Skip to main content

← SiftingSignal · Methodology · updated 06 Jun 2026

How we build each synthesis.

Every topic gathers what a wide range of sources are saying, ranks each source by the kind of voice it is, and publishes the read with citations, a contestedness score, and a record of who disagrees.

What we cover

MediaBooks, TV, film, music, games. Critics, prize juries, trades, communities.
AIResearch, the major labs, named practitioners, AI policy. First-party + the skeptics.
MoneyMacro, markets, personal finance, crypto. Central banks, named analysts, retail talk.
HealthClinical evidence, nutrition, longevity, public health. Journals, regulators, evidence-checkers.
SportsUS majors, college, global soccer, F1, combat sports, golf, tennis.

How we rank sources

Every source sits in one of five tiers. The tier names what kind of voice is in the room, not whether it's right. A popular voice can be brilliant and a top expert can be wrong; the tier just makes the structural call visible.

Top expertsPeer-reviewed research, named specialists, first-party institutional data.
InstitutionalEstablished outlets producing structured analysis: medical bodies, central-bank research, investigative desks.
Industry · pressTrade press, industry analysts, working practitioners writing professionally.
Popular voicesCreators, popularizers, mainstream interpreters. High reach; audience-shaped framing.
ForumsAnonymous + community signal. Useful for sentiment; never the basis for a claim on its own.

Every assignment is in our source directory with a one-line reason. You can challenge any of them.

The aggregator

For each topic, the engine pulls every recent signal from a curated source list across a rolling 30–90 day window. Signals are deduplicated, classified, embedded, and grouped by claim. A primary language model produces a synthesis — what each tier is saying, which claims are best-supported, which sub-questions stay contested. A second model from a different vendor independently re-reads the same signal set. If they substantively agree, the synthesis is published. If they disagree beyond a tolerance, the synthesis is held for editorial review rather than shipped.

What every synthesis shows

Each published read carries a contestedness score (range of opinion across cited sources), an evidence ratio (how much of the cited material came from the higher tiers), and a citation footer that names every source it pulled from. No claim ships without a source — cite or die.

When we get it wrong

Tier assignments and published syntheses are public and audited. Every reviewed change goes to the corrections log with the reason. We don't quietly edit; we mark.

Going deeper The reader walkthrough lives at /how-it-works. The full tier criteria are at /methodology/tier-criteria. The live source list with reasons sits at /sources.