About

What this is

We run an extraction pipeline over financial Twitter. Every post from the accounts we track is parsed into structured claims — ticker, direction, claimed magnitude, stated horizon — and stored in a database. Each claim is then scored against what actually happened, measured as excess return over SPY across the claim's own horizon. Every quote on this site links back to the original tweet, so you can check us against the source.

The output is the site you're reading: the Hypotheses database and its Track Record, weekly Deep Read synthesis and Stories, per-ticker and per-author briefs, a mechanics Alerts feed, and The Book — a daily model-consensus ledger whose live, pre-market-timestamped record begins Monday 2026-07-06.

How the scoring works

We hold claims to their own words. A tweet that says a stock will rip 30% in a month gets adjudicated against that magnitude and that horizon, in SPY-excess terms — beating a rising market is the bar, not just going up. Confirmed and refuted claims are published alike; the honest headline is that most tweeted claims do not clear their own bar. Author grades (A/B/C, how their written analysis reads) and trader scores (how their last-20 timestamped calls performed vs SPY) are two deliberately separate measures — a great read can be a bad trader, and the site says so when that happens.

Who runs it

An individual operator plus an automated pipeline. Extraction, scoring and page generation run on a schedule; the weekly analyst pass and the editorial choices are reviewed by one person. There is no team, no fund, and no third party paying for placement.

What it costs

Free during the live-validation period. We think the right time to ask anyone for money is after the live record — not the replay backtest — has had time to prove or embarrass us.

Disclaimers

Nothing on this site is investment advice or a recommendation to buy or sell any security. We may hold positions in names discussed. Extraction and pricing pipelines have bugs and gaps; where data is missing we try to say so on the page rather than fill in a number. Past performance — especially replayed, in-sample performance — tells you little about the future.