VerifiedNews
How often do week-one verification verdicts hold up against a later verdict built from cited evidence?
Share of week-one Contested verdicts that held up against a later cited verdict
23.9% [17.3%, 30.5%]
Sidestreet
What does each retrieval layer buy, and what does strictness cost a discovery engine?
Share of the engine's good answers put out of reach by restoring the strict filter
16.7% [8.0%, 26.6%]
Understudy
Can findings from synthetic user populations be trusted?
Share of sessions that reached the fully dead chip and were caught, activation mode, first population
47.4% [26.3%, 68.4%]
Every study cites its code, its data, and the decision it fed.
How to read these
- Verifiable artifact
- The claim reruns or recomputes from a committed file in the study repo, such as data, code or a frozen record, so you can check it without relying on my account.
- Published write-up
- The claim rests on an account already published outside the study repo, which you can read and compare but cannot rerun.
- Self-reported
- The claim rests on my own account, such as a decision I made or a figure no published file reproduces, and nothing on the page lets you check it.
Each study's verify block names the commit its design was frozen at, prints the commands that check it, and says whether anything beyond git metadata dates that freeze.