How we report
These are the rules the show is made under. Each one comes with a real example from a published episode, because a methodology page that can't point to its own work is just another headline.
Settlement terms come from the executed agreement, not the press release. A court holding comes from the opinion, with the case number. If the only source is a headline, we say it's a headline.
From the show
The August 31 episode took the settlement terms from the fully executed agreement filed as ECF 572-1, and the Sixth Circuit's holding from the published opinion in NetChoice v. Skrmetti, No. 25-5660.
"At least $12.1 billion, guaranteed" and "up to $17.1 billion, contingent" are different sentences. We do not collapse them to make a cleaner headline.
From the show
Four settlement totals ran in the same week - $16.7B, $17.1B and ~$18B. We reconciled them from Exhibit B rather than picking one, and explained that Meta's ~$10B legal accrual is an accounting charge, not the payout.
Before the news, not after it, and not buried in show notes. If we were careful in the wrong direction, we say that too.
From the show
August 31 opened with two corrections to August 24: the settlement had in fact been signed on August 26, and Meta's New Mexico appeal deadline was Tuesday September 8, not "around September 5" - because the 30th day fell on a Saturday before Labor Day.
Every episode closes by naming the categories that came back empty or that we did not reach. A quiet system is not the same as a quiet world.
From the show
In one week our research pipeline reported five of seven categories empty. It was wrong - four real stories had been found and dropped before verification, including the New Jersey phone law. We caught it by hand and told listeners why.
Candidate statistics go through an adversarial check before they reach the script. The ones that fail are dropped, and the episode says how many.
From the show
"Four candidate statistics were refuted in verification and are not used." Two screen-time studies surfaced the same week and both were voted down in our own fact-check, so neither number appears.
Some errors are so predictable we wrote them into the tooling. The scripts that generate captions and social posts refuse to emit them.
From the show
Never "NYC banned AI" without the grade scope (2-K through 8) in the same sentence. Never imply the Sixth Circuit blocked Tennessee's law - it vacated and remanded on one prong, and the law is in force.
When an episode touches self-harm or youth mental-health findings, no method or circumstance is described, and the 988 line is read verbatim.
From the show
"If you or your child are struggling, the 988 Suicide & Crisis Lifeline is free, confidential and available 24/7 - call or text 988."
Weston is an AI-generated voice. We say so on every page rather than letting you find out. What he reads is produced by a research process that pulls primary sources, runs candidate claims through an adversarial verification pass, and drops the ones that fail - and a person reviews every episode before it publishes.
AI-assisted, human-accountable. If something is wrong, it is our fault, and the next episode will say so.
Found an error? Tell us. Include the episode date and, if you have it, the source that contradicts us. Corrections are issued at the top of the next episode and noted on the written report.
Reporters: transcripts are citable and carry the docket numbers. Browse transcripts →