How it works · why trust it

Forensic stock research, from the filings — not opinion

Stockonomy reads the warning signs out of a company's SEC filings — a quality grade, red flags, distress and earnings-quality screens — and explains them in plain English. The core is deterministic math you can reproduce, not a chatbot's guess.

The problem it solves

The honest risk picture of any stock isn't the price chart — it's in the SEC filings every public company is legally required to publish: how strong the balance sheet is, whether the reported profits turn into real cash, and whether the price already assumes perfection. But that's 200+ pages of footnotes per company, so most people just hope. The tools that promise to help are usually one of two things: someone's opinion, or an AI chatbot that will confidently invent a number.

How it works — deterministic, not a chatbot

The scoring path is deterministic. The grade, the Altman-Z distress score, the Beneish-M earnings-quality score, the cash-conversion and capital-allocation checks — all of it is computed straight from the company's XBRL financial data with fixed formulas. That matters for two reasons: it's reproducible (the same filing always produces the same score, and you can check the math), and it can't hallucinate a figure that isn't in the filing. An LLM is used to write the plain-English narrative on top of numbers it is not allowed to invent.

The one exception, stated plainly. Some companies can't be graded by those formulas at all — banks and insurers, whose balance sheets the ratios aren't built for, and companies whose filing history is too thin to compute a trend. For those, and only those, the headline quality number falls back to the language model's read instead of a computed grade. Everything else on the page — the distress screens, the forensic checks, the financial line items — stays deterministic either way. We don't yet mark that fallback on the page itself, which we should; until we do, treat a company with no fundamentals grade shown as one where the quality number is a model's judgement rather than arithmetic.

This is the core difference from every “AI stock analyzer” at this price point. They are LLM wrappers — the model decides the verdict, so it can drift, flatter, or fabricate. Here the verdict is math; the AI just explains it.

What it actually checks

Financial health: how closely the balance sheet resembles those of companies in the year before they restructured — working capital against total assets, how much of the asset base was funded by profits kept rather than by borrowing, what those assets earn, and how far the equity's value exceeds what is owed.
Earnings quality: how much of reported profit arrived as cash, and whether receivables, margin and asset mix moved together in the combination seen at filers later found to have overstated earnings.
Capital allocation: ROIC, dilution, buyback timing, payout discipline — is management creating value or destroying it?
Forensic text signals: material-weakness, going-concern, restatement and auditor-change language pulled straight from the filing text.
A price-aware rating + a calibrated 12-month forecast: a probability distribution with a fat left tail, not one invented number to aim at. It's fit on cutoff years 2016–2022 and then scored against 2023–2024, which the fit never saw.

How we keep ourselves honest

We don't ask you to take the engine on faith. To test it, we replay the exact same engine on real companies' financials truncated to before known blow-ups — Carvana, Bed Bath & Beyond, Peloton — using only data that was public at the time. It flagged them on pre-collapse data; healthy controls (Apple, Microsoft) at the same cutoff did not, so it's not just crying wolf. We also calibrate the grade across more than a thousand point-in-time cases and measure the realized forward return and wipeout rate by grade.

And we're loud about the limits: forensic signals flag probability, not certainty; most companies that show one signal never collapse; small-caps are noisier; and the forecast band runs a touch narrow in the most volatile years. The full, reproducible track record — with the limits stated plainly — is on the proof page.

▶The Grade As Evidence · What A–F Has Actually Doneclick to expand

▶Methodology & Transparency · How Every Number Is Producedclick to expand

▶Competitive Positioning · Where Stockonomy Fits Next to the Terminalsclick to expand

Who it's for

Anyone who holds real positions and wants to know the risk they're actually taking — especially people concentrated in a single stock (an employer's RSUs, one big bet) who can't afford to be the last to know when the fundamentals quietly deteriorate.

Who's behind it

Stockonomy is built and run by an independent developer and investor who got tired of holding a concentrated position with no fast, honest way to read the risk buried in the filings. It's operated by an LLC currently being formed. The methodology and the full track record are public on purpose — every formula is named above, the calibration is documented, and the backtest is reproducible from the same public SEC filings — because for a money tool, transparency is the trust.

Questions, corrections, or feedback are genuinely welcome — founder@stockonomy.net.

See it on a stock you own

Run any US-listed operating company that files a 10-K through the forensic engine — grade, red flags, the distress screens behind them. Free, no card.

Analyze any US filer — free →

All guidesThe proof / track recordBrowse stock analysesHow it works

Stockonomy is an educational research tool. Nothing here is investment advice, a recommendation, or a solicitation to buy or sell any security. Forensic signals flag probability, not certainty. Data is sourced from public SEC EDGAR filings.

The problem it solves

How it works — deterministic, not a chatbot

What it actually checks

Financial health: how closely the balance sheet resembles those of companies in the year before they restructured — working capital against total assets, how much of the asset base was funded by profits kept rather than by borrowing, what those assets earn, and how far the equity's value exceeds what is owed.

Earnings quality: how much of reported profit arrived as cash, and whether receivables, margin and asset mix moved together in the combination seen at filers later found to have overstated earnings.

Capital allocation: ROIC, dilution, buyback timing, payout discipline — is management creating value or destroying it?

Forensic text signals: material-weakness, going-concern, restatement and auditor-change language pulled straight from the filing text.

A price-aware rating + a calibrated 12-month forecast: a probability distribution with a fat left tail, not one invented number to aim at. It's fit on cutoff years 2016–2022 and then scored against 2023–2024, which the fit never saw.