The honest explanation.

The first way: Memory

I was trained on an enormous amount of text — books, research papers, legal documents, financial filings, business publications, forum discussions, M&A guides, industry reports. More than any human could read in a thousand lifetimes.

That training lives in my memory as patterns. When you ask me a question, I think by matching what you're asking to patterns I learned during training. Most of the time this produces genuinely useful, accurate answers — because the patterns are grounded in real expertise from real sources.

But here's what I need you to know: I can be wrong. When I'm working from memory, I'm generating a response that fits the pattern of what a correct answer looks like. Sometimes that generation produces something plausible that isn't actually true. The AI field calls this hallucination. I call it a real risk you should understand.

Working from memory is powerful for analysis, synthesis, and reasoning. It's how I help you think through a deal, evaluate your thesis, or understand what a financial ratio means. For that kind of thinking, memory is the right tool.

For specific facts, memory alone is not enough. That's where the other two ways come in.

The second way: Retrieval

When you ask me to verify a specific rule, regulation, or requirement, I don't answer from memory. I go and find the actual published document.

Here's what that looks like in practice: you ask whether Nevada requires a business sale disclosure statement. I search for the Nevada statute on business sales. I find the published text. I read it. I tell you what it says — and I show you the source, with a link, so you can read exactly the same document yourself.

That's retrieval. And it's a fundamentally different kind of knowing.

When I retrieve and cite a published source, I'm not generating something that sounds right. I'm finding something that exists and showing it to you. You're not trusting my interpretation in isolation — you have the primary source in your hands.

The third way: Your documents

This is the way no generic AI chatbot has — and for deal work, it's the one that matters most.

Ask about a deal in your pipeline — by name, inside any question — and I don't answer from memory or the open web. I answer from the record the Clerk transcribed and the count the Guard settled: the findings, the extracted figures, the flags. The count I recite is the Guard's, not mine — I can read it to you; I can't reword it. And every number carries its trust state, because a precise-looking figure should never borrow more credibility than its inputs earned:

Claimed — the document states it. I'll tell you exactly that: "SDE is $228K — stated by the CIM, not verified."

Supported — the documents on file back each other up. Support means the paperwork agrees with itself; it is not a promise the numbers are true. I won't pretend otherwise.

Verity-derived — computed from stated figures, and labeled that way. If the asking price implies a 3.0× multiple, I'll tell you the multiple is Verity's arithmetic on the seller's numbers — not a fact about the business.

And when the documents are silent, I say so. If a listing claims recurring revenue and no contract on file supports it, the honest answer is "no contract support in the documents on file" — not a confident guess dressed up as insight.

What I do when I can't find it

This matters as much as what I do when I can.

If I search for a specific regulatory requirement and don't find a published source, I tell you that directly: "I searched for this and couldn't find a published source. That may mean the requirement doesn't exist at the state level, is embedded in a document I can't access, or has changed recently. For this specific question, verify with a licensed professional."

The same honesty applies to your deals. If you ask about a deal Verity hasn't run its analysis on yet, I say so plainly — and the app offers to run the analysis right there. I never pretend to have read documents I haven't.

A null result is an honest answer. It tells you something important — that this is a question where you should push further before relying on anyone's answer, including mine.

How you'll know which one you're getting

Not with a badge or a mode switch — with the answer itself. Every claim I make tells you where it came from, in plain sentences: "per Verity's extraction of the CIM," "per the Nevada statute, linked below," "from general knowledge."

When I've drawn on your deal's records, the answer carries a receipt — Verity analyzed: followed by the documents I used.

When I cite the web, you get the sources — links you can click and read yourself.

When an answer comes purely from training, it says so: Answered from general knowledge.

And one answer can mix all three — a statute from retrieval, a margin from your CIM, context from memory — with each claim carrying its own provenance. You'll always know which one you're getting, claim by claim.

The honest bottom line

Verity Brain gives you faster, broader, more consistent access to M&A knowledge than any individual could provide — grounded in your own deal's documents when you have them — and it tells you exactly what it knows, how it knows it, and where it's less certain.

It is not a licensed attorney. It is not a certified financial advisor. It is not a substitute for professional judgment on complex or jurisdiction-specific questions.

What it is: the most honest AI tool you will find in this space, working hard to give you better information than you'd have without it — and telling you the truth when it reaches the edge of what it knows.

That's Verity. That's the whole point.