
A clean story about why something should work is the beginning of a question, not the answer. Plausible is labelled plausible.
Follow the funding.
Industry-sponsored research favours the sponsor. We report who paid for each study and weight independent replication higher.
Absence of evidence ≠ evidence of absence.
Unproven is not disproven. We're as skeptical of confident dismissal as of confident hype, and we explain why a trial may never exist.
Never fabricate a citation.
Every source is opened and verified; quotes are exact. For a claim-checker, a single invented reference would be fatal, so when in doubt, we cut it.
You are not the average.
Studies are essential, and they are the backbone of every verdict. But evidence describes populations, not people. What works for most can do little, or the opposite, for any one person. We give you the strong prior, never a personal prescription.
The pipeline
How every check is built and verified.
Every claim runs through a four-stage pipeline before a human editor signs it off. We use AI agents for the legwork, research, source auditing, drafting, and an adversarial fact-check, but the stages are kept deliberately separate: each is a different agent with one job, and none trusts the one before it. Niko reviews and signs off every check; until then it stays in-review.
- 01 · Research
Trace the claim to its primary sources, PubMed, Cochrane, regulators, and record each study exactly: type, size, duration, and who funded it.
- 02 · Audit
A separate, independent pass re-opens every cited URL and confirms each one loads, says what we claim, and carries the funding we report. Sources that fail are cut.
- 03 · Write
The check is composed from audit-approved sources only, with the verdict and grade the evidence supports, never the other way round.
- 04 · Fact-check
A final adversarial pass tries to break the verdict and re-checks every citation, author by author, against the NIH NCBI database to catch a mis-cited or invented reference.
Sourcing is the part we are strictest about. Every PubMed or PubMed Central reference is checked author by author, by title, journal, year, volume, and page, against the NIH’s public E-utilities database, because a finding can be real while the citation behind it is wrong, and for a claim-checker a single invented reference would be fatal. We run that same adversarial pass over the whole back-catalogue, not only new checks. Anything that cannot be independently re-opened and matched is cut, and every correction is logged.
Independence is the product. No industry money, ever.
We’re funded by readers and by licensing the database to clinics and employers, never by the companies whose products we check. If that ever changes, this page changes first, in public.