What the S&P 500's highest-paid CEOs took versus what they delivered: descriptive, primary-sourced, reproducible.
Interactive front-end over the frozen v2.2 S&P 500 dataset · data as of 2026-08-09 · concept record: DOI 10.5281/zenodo.20680108 · NM AI Research, ORCID 0009-0003-4213-7769 · mixed open licences
Licence: this interactive and the analysis files are CC BY 4.0. universe_sp500.txt is CC BY-SA 4.0. AI disclosure: parts of this text were artificially generated with AI assistance and reviewed by the author. The model and conflict are named in the Disclosure note.
| # | Company | CEO | Pay | Peer-relative TSR | Realized ÷ Granted | Board target | Note |
|---|
What the toggle shows. Granted is the Summary Compensation Table figure the press quotes. Realized (CAP) is Compensation Actually Paid, the SEC measure that re-marks equity to the share price; it is an accounting fair value, not cash-in-pocket. Flip the toggle and the league table reorders: that reordering is the central finding (F1).
Peer-relative TSR is the company's cumulative total shareholder return ($100 invested at the window start) minus the board's own chosen peer index. A lag can mean the benchmark ran hot, not that the company cratered, e.g. semiconductor peer indices inflated by Nvidia. Green = beat the chosen peers; red = lagged them.
Attribution: TSR is not caused by the CEO (macro, sector, luck, predecessor). This is a co-incidence of pay and peer-relative return, not causation, never read it as "overpaid." CAP caveat: swings with the share price; not cash received. Peer-group choice is the company's own and drives the gap. Sample: this is a 495-issuer view from 498 resolved securities in a dated 503-security universe. Five securities are explicit placeholders. The separate board-target layer contains 29 original S&P 100 cases and 18 findings extensions, all primary-verified. The observed peer-beat share is 227/493, or 46.0%. No statistical baseline or causal model was run.
Verification. Each row carries the SEC accession or accessions used. The 47 board-target cases have primary-verified accession, form, document, date, period and metric metadata in board_targets.csv. The 503-security universe is adapted from Wikipedia revision 1364118843 under CC BY-SA 4.0. Five securities are explicit placeholders. No statistical baseline or causal model was run.
Disclosure. The author holds no direct position in any company named. A workplace defined-contribution pension may hold some indirectly through pooled funds, not directed by the author. No third party funded or directed this work. Anthropic Opus 4.8 assisted with retrieval and drafting. OpenAI GPT-5.6 Sol checked and revised v2.2. OpenAI is a model provider and competes with some companies discussed or their products; it is not an issuer in this dataset.
Reproducibility. Built free from SEC EDGAR DEF 14A Pay-versus-Performance disclosures (Item 402(v)); 495 issuers and 2,768 populated company-year rows. python3 assemble_sp500.py uses the Python standard library. This page reads the frozen scorecard_sp500.csv verbatim. Row-level SEC accessions are included in the CSV.