The method
Verify the source, not the story.
Most diligence trusts the pitch deck, the company narrative, the sell-side note. The detail that breaks a thesis is usually public, and usually unread. The engine reads the source instead. The thesis-breaker surfaces in your diligence, not after the wire.
01
Source
Every figure traced to its filing, trial registry, or peer-reviewed paper.
A number without a primary source is a rumor. Each claim in a report carries the document it came from, the verbatim line that supports it, and the timestamp it was retrieved — so you can re-verify any digit yourself. Nothing rests on trust, and no public fact ambushes you later.
02
Break
An adversarial model of a different lineage tries to refute every claim.
A model checking its own work shares its own blind spots. So a separate, independent model is pointed at each finding with one instruction: refute it. Only claims that survive a genuine attempt to break them remain — so the weak claim dies in private, before your investment committee sees it, not after you wire.
03
Pre-register
The call is published before the data lands.
A track record assembled after the fact proves nothing. Each read is pre-registered — in public, dated, before the readout — so the record is auditable, not asserted. You see the misses as well as the hits, and you never have to take the batting average on faith.
The result is a report where every claim is auditable to its source, every number has survived an attempt to break it, and every forward call is on the public record — a document you can take into a committee and defend line by line.
Point me at an asset