A living trust index of everything that informs — outlets, journalists, creators, podcasts, AI models. Computed by the Oya algorithm — the algorithm's output is the Index, definitionally. Six structural axes, one behavioral burn record, full history. No composite, ever.
Default sort: evidence strength (the trust axis) — locked by doctrine: it does not change per subject type or editorial mood. No composite exists and none ever will. Click any column to re-sort. Framing values are observed lean, disclosed — never punished.
| Subject | Type | Evidence | Diversity | Account. | Opinion | Political | Economic | Burn | Flags |
|---|
Readers nominate subjects the Index should score: intake (name, type, links, why, contact) → public review queue → notability check (real publishing subject, sampleable body of work, not a duplicate) → decision + notification to the nominator. Every nomination carries its status and the written reason. No pay-for-placement, ever.
Outlets, journalists, creators, podcasts, AI models. How we research each type and where we draw the line is published in OYA-INDEX.md: the rated unit is the body of work (the byline, the channel, the show, the model version) — never a whole person, never a rumor.
How well a subject's aggregated claims are supported by non-social evidence, with honesty calibration: stated certainty above measured support is penalized per claim. The default sort is evidence strength and the doctrine locks it — no composite, ever.
Breadth across cited outlets, countries, and source tiers. One subject repeating itself has zero breadth, no matter the volume.
Is it clear who stands behind the content and how it's funded? Outlets: ultimate parent, chain depth, funding. Journalists: byline, employer, corrections practice. Creators/podcasts: operator, funding model, corrections. AI models: builder, training disclosure, published evals.
Share of claims that are opinion/normative rather than checkable. Recorded as signal, never as accusation.
Observed lean of framing. Published as observed signal — the Index does not punish lean, it discloses it.
Every burn carries severity = reach × (1 + log10(volume)) × intent (honest 1.0 / reckless 1.5 / knowingly-false 2.0). A years-long mass-reach fabrication is a forest fire (years to recover); a 2-follower honest mistake is a grilled plant (~a week). The scoring penalty decays exponentially per severity band. A burn on a claim later verified TRUE is exonerated immediately and fully. The event history stays visible permanently — only the scoring penalty decays or is expunged.
Repetition is not corroboration. Social-tier evidence scores zero. Evidence floods raise public GAMING_SUSPECT flags. Burn history is permanent and follows rebrands. Same inputs always reproduce the same scores — input digests (SHA-256) are published per version.
Align Newsroom carries a scorecard like any subject. If we fail our own standard, it shows up here.