Method and data
How we scored it
Every call gets one of six verdicts, checked against a public source. Here is what each verdict means and where the data comes from.
Scoring
- HitThe main claim happened, within the stated or reasonable time frame, confirmed by a public source.
- PartialThe direction was right but a key detail (size, timing, actor) was wrong.
- MissIt did not happen, or the opposite happened.
- VagueNo checkable target or deadline. Not counted in hit rates.
- UnresolvedToo recent, or we could not confirm it.
- After the factPosted once the event was already under way.
Strict hit rate counts hits only. Half credit counts a partial as half a hit.
The datasets
- CuratedHis best-known calls plus ones readers shared. Biased toward hits, which is why we also use the others.
- Random sampleRandomly drawn prediction-like posts.
- Recent censusEvery prediction-like post from Apr to Oct 2026.
- India protestsCalls about unrest in India, matched to the first public report of any matching event.
The original study snapshot covered 8,081 public posts from Oct 2024 to Oct 2026, read through the X API. This raw-post snapshot is separate from the growing library of audited claims; current audited-claim totals appear on Overview and Archive. We did not post, like or reply to anything. All data comes from public posts read through the official X API. We only show short excerpts and link to the original posts.
Our forecasts
Each has a checkable claim, a deadline, a probability and a real trigger source. Status is updated with every data refresh. Calibration and the Brier score fill in as they resolve.
Caveats
- Judging “hit” involves judgement, especially for partials. Every verdict links its source so you can disagree.
- Small samples: a few calls can move a percentage a lot.
- Deleted posts and replies outside the archive are not covered.
- This is about public posting behaviour only. We make no claims about the person behind the account.