The signal
This is a worked example — an autopsy of one signal, months later, so you can see exactly what our engine looks for and what it ignores.
Start in February. On the 2nd, Rep. Julia Letlow disclosed a sale of United Rentals (URI) under the STOCK Act; on the 18th, Rep. Thomas Suozzi did the same. Two members selling the same industrial name inside a month is mildly interesting and easily coincidental. We logged it.
What survived the filter
Ten weeks later, it stopped looking coincidental. Between April 24 and April 27, four United Rentals insiders sold a combined $25.6 million of stock:
- Matthew Flannery — 22,768 shares at $984.98, roughly $22.4M — the company's chief executive
- Craig Pintoff — 2,466 shares, ~$2.4M
- Andrew Limoges — 548 shares, ~$536K
- Joli Gross — 306 shares, ~$292K
Grade A, two independent signals
Four distinct insiders, one four-day window, with the CEO supplying the bulk of the capital. And the detail our engine cares about most: not one of these sales was marked as a 10b5-1 plan trade. ClusterSignal reads the SEC's plan-trade checkbox on every Form 4 and throws out pre-scheduled, mechanical sales before clusters can form — roughly half of all insider transactions market-wide. Everything here survived. Discretionary decisions, made that week, at prices near $955–985.
Our engine graded the cluster A — its highest tier — on insider count, capital, seniority, and velocity. Combined with the earlier congressional sales, URI carried a bearish convergence flag: two independent signal types, no mechanism connecting a Louisiana congresswoman, a New York congressman, and four Connecticut executives, all selling the same stock within a season.
Why you're reading this in July
Honesty about our own mechanics: those events have now aged out of our live 90-day convergence window, which is why URI no longer tops the board. Signals are timestamped by when the trades actually happened — not when we detected them — because a disclosure product that fudges dates is worthless. What patterns like this have historically preceded is an empirical question; our public backtest tracks 30/60/90-day forward returns for every graded cluster, sample sizes disclosed. Look it up and decide for yourself.
Every transaction above links to its primary source — the Form 4 filings on SEC EDGAR and the House disclosure PDFs. That's the standard for everything we publish.