The SEC, the AMF and BaFin publish what company insiders buy and sell. We read all of it, fold it into one schema and rank it with deterministic statistics — so you read the ten trades that matter, not the ten thousand that don’t. Statistics predict; AI explains.
The last filings the machine processed. No account needed to look.
The public record is free. The machine isn’t.
Every raw transaction, company and insider stays free, for everyone. Plans unlock what we compute on top of it — statistics, signals, alerts and AI search. Seven-day trial, cancel any time.
A disclosure lands: free text, footnotes, share counts in whatever convention the filer used, prices sometimes missing altogether.
Normalised
Type
Other
Qty
-65,620
Price
0.00 USD
Value · USD
$0
Issuer, insider, role, side, quantity, price, currency, value in USD. One schema for three regulators and twenty years of filings, so a French disclosure and an American one sit in the same table.
Scored
Cluster buy3 or more distinct insiders trading the same stock in the same direction within 90 days
Unusual sizeAt least 2 standard deviations above this insider's own typical trade size
First buy in 12moFirst open-market buy after 12+ months without buying
The trade is measured against the insider’s own history and the company’s. Three deterministic tests, no model in the loop — the same input always produces the same score.
Explained
Only then does a language model write the sentence. It never invents a figure: it reads the ones above and says what they mean.
Every signal we compute is tested against what the stock actually did next, and the whole registry is published — including the horizons where the edge disappears. The insider edge is real and it is short-lived: strongest in the first weeks, gone within a year.