Why trust this?
Because it tells you what it knows, how it knows it, and — just as loudly — what it does not know. This page is the trust contract.
The Fair Witness discipline
A fair witness testifies only to what can be observed: "the house is white on this side." Applied here: every statement on this site is either a provenance fact (we hold N articles from M source domains), a corroborated claim with its evidence chain attached, or the word Unknown. There is no fourth category. Fair Witness is a consensus instrument, not a verdict engine: it reports what independent sources corroborate, and it refuses to convert "widely repeated" into "true" or "alarming" on its own authority.
Deny by default
A score, claim status, or alert level appears only after it clears a publication gate: independent corroboration across source tiers, a confidence floor, and a human-certified calibration epoch. Right now nothing has cleared that gate, so every judgment field on this site reads "Unknown." That is the system working, not the system failing. An instrument that guessed while uncalibrated would be worse than no instrument.
The claim-status ladder
Every fact, once extracted, will carry exactly one of these reader-facing statuses:
| Status | Meaning |
|---|---|
| Verified | Corroborated by primary documents or multiple independent tier-1 sources. |
| Reported | Asserted by credible sources; independent corroboration not yet complete. |
| Unverified | Circulating, but no source we hold meets the corroboration bar. |
| Unknown | Not yet evaluated — the honest default, shown plainly, never smoothed over. |
In the current phase every claim status is "Unknown" because fact extraction has not run. The ladder is published now so you can hold us to it later.
Source tiers (S1–S4)
Each ingested article carries a provenance tier: S1 wire services and major outlets with corrections practices; S2 established national/ regional outlets; S3 partisan-leaning or advocacy-adjacent outlets; S4 unvetted or unknown provenance. A tier describes where a claim came from, never whether it is true. Counting distinct source domains is not independence either — syndication can make one report look like twelve. Until independence analysis lands, the independent-source count reads "Unknown."
What we deliberately don't do
- No verdicts. We report corroboration, not conclusions.
- No partisanship. Institutions and mechanisms, not parties and personalities.
- No prediction. We describe what has happened, not what will.
- No activism. No calls to action; you decide what to do with a calibrated readout.
- No doomscroll mechanics. Bounded lists, no infinite feed, no urgency counters, no engagement tracking. Red is reserved for the single highest alert level — which has never been used.
- No invented numbers. If a value hasn't cleared the gate, the site says "Unknown." Always.
The calibration promise
When scores do publish, they arrive on a five-level scale explained in plain language, with the derivation attached — never a naked number. The contract is symmetrical: if this instrument says pay attention, pay attention; if it says normal, exhale. It can only keep that promise by refusing to speak before it is calibrated, which is what you are seeing now.