WHY TRUST THE OUTPUT
Trust is a property of the pipeline, not a promise.
Nothing here asks you to take the output on faith. Every verdict is built so you can check it — and so the system can't quietly pretend to know things it doesn't.
Evidence is what the model was given
The evaluator receives targeted sections of SEC EDGAR filings and pre-computed metrics, including market prices where a framework uses them. That is the evidence base. The model is instructed never to use what it remembers about a company as a source; a missing section becomes an unknown, not a guess.
Citations, checked against the filings provided
Citations name the filing section they rely on, and each is checked against the sections the model was actually given. A citation that names no section is shown as Unattributed and marked unverified — never presented as sourced.
Item 1A, Risk FactorsUnknowns are a feature
When a filing doesn't disclose something the methodology needs, that dimension is marked unknown and the verdict is capped accordingly. Better an honest gap than a confident fabrication.
segment-level margin split not disclosed in source materialsYou confirm the cost before it runs
Scans, deep dives and Strategy Lab runs show an estimated cost you confirm before they start, and every run reports what it actually spent. Research you can audit includes the bill.
What this can and can't do
The output is AI-generated research. It can misread a filing; it cannot know what a filing doesn't say; and it never knows anything about you. Treat every verdict as a structured starting point for your own reading — the citations exist so you can do exactly that.
- Screens over 5,000 U.S.-listed companies on the numbers.
- Reads the SEC filings of the top-ranked companies that pass each framework's screen.
- Names the filing section behind each citation.
- Names what the filings don't say.
- Reuses a recent read when a company's filings haven't changed.
- Shows an estimated cost you confirm before a scan starts.
- Sends price-level alerts (above or below a price), optionally combined with a screen's filters, by email and in-app, plus webhook on Mother Lode.
- Doesn't tell you what to do. A verdict describes fit to a framework; it is not a recommendation to buy, sell, or hold anything.
- Doesn't take anything about you as an input.
- Doesn't place trades, connect to a brokerage or hold investments.
- Doesn't cover ETFs, funds or listings outside U.S. exchanges today. We're looking at adding them in 2027.
- In a scan, doesn't run an AI read on every company, only on the top-ranked ones that pass the screen.
- Isn't allowed to fill gaps from a model's memory: what the filings don't say is marked unknown.
- Doesn't publish a track record of its verdicts, and doesn't forecast returns.
See the discipline applied, end to end
One sample verdict, dissected with every citation visible.