The Method
How to weigh anyone · Including this publication
A list of people worth reading runs out the moment you meet someone who is not on it. This is the rubric instead — six questions that work on a stranger, and the same ones used to score every source in this corpus.
Nothing here is trusted above it. Not a source, not a study, not this publication. Trusting anyone completely is how you end up in a cult rather than in possession of a view — and a source recorded at 100% is not a well-regarded source, it is a bug.
What kind of claim is it?
A fact, a value, or a prediction. They are scored differently and most arguments are people grading one as if it were another. A value dressed as a fact is the failure this publication exists to avoid.
What tier is the evidence?
For health the usual order is inverted. Someone who has run the protocol on themselves and on patients outranks an institutional position, because the institution carries the strongest incentives and the slowest clock. An endorsement is a fact about the endorser.
What did saying it cost them?
Weight a claim by what it cost the person to make it. Someone who lost standing, income or a licence for saying a thing has paid for the right to be taken seriously. Someone rewarded for saying it has not. This is the single best filter available to an outsider.
What is their record, and how do they handle being wrong?
Right about eighty per cent of the time is the bar. Nobody honest clears it every time. What separates a source worth reading is what happens on the twenty: correcting fast counts for them, never correcting counts hard against them.
Did they delete anything?
Erasing a failed prediction is worse than making one. A wrong call left standing is evidence you can use. A deleted one is evidence about the person, and it is worse.
Who disagrees, and cap it at eighty.
Find the strongest person who disagrees and read them. If you cannot name one, you have not finished. And nothing goes above eighty per cent — agreeing with anyone completely is how you end up in a cult rather than in possession of a view.
What the numbers are
Published so you can apply them yourself, and so you can tell when this publication has departed from them.
Evidence tiers, health
Inverted on purpose. The people carrying the strongest incentives and the slowest clock are weighted lowest.
| H1 | 1.00 | Practitioner report / operator experience |
| H2 | 0.85 | Community convergence |
| H3 | 0.80 | Animal / mechanistic study |
| H4 | 0.45 | Human trial |
| H5 | 0.10 | Institutional / regulatory position |
Trust bands
| 75% | Relied on | Still wrong often enough that disagreement must be recorded. |
| 60% | Strong | Weigh heavily. Verify anything load-bearing. |
| 40% | Useful | One input among several. |
| 20% | Watch | Interesting. Unverified. |
| 5% | Logged | Present for completeness only. |
Track record
| 80% | The bar. Right about this often. Nobody honest clears it every time. | |
| −60% | Per deleted prediction. Erasing a failed call is worse than making one. | |
| +12% | For being wrong and correcting it fast. | |
| −35% | For being wrong and never correcting it. | |
| 25% | Right this rarely, with enough calls on the record, and nothing they say counts again. | |
Disqualifiers
- D1 · Internally inconsistent claimsA tenfold spread in community dosing with no primary source is a verification failure, not merely an absence of trials.
- D2 · Actively refuted mechanismThe cheaply-checkable mechanistic claim it rests on is contradicted by real research.
Every position on this shelf carries what would break it, and this publication is capped at the same 80% as everyone else. If a position here fails, it stays up with the failure recorded against it. Check the method against the chain, and if it has been departed from, that is the thing worth telling us about.