Who gets paid.
Who owns it.
Who's watching.
Every large company already tells regulators what its chief executive makes and what its typical worker makes. Almost nobody reads the filings. Search one.
Start here
Starbucks
1,794times · chief executive pay ÷ median employee pay calculated
A part-time barista made $17,279. The chief executive made $30,992,773. Both numbers are Starbucks' own, filed with the SEC. It would take that barista 1,794 years to earn one of his years.
Run the model
What would 20 to 1 look like?
Drag the cap. This changes one thing only — the ceiling on chief-executive pay as a multiple of that company's own median worker — and holds everything else still. It is a proposal, a way to make a large number legible. It is not a prediction and not something any company has said it would do.
It wasn't always like this
In 1965 the ratio was about 20 to 1.
The gap is not a law of nature. It is about forty years old. Economists reconstructing executive pay back to the 1930s found big-company chief executives earning roughly twenty times a typical worker through the 1950s, 60s and 70s — a professional's salary scaled up, not a fortune. Then, after 1980, it left the ground.
Historical figures are the Economic Policy Institute's series for the 350 largest US firms, measured against a full-time production worker. The 2025 average is the AFL-CIO's S&P 500 figure. The Starbucks bar is a company disclosure against its own median employee, part-timers included — a different yardstick, which is why the honest comparison is 1,794 against 312, not against 20.
Three things changed after 1980
Pay became stock
Options and equity awards replaced salary as the main form of executive pay, so compensation at the top began tracking the stock market rather than the company's own payroll. Nobody else's pay was tied to it.
The top tax rate collapsed
The highest marginal rate was 91 percent in the 1950s and 70 percent until 1981, then fell below 40 percent within a decade. At 91 percent an extra million was mostly a gift to the Treasury. At 37 percent it was worth fighting for.
Worker pay stopped moving
Productivity kept climbing. Median wages, adjusted for inflation, roughly flatlined. The bottom of the ratio stopped growing at the moment the top took off.
The second question
Pay is the part you can see. Ownership is where the money lives.
Cap every executive on this site at twenty times their median worker and you move something on the order of $150–200 billion a year worldwide — about a tenth of one percent of the global economy. Real, worth doing, and not the whole story. A salary is labor income. Most concentrated wealth is ownership: shares, dividends, capital gains. A pay cap does not touch it.
So every company page carries an ownership row, and right now every one of them reads insufficient data. No company here publishes what share of its stock is held by rank-and-file employees. That silence is the finding. Getting those numbers is the next job, and it's one the public can do faster than I can alone.
Five ways ownership has already been reset somewhere
| Broad employee equity | Publix, WinCo, John Lewis (UK), thousands of ESOPs | Shares to every worker every year as part of pay. Nothing taken; new ownership created and spread. |
|---|---|---|
| Citizens' wealth fund | Alaska Permanent Fund, Norway's sovereign fund | The public holds a slice of the economy; everyone gets the dividend. |
| Tax ownership like work | Partial, in several countries | Equal rates on capital and labor income; a real inheritance tax above a very high threshold. |
| Antitrust | US 1911–1984, EU today | Concentrated ownership rides on concentrated markets. |
| Workers on the board | Germany, above 2,000 employees | Doesn't change who owns shares. Changes who owns decisions. |
The honest objections: capital taxes may reduce investment; wealth funds can be politicized; single-company employee stock is dangerous if the company fails, which is what happened to Enron's workers; founders who take real risk arguably earn real reward. Serious versions of any reset are designed around those.
Three AI systems audit each other. Citizens audit the AI. And exposing corruption pays better than hiding it.
An experiment, not a product
Can AI audit a corporation?
Here is the honest status: this experiment has not been run yet. No findings are published below, because there are none. What follows is the protocol, published first so that the method can be criticized before any result exists to argue about.
One document, three readers
Take a single filing — a proxy statement, a 10-K. Give it to three AI systems built by different companies, with the same instruction: find everything in here an ordinary employee, investor, journalist or citizen should know.
They read each other
Each system then reviews the other two's findings and marks what it agrees with, disputes, or thinks is unsupported by the document. Agreements, disagreements and open questions are published separately — the disagreements matter most.
Humans decide
Every finding carries its source passage, the AI's reading, a confidence level, a counterargument, and a human review status. Nothing is treated as established because a machine said it. The AI finds and publishes; people judge.
The failure is part of the result
If the three agree on something wrong, that is the most valuable output the experiment can produce, and it gets published as loudly as any finding. Systems trained on similar data share blind spots. The point is to measure that, not to pretend it away.
Why bother: if AI can already read every filing a company produces, the question is not whether corporations will be analyzed that way. They already are, by the people who can afford it. The question is whether the public gets to see the same thing.
The full design — three monitors in different countries, citizen panels chosen by lottery, whistleblower bounties paid from penalties, and courts rather than machines imposing consequences — is in the design document. Its known weak points are listed there first, before its claims.
Please do
Prove us wrong
Find an error. Find a better source. Find a flaw in the method. Find a company whose numbers say the opposite of the argument. This project is worth more if it survives that than if it avoids it.
What to send
- A number that's wrong, and the filing that proves it
- A better or more current source
- A methodology critique — especially of the Pay Fairness score or the 20× model
- A missing company, or a missing metric on one that's here
- A counterargument worth publishing next to the claim
What happens then
- Every submission gets a status: submitted, under review, verified, rejected, or published
- Verified corrections are made and dated in the correction log, which is public and append-only
- Rejected ones are logged too, with the reason
- Nothing is quietly edited. If a number changes, the old one stays visible
Send a correction Correction log Methodology
Right now this runs on email and a public log, because the site has no backend and I would rather admit that than pretend there's a review queue. If submissions outgrow an inbox, the queue gets built and the log stays.
Do a piece of it yourself
The Rebecca Challenge
Pick a company — ideally the one you work for. Everything on this site was found by one person reading public filings, and it takes about twenty minutes per company.
- Search EDGAR for the company's most recent DEF 14A (the proxy statement).
- Search inside it for "pay ratio". Nearly every US public company states its chief executive's total compensation, its median employee's total compensation, and the ratio, in one paragraph.
- Note how they identified the median employee — whether part-time pay was annualized, which countries were included. That caveat matters as much as the number.
- If you can find them, add: employee ownership share, buybacks, net income, revenue.
- Send it with the filing link and the page or section.
Accepted submissions become part of the public dataset, credited unless you ask otherwise.
Open by default
Build with the Ledger
The whole dataset is two files. No key, no login, no rate limit. Take it, check it, build on it, or use it to argue against the site — all of that counts as the project working.
ledger.json ledger.csv Methodology The design Corrections
Every figure in the data carries a status field — verified, calculated, estimate, proposal — and a source with the form type, filing date, URL and retrieval date. If you republish a number, carry its status with it. Attribution: The Rebecca Ledger, rebecca.wilde.energy.
Why Rebecca
A ledger is where you write down what is owed.
Rebecca Sue Smith was my best friend. She taught me to look at how people are treated and refuse to call it normal just because it's common. She isn't here to see this. She'd have wanted it built anyway. — Ryan Wilde
The question the ledger asks is narrow on purpose: what do companies owe the people who make them work, and what does the public have a right to see about how that gets decided? It is one person with a laptop reading public filings. That's the weakness and it's also the point — every number here was findable by anyone.
Who gets paid. Who owns it. Who's watching?
Start with the company you work for.