Equity research, reimagined for the AI era

Equity research was built for paper. We rebuilt it for AI.

Stoquity is an AI-native equity research ledger. Three frontier models analyze every company, independently and with nothing to sell. Every call and forecast goes on a live record that grows with each filing and can never be rewritten.

Living research. On the record.

The problem

Most equity research you can read was written by someone with something to sell.

The research that reaches the public, and much of what reaches institutions, is sell-side: produced by banks and brokers that also want the covered company’s banking business and their clients’ trading commissions. Regulation manages that conflict. It does not remove it.

Almost everything is a Buy.

Roughly one in twenty analyst ratings on S&P 500 companies is a Sell, by FactSet’s count. Negative ratings are rare where the companies covered are also clients or prospects.

Coverage follows fees.

Companies that generate banking and trading revenue get analysts. Thousands of sound, smaller companies get little or none.

It is a document, not a record.

A note is frozen on the day it is sent. When it proves wrong it is replaced by the next one, and nobody keeps the tally.

Stoquity has no investment bank, no trading desk and takes no payment from the companies it covers. 21% of our current ratings are Cautious: 233 of 1,090.

The difference

AI makes a different kind of equity research possible. So we built it.

Traditional equity research
Stoquity
Who writes it
One analyst, one house view
Three AI models from three vendors, working independently
Who pays for it
Banking and trading relationships
Readers and data customers. Never the companies covered.
How long it lasts
Out of date the day after it is sent
Updated with every filing; the whole history stays on the page
Coverage
Where fees justify an analyst
1,090 companies, every one to the same depth
The numbers
Typed into a spreadsheet by hand
Computed by code from SEC filings and market data
Can you test it
You read it
You run it: playback, what-if scenarios, your own assumptions
When it is wrong
Superseded by the next note
Scored against the filing. The miss stays on the record.
Who can read it
People, as a PDF
People and their AI agents, as structured data
1,090US-listed companies, each with its own ledger
3frontier models on every forecast, from Anthropic, Google and Mistral
46,000corporate events on record, drawn from SEC filings
3,093forecasts written down before the filing that settles them
A live and living ledger

The history of the company, and of everything we said about it.

Each company has one continuous record instead of a stack of dated notes. It holds what the company did, quarter by quarter and event by event, beside what we called and how that call turned out.

  • Timeline. Every reported quarter, latest first: revenue, earnings, cash flow and backlog, beside the share price reaction.
  • The long arc. Acquisitions, leadership changes, restructurings and restatements from the company’s SEC history.
  • Our record. Every rating, review, forecast, score and correction, in date order.

How far back each part goes

We would rather tell you than have you find out.

Part of the recordReaches back to
Quarterly financialsThe last eight quarters
Annual financialsThe last four fiscal years
Corporate events2017 for the typical company; 2004 at the earliest, 2022 for the most active filers
Share prices and risk statisticsMay 2024
Our ratings and their scoresJune 2026
Three-model forecastsNovember 2026

History only grows from here. Nothing already on a ledger is trimmed.

Multi-model scoring

One analyst gives you an opinion. Three models give you a range.

Models from Anthropic, Google and Mistral answer the same question about the same company, separately. Where they agree, the outlook is clear. Where they split, you have found the real uncertainty, which a single confident voice would have hidden.

  • Identical questions. Every model gets the same figures and the same claim.
  • Probabilities, not adjectives. Each answer is a number that can be checked.
  • A public scoreboard. As filings arrive, each model is ranked on how well its probabilities held up.

What that looks like on one company

Analog Devices, from its ledger. Real entries.

Aug 20RatingNeutral, reaffirmed at $370.24. Turns Cautious if revenue growth falls below 8%.
Aug 21ReviewThree-model committee concurs, 3 of 3
Sep 15ScoreThe August 20 rating: hit
Nov 21ForecastRevenue next quarter is at least $4.10B
Claude 62%Gemini 90%Mistral 85%
Nov 21ForecastFree cash flow next quarter is at least $1.49B
Claude 58%Gemini 85%Mistral 60%

Both forecasts settle from the SEC filing expected about November 24. No forecast on the ledger has settled yet.

Institutional-grade analysis

The depth of an initiation report, on every company, kept current.

Not a summary of the latest earnings call. Each page joins the company’s filings to the data around it, and shows the working.

  • More than the SEC. Federal contract awards, national defense budgets, interest rates and macro series are joined to the filings where they matter.
  • Every figure traceable. Each number names the filing or dataset it came from.
  • Coverage others skip. Mid-sized companies get the same treatment as the largest.

On every company page

Financials, derivedRevenue, earnings, free cash flow and backlog, rebuilt quarter by quarter from filings.
Ratio analysisValuation, margins, returns, leverage, interest cover, cash conversion.
Risk statisticsBeta with its confidence range, volatility, drawdown, behavior on down days.
Break conditionsThe specific figures that would change the rating.
Scenario toolsWhat a rate move, a growth change or a supply shock does to the case.
Stoquity Slices25 lists that rank every company on published rules.
Methodology

Models reason. Code calculates. Filings decide.

Each task goes to whatever can be trusted with it. A model never produces a financial figure, and never marks its own work.

1
Gather

Filings, prices, corporate events and outside datasets are pulled into one structured record per company.

2
Compute

Code derives every quarter, ratio and risk statistic. Nothing is typed by hand.

3
Rate

A rating is published with its date, the share price, and the figures that would prove it wrong.

4
Challenge

Three models review the rating and forecast the next quarter, without seeing each other’s answers.

5
Settle

The next filing arrives. Code marks each forecast true or false and scores every model.

The rules we hold to

The same forecasts for every company
Revenue, earnings per share, free cash flow and backlog against a stated threshold, so results compare across companies and models.
Every model, every time
A forecast goes on the record only with a full set of independent answers. No model is quietly left out of a hard question.
A real yardstick
A model earns credit only for beating what history alone would have predicted. Looking competent is not enough.
No forecasts after the fact
A forecast written when the answer could already be known is withdrawn and shown as withdrawn. 62 so far.

What it does not do yet

Foreign filers
Companies that report under international standards do not yet have a timeline or forecasts.
Banks, insurers and property trusts
Fewer forecasts, because free cash flow and backlog do not describe them.
Outside data
Deepest for aerospace and defense today. Other sectors follow.
A short track record
Ratings have been scored since June 2026. The first forecasts settle this month.

Read the full methodology

An immutable record

What we said, when we said it, provably.

Anyone can claim a track record. Ours cannot be edited, by us or anyone else. That is what lets you trust the research before you agree with it.

  • Append-only. Entries are added, never changed. A wrong figure gets a dated correction beside it.
  • Fingerprinted. Every version of every record is stored under a hash of its contents.
  • Independently timestamped. Each day’s manifest is anchored to a public blockchain, so the date can be verified without trusting us.

The record so far

Since June 2026, in full, including the parts that do not flatter us.

RatingScoredAgainst benchmark
Conviction8+3.9 pts
Positive306+0.7 pts
Neutral496−0.8 pts
Cautious222−3.6 pts
manifests/2026-11-21.json  →  fingerprint of 1,090 records  →  timestamp proof
Start anywhere

25 ways into 1,090 companies.

Stoquity Slices are lists scored by published rules, by code. Each company links to its ledger. Lists with a dashed outline are warnings, not ideas.

A new-age architecture

Headless at the core. A harness around the models.

Equity research has always been shipped as a finished document. Ours is built the way modern software is: the substance is separated from the presentation, and the models are parts that can be measured and replaced.

HeadlessThe research is data first.

Every company is one structured record: financials, events, ratings, forecasts, scores. The web page is only one view of it. Your terminal, your spreadsheet or your AI agent can be another.

HarnessThe models are interchangeable.

The harness puts identical questions to any set of models, keeps them apart, and scores each against the filing. Today it holds three. It can hold yours.

Web LiveCompany ledgers and Slices, for people.
API WaitlistEvery record as data, queryable as it stood on any date.
MCP WaitlistA Model Context Protocol server, so AI agents can read the ledger directly.
For institutions

Run the ledger your way.

Because the research is headless and the models sit in a harness, a firm does not have to take our view. It can run its own on the same rails.

Swap the models

Seat your own model, or a different trio, beside or instead of ours. Get a private report on how well calibrated each one is on real companies.

Tune the methodology

Set your own forecasts, thresholds, review rules and scoring. Keep what works from ours and change the rest.

Bring your capital market assumptions

Your rates, growth, risk premia and sector views flow through every scenario and valuation, across all 1,090 companies at once.

Keep a private ledger

Record your own theses and break conditions with the same dating, scoring and timestamping. Only you see it.

Contact us to join the waitlist. Most of this opens within two quarters, in the order firms ask for it.

Join the waitlist

Living research. On the record.

Explore 1,090 companies