How a Stonck report is built

You type a ticker. A few minutes later you get a full equity research report with a rating you can actually act on. This post is about the part you don't see — what happens between those two moments, and the design decisions that make the output worth trusting.

It starts with the numbers, not the narrative

Every report is anchored to real fundamentals. When you submit a ticker, Stonck pulls the underlying financial data — income statements, balance sheets, cash-flow statements, price history, analyst estimates, insider transactions, and recent news — from a market-data provider. Nothing in the report is invented from memory or vibes; it's all traceable back to a filing or a quote.

That data is the foundation. Everything built on top of it is reasoning about what the numbers mean.

A team of specialists, not one generalist

Rather than ask a single model to "analyze this stock" — a prompt broad enough to produce confident nonsense — Stonck runs a team of specialist agents, each with one job and the data to do it.

There's an agent that reads the technicals (trend, momentum, where price sits relative to its own history). One that works the fundamentals (margins, returns on capital, leverage, the quality of the balance sheet). One that handles valuation — what you're paying for what you're getting. Others cover the earnings call, capital allocation, analyst estimates, insider behavior, news, and the increasingly unavoidable question of how AI reshapes the company's moat.

Each works its own corner in parallel, then hands its findings forward. The point of the division of labor is the same reason a real research desk has analysts and not one person: depth beats breadth, and a specialist is harder to fool.

The one rule we never break: the machine doesn't do the math

This is the most important design decision in the whole system, so it gets its own section.

Language models are extraordinary at reasoning and unreliable at arithmetic. So Stonck never asks one to calculate.

Every ratio in a report — every margin, every multiple, every coverage figure, every growth rate — is computed in plain, auditable code before any agent sees it. The numbers are handed to the agents pre-calculated. Their job is to interpret those numbers — to tell you whether a 28% operating margin is a fortress or a flash in the pan — not to derive them.

That single rule eliminates an entire class of failure. A model can't fat-finger a P/E it was never asked to compute. When you read "debt is 1.4× EBITDA" in a Stonck report, that figure came out of a calculator, not a hunch.

A skeptic, a scorecard, and a verdict

Once the specialists have weighed in, the report goes through three more stages.

First, a risk pass — an agent whose entire mandate is to find what the optimists missed and to argue the bear case hard enough that it can override a rosy consensus when the asymmetry warrants it.

Then a scorecard turns the qualitative work back into something rigorous: deterministic, math-driven grades that don't drift with the model's mood.

Finally, a portfolio manager weighs the whole picture and writes the verdict — the part you read first. It's editorial in tone but disciplined underneath, and it commits to a clear stance rather than hiding behind "it depends."

What you get, and what it isn't

The result is a structured report: a cover-page grade, the supporting analysis section by section, the bear case stated plainly, and a final take that doesn't equivocate. It reads like something a research analyst wrote, because the workflow is modeled on how one actually works.

What it isn't: a crystal ball, or financial advice. Stonck is a research tool. It's built to make you a faster, better-informed reader of a company — to do in a few minutes the legwork that would take you an afternoon — not to tell you what to do with your money. The market gets a vote too, and it doesn't read our reports.

That's the whole machine: real numbers in, a team of specialists arguing over what they mean, arithmetic kept strictly out of the model's hands, and a verdict that's willing to be wrong out loud. We'll keep writing here as the system evolves.