Hypothesis in,
proof out.
A local research cockpit that refuses to believe an AI until it has receipts. Every AI-surfaced stock idea is a hypothesis — checked against the primary SEC filing before it earns a single line of the journal. Nothing trades. Ever.
An AI speaks with total confidence.
Confident isn't true.
So the claim doesn't get believed — it gets dragged to the primary source. Watch a fabrication die on camera, then watch a real signal get rejected for a real reason. The gate runs continuously; this is what it does.
—
It remembers every lie it's caught.
The gate keeps a standing catalog of fabrication signatures — the way an immune system remembers a pathogen. A match only ever accelerates suspicion, never trust. Fast path to "look harder or kill it." No fast path to "believe it."
Four steps. No shortcuts.
A claim arrives
An AI surfaces a ticker with a confident, specific thesis. It lands as a hypothesis — believed by no one yet.
→Check the source
Every load-bearing number is pulled from the actual SEC filing — not the AI's word. The claim is shown beside the proof, never instead of it.
→Render a verdict
Confirmed, pending, or killed — on the claim's truth, not on the stock. Fabrications die here.
→Seal it
Survivors are written to a tamper-evident, hash-sealed journal — timestamp, source, and the falsifier that would kill the thesis later.
Even a real signal dies
when the facts don't hold.
Catching lies is half of it. The harder half is refusing a convenient yes — a signal that's genuinely real, but fails on the fundamentals.
In a genre full of fake backtests,
showing nothing is the flex.
It never invents a track record. The calibration score reads a dash until a real decision has a real outcome. The proof comes now; the grade comes later.
Signals decay.
Proof compounds.
Not financial advice · a research demo