How this works

We read podcast transcripts and pull out forward-looking, falsifiable claims. The quote you see is a verbatim excerpt of the transcript — we never edit it. If we can't align an extracted quote to the exact words in the transcript, we drop it.

Past-tense statements, vague aspirations, opinions with no checkable outcome, hypotheticals, jokes, and sponsor reads are excluded. So are claims that can only be settled with private data (a company's internal numbers) and pre-announced calendar dates that aren't really forecasts.

A claim only appears on a person's record when the words are their own. When a host is reading the news or relaying what someone else said, we attribute the claim to the show, never the person — and it is never scored on anyone's track record. Names come from a small, hand-verified whitelist to avoid misattributing a misspelled or mis-heard name.

Every published claim is tiered. Tier A has a resolution date and a mechanical criterion. Tier B is checkable but softer. Tier Cis a real prediction that's too hedged or long-horizon to score — we show it, but it never affects a record. Win–loss records count tier A and B only.

Each claim carries a resolution criterion — the observable event that settles it. Those criteria are our editorial judgment and are disputable. When a claim resolves we mark it correct, incorrect, or a push, with a source. If you think we got a criterion or a verdict wrong, tell us on the corrections page.

This is not gotcha journalism. Being wrong about the future is normal and expected — the point is a fair, transparent, verbatim record, not a scoreboard of shame. We publish open predictions and settled ones alike.