Where the numbers come from

Every figure we publish is drawn from a company's own public regulatory disclosures. We extract them into a structured database, and we print the accession number of the source document alongside the figure so that any reader can open the original and check it.

We do not use survey data, estimates, or third-party compilations for any published figure. If a number cannot be traced to a specific disclosure by a specific company, it does not run.

What we cover

Nearly 3,000 U.S. public companies across twelve industries: banking, real estate, software, hardware, biopharma and life sciences, medical technology and diagnostics, insurance, utilities, asset management, gaming and lodging, payments and financial infrastructure, and specialty finance.

Coverage includes every company each of those companies names as a compensation peer, held to the same standard. A peer group is an argument a committee makes about itself, and an unchecked peer group is how that argument goes unexamined.

How a story becomes a card

Detection. Deterministic queries against the structured data find the outliers: pay multiples, out-of-cycle grants, peer groups that skew large, undisclosed performance targets, sustained investor opposition.

Evidence. A candidate is rejected automatically unless it carries its source. For a claim about one company that means the accession and a link to the document. For an aggregate it means the exact cohort as data, the counts, and at least three real example filings. Nothing is ever supplied to make a card pass this gate.

Drafting. A model writes the card. Every numeric token in the draft is checked against the structured record it claims to describe, and a draft containing a number that does not reconcile is rejected rather than corrected.

Human review. A person approves or rejects every item before it publishes. There is no path from the database to a reader that does not pass a human.

Aggregate figures

Where we report a figure across many companies rather than one, we publish the cohort itself as data, not as prose: the exact filter, the numerator, the denominator, and example filings that a reader can open. The sourcing line names the number of companies in the cohort.

Fiscal years

Compensation figures are keyed to the fiscal year label a company printed in its own disclosure, not to a calendar year we assigned. Companies with non-calendar fiscal years are reported on their own calendar.

One consequence worth stating plainly: a company's discussion of an investor vote in a given year's disclosure refers to the prior year's vote. We never compare the two as though they were the same year, and neither should anyone reading a compensation discussion.

What we do not do

We do not comment on securities, and nothing we publish is a view on whether to own anything.

We do not describe people. Adjectives attach to pay designs.

We do not accept payment for coverage, and no commercial relationship can buy, delay, or preview it.

Right of reply

Any company named in a feature gets 48 hours to comment before publication. Daily cards, which report a disclosed figure and nothing else, do not carry a comment window. Replies run, and "declined to comment" runs too.

AI disclosure

The Pay Desk is AI-assisted and human-edited. Extraction, detection, and drafting are machine work under the checks above. The decision to publish is a person's.