ESG measures what companies say. We measure what they do.
ESG ratings are built on corporate questionnaires and voluntary disclosures. Across providers, they correlate at just 0.54 — compared to 0.99 for credit ratings (Berg, Koelbel & Rigobon, MIT Sloan, 2022). That data is priced in at publication and has weak predictive power for returns.
Our data comes from sources companies can't control: court filings, regulatory enforcement actions, and investigative journalism. The gap between what ESG reports and what actually happened is structured, measurable, and quantifiable.
6,000+ companies. 11 risk dimensions. Every score sourced to a specific filing. API and SFTP delivery. Built by a team from Goldman Sachs, Deutsche Bank, and Arthur D. Little.
Where ratings and reality diverge
ESG providers disagree nearly as often as they agree. The same company can be an MSCI “Leader” and a Sustainalytics “High Risk.” That's not a data source — it's noise. Here are four cases where ESG ratings and the documented record contradict each other.
| Company | ESG Rating | What the record shows | Our Score |
|---|---|---|---|
| Schlumberger | MSCI AA “Leader” | Designated international war sponsor by Ukraine (April 2023). Continued Russian operations while 1,000+ Western companies withdrew. Russian employees received military draft notices through the company. | -90 War & Weapons |
| Meta | MSCI AA “Leader” | ~€4 billion in ethics-related fines over 3 years. €1.2B GDPR violation. $1.4B facial recognition settlement. Sustainalytics rates the same company “High Risk.” | -70 Data Privacy |
| Microsoft | MSCI AAA (highest) | 11.9% of annual revenue from defence contracts. $22 billion HoloLens contract with U.S. Army. Sustainalytics: “Low Risk.” | -50 War & Weapons |
| NVIDIA | In 8 of 10 “sustainable” ETFs | Emissions rose 87% in a single year. Military AI applications documented across multiple defence programmes. | -70 War & Weapons |
These aren't edge cases. In our analysis of the S&P 500, 72.5% of companies scored negative on average ethical conduct. Mean score: -6.2. The divergence between ESG labels and documented conduct is systematic.
$30 billion flowed into ESG ETFs in 2025. Here's what's inside them.
We scored the top holdings of the 10 largest ESG ETFs by AUM — including iShares ESGU ($12.7B), Vanguard ESGV ($9.1B), and SPDR EFIV ($4.3B).
| Holding | Weight | Mashinii Avg Score | Worst Dimension |
|---|---|---|---|
| Apple | ~7% | -3 | Climate: -40 |
| Microsoft | ~6% | -5 | Weapons: -50 |
| NVIDIA | ~6% | -10 | Weapons: -70 |
| Amazon | ~3% | -25 | Workers: -50 |
| Meta | ~2% | -39 | Privacy: -70 |
| Alphabet | ~2% | -29 | Weapons: -80 |
Average score across top 10 holdings: -13.6. Eight of ten score negative. Only Visa (+9) and Mastercard (+1) score positive.
ESGU holds ~300 of the S&P 500's ~500 companies. Overlap with a standard index fund: over 90%. Fee premium: ESGU charges 0.15% vs VOO at 0.03%. The ESG label doesn't change portfolio composition meaningfully — but it does change investor perception. That perception gap is structured and measurable.
Adversarial conduct data as a risk factor
Traditional ESG sources 50–80% of its data from corporate self-reporting. Our data sources 100% from adversarial or independent channels:
Court filings
Settled litigation, ongoing proceedings. Docket numbers included.
Regulatory enforcement
OSHA, HSE, GDPR, SEC, NLRB, CFTC penalties with enforcement dates.
Investigative journalism
Reuters, Financial Times, Wall Street Journal, Guardian. Publication dates cited.
NGO field reports
Amnesty International, Human Rights Watch, Greenpeace. Documented findings.
Each company scored across 11 dimensions on a -100 to +100 scale. Dimensions map to identifiable risk categories: weapons and military exposure, regulatory and environmental risk, labour and supply chain risk, litigation and privacy risk, governance and corruption.
Self-reported data is priced in at publication. Adversarial data enters the market through news cycles with variable lag. The 0.54 correlation between ESG providers means the market has no consensus on what “ESG risk” even means. A signal built on verifiable public records offers a testable alternative.
Coverage, delivery, methodology
| Coverage | 6,000+ public companies. Primary universes: MSCI World, FTSE 350, S&P 500, Euro Stoxx 600. |
| Dimensions | 11 risk dimensions per company, scored -100 to +100. Not blended into a single score. |
| Sources | Court filings, regulatory databases, investigative journalism, NGO reports. 100% adversarial — zero corporate self-reporting. |
| Update frequency | Within 48 hours of new evidence. Not quarterly. Not annually. |
| Delivery | REST API, SFTP, flat file export. |
| Format | JSON (API), CSV (flat file). |
| Historical data | Point-in-time reconstruction available on request for backtesting. |
| Methodology | Published, stable, auditable. No black-box ML scoring. Every score links to its source document. |
Test the signal yourself.
Request a sample dataset — 100 companies, all 11 dimensions, full source citations. Run your own regressions. No commitment.