Methodology

How Terra Detects Greenwashing

Terra uses an NLP-powered analysis engine to cross-reference corporate sustainability claims against independent data sources. Every score is evidence-backed, weighted by category, and reproducible.

Data Sources

Terra ingests and cross-references data from multiple independent sources to ensure findings are grounded in verifiable evidence, not a single dataset.

SEC EDGAR—10-K, 10-Q, and DEF 14A filings for financial and environmental disclosures
Banking on Climate Chaos—Annual report on fossil fuel financing by the world's largest banks
CDP (Carbon Disclosure Project)—Standardized corporate environmental disclosures and scores
EPA GHGRP—EPA Greenhouse Gas Reporting Program — facility-level emissions data
Company ESG Reports—Self-published sustainability and ESG reports, press releases, and investor presentations

Analysis Pipeline

Each investigation follows a five-stage pipeline designed to systematically separate genuine sustainability efforts from greenwashing.

01

Document Ingestion

SEC filings (10-K, 10-Q), ESG disclosures, sustainability reports, and press releases are collected and parsed into structured text. For SEC-registered companies, filings are retrieved directly from EDGAR.

02

Claim Extraction

NLP models identify and extract specific sustainability claims — emissions targets, renewable energy commitments, net-zero pledges, supply chain standards, and marketing language. Each claim is categorized and tagged.

03

Reality Cross-Reference

Extracted claims are compared against independent databases: EPA emissions records, Banking on Climate Chaos financing data, CDP disclosures, and tracked corporate pledges. Discrepancies are flagged automatically.

04

Contradiction Mapping

Each discrepancy is analyzed for severity (Minor, Moderate, Major, Critical) based on the magnitude of the gap between claim and reality, the materiality of the issue, and the availability of counter-evidence.

05

Weighted Scoring

Contradictions are weighted by category importance and severity, then aggregated into a composite Greenwashing Score (0-100). Reports are generated with audience-specific summaries for investors, regulators, boards, and consumers.

Scoring Categories

The Greenwashing Score is not a single metric — it is a weighted composite across six categories. Categories are weighted by their materiality to actual environmental impact.

25%
Emissions—Scope 1/2/3 emissions claims vs. reported data and third-party estimates
20%
Financing—Fossil fuel financing, divestment pledges, and capital allocation patterns
20%
Targets—Net-zero timelines, interim milestones, and progress against stated goals
15%
Supply Chain—Upstream/downstream sustainability claims and supplier audit disclosures
10%
Marketing—Public-facing sustainability messaging vs. internal practices
10%
Lobbying—Political donations, trade associations, and regulatory lobbying positions

Scoring Rubric

0-30
Low Risk—Claims are largely consistent with available data. Minor discrepancies only.
31-60
Medium Risk—Notable gaps between claims and reality. Some commitments unmet or vague.
61-80
High Risk—Significant contradictions. Major claims undermined by available evidence.
81-100
Critical—Systemic greenwashing. Claims are fundamentally at odds with company actions.

Validation & Limitations

Multi-source verification

Every contradiction requires corroborating evidence from at least one independent source. Claims are not flagged based on a single dataset — cross-referencing reduces false positives and ensures findings are defensible.

Severity calibration

Severity ratings account for the magnitude of discrepancy, the materiality of the claim (e.g., Scope 1 emissions vs. marketing language), and whether the company has disclosed corrections or updates.

Known limitations

Analysis depends on publicly available data. Private companies with limited disclosure may receive incomplete assessments. Scores reflect contradictions in public claims — they do not capture undisclosed initiatives or in-progress improvements not yet reported.

Team

Built by Jean Lin (Wharton, University of Pennsylvania) and Abhi Chundru (University of Michigan). We're current juniors on leave to work full-time at an AI research lab backed by leading groups including Excel and Google DeepMind, where we investigate the boundaries and real-world capabilities of artificial intelligence. As Members of Technical Staff focused on reinforcement learning environments, we build high-fidelity training and testing systems that improve model performance on demanding tasks and support multi-million-dollar research partnerships.

We bring both AI depth and finance rigor. Abhi has experience spanning distressed credit research and private equity, and spent four years building Sidereal, a biodiesel project recognized by the U.S. Department of Energy, including a presentation at an energy tech conference in Houston (2024). Jean has worked at a $21B AUM private equity firm and founded a sustainability startup that uses blockchain to make environmental impact more measurable and accountable.

Environment and sustainability have been core to what we've wanted to build for years, and we're grateful to express that here.