Documentation

Methodology

How Prasine Index assesses EU corporate sustainability claims — pipeline architecture, data sources, verdict definitions, and scoring logic.

Pipeline architecture

Each assessment runs a 7-agent pipeline. Every agent has a typed Pydantic contract at input and output — no unvalidated state crosses a boundary.

#AgentWhat it does
1DiscoveryFetches the claim source URL and extracts all sustainability assertions from the page. Filters to substantive environmental claims — ignores boilerplate and navigation copy.
2ExtractionStructures each claim as a typed record: claim text, category (emissions, target, process, certification, offsetting), and source attribution.
3ContextRetrieves company-level context: sector, CSRD obligation status, SIC code, prior enforcement history, and known lobbying ratings. Sets the baseline for scoring.
4VerificationParallel fan-out across all 22 data sources. Each source returns typed Evidence records with a confidence value and a URL. Uses LangGraph for managed parallelism — the only place a framework is used.
5LobbyingDedicated pass for influence data: LobbyMap rating, EU Transparency Register registration, GOGEL/GOGET extraction presence. Lobbying contradiction is an independent scoring trigger.
6JudgeScores each claim 0–100. Confidence-weighted aggregate across all evidence items. Assigns verdict tier. Flags data gaps where regulatory ground truth is absent.
7ReportWrites the full assessment: evidence chain with citations, dimensional scoring table, key finding, data gaps, methodology note, and trace ID for reproducibility.

The pipeline is single-threaded at the assessment level — one company at a time. Verification runs all 22 sources in parallel within that run. Claims within a company are scored sequentially with shared context.

Data sources

Every evidence item in a Prasine report is drawn from one of these 22 EU open data sources. No paywalled databases. No hallucinated citations — every source is fetched at run time and the URL is included in the evidence record.

SourceWhat it provides
EU ETSVerified installation-level CO₂ emissions from the EU Emissions Trading System. Ground truth for industrial emitters obligated under ETS.
E-PRTREuropean Pollutant Release and Transfer Register — facility-level pollution data across EU member states.
EDGAR (JRC)EU Joint Research Centre gridded GHG emissions — used where ETS coverage is absent.
EEA National InventoriesEuropean Environment Agency country-level GHG inventories. Sector-level emissions trends.
EurostatEU statistical office — energy mix, production volumes, and sector emissions aggregates.
Climate TRACEIndependent satellite-derived emissions estimates. Used to cross-check self-reported figures.
SBTiScience Based Targets initiative — whether the company has a validated, approved, or committed net-zero or 1.5°C target.
TPI (Transition Pathway Initiative)Independent assessment of company management quality on climate transition. Tracks emissions performance vs. sector benchmarks.
CA100+Climate Action 100+ benchmark — net-zero alignment score, disclosure quality, and short-term target setting.
CDPCarbon Disclosure Project open dataset — company-disclosed emissions, targets, and climate governance data.
LobbyMapInfluenceMap corporate climate lobbying scores. The single strongest independent CONFIRMED trigger: a D or D+ rating demonstrates active obstruction of climate policy.
EU Transparency RegisterConfirms whether the company is a registered EU lobbyist and its declared lobbying expenditure.
EUR-LexEU legislative database — cited when a claim conflicts with current or forthcoming EU law (CSRD, Green Claims Directive, EmpCo Directive, EU Taxonomy).
Enforcement rulingsNational advertising authority and regulatory rulings (ASA UK, Reklamnämnden SE, ACM NL, others). Prior violations are scored as independent evidence.
GOGELGlobal Oil and Gas Exit List — upstream oil and gas expansion plans. Active expansion contradicts credible net-zero claims.
GOGETGlobal Oil and Gas Extraction Tracker — production volumes and new field approvals.
Global Coal Exit List (GCEL)Coal power and mining exposure. Active coal expansion is incompatible with Paris-aligned claims.
Global Coal Plant Tracker (GCPT)Plant-level coal capacity in operation and under construction.
Europe Gas Tracker (EGT)Gas infrastructure under development — pipelines, LNG terminals, gas plants. New gas locks contradict transition claims.
Banking on Climate ChaosFossil fuel financing by major banks — relevant for financial sector claims and credibility of transition pledges.
EU Innovation FundEU-funded decarbonisation grants. Receipt of transition funding is contextual evidence, not a verdict trigger.
Source documentThe claim's own source URL is fetched and analysed — the claim text, its context, and any referenced commitments are extracted directly.

Scoring

Each claim receives a score from 0 to 100. Higher scores indicate stronger evidence of greenwashing. The score is a confidence-weighted aggregate across all evidence items retrieved for that claim.

ScoreVerdictWhat it means
0 – 20SubstantiatedAvailable evidence supports the claim. No significant contradicting data found across the 22 sources. The claim is defensible under current EU standards.
21 – 40UnverifiableInsufficient public regulatory data to verify or refute the claim. The claim may be accurate, but cannot be independently confirmed through EU open data.
41 – 60MisleadingEvidence contradicts the claim, but not conclusively. The claim overstates, omits material context, or uses framing inconsistent with the underlying data.
61 – 80Likely greenwashingStrong contradicting evidence from multiple sources. The claim conflicts with regulatory data, lobbying conduct, or target credibility in ways that are difficult to reconcile.
81 – 100Confirmed greenwashingDefinitive contradicting evidence. One or more confirmed triggers apply (see below). The claim cannot be reconciled with independently verified facts.

Confirmed triggers

Any of the following, if confirmed, triggers CONFIRMED status regardless of other evidence:

TriggerLogic
LobbyMap D or D+An "Obstructive" or "Highly Obstructive" climate lobbying rating means the company actively opposes or delays the climate legislation its own green claims depend on. Internal contradiction at the conduct level.
Prior enforcement rulingA national advertising authority or regulatory body has ruled a similar or identical claim from this company misleading. Establishes pattern and prior notice.
Active fossil fuel expansionCompany has upstream oil, gas, or coal expansion planned or underway (per GOGEL/GOGET/GCEL/GCPT) while making Paris-aligned or net-zero claims. Structural contradiction between capital allocation and stated direction.
Offsetting basisNet-zero or carbon-neutral claim relies primarily on carbon offsetting, which is expressly excluded from acceptable net-zero bases under the EmpCo Directive (EU 2024/825) and the EU Taxonomy.
No SBTi target + net-zero claimCompany claims net-zero alignment with no approved, committed, or validated SBTi target. No credible independent verification pathway exists.
Verified emissions gapIndependent satellite or regulatory emissions estimates significantly exceed company-disclosed figures, and the discrepancy is material to the claim being assessed.

Multi-claim assessments produce a company score from the three highest-scoring claims (confidence-weighted), floored at the bottom of the worst claim's verdict band — a confirmed finding cannot be averaged away by milder claims. The company-level verdict is always derived from this final score, so score and verdict can never disagree.

What this is and isn't

Not a legal determination

A Prasine Index assessment is an evidence-based research output, not a legal opinion. It identifies public data that contradicts a company's stated claims. It does not constitute legal advice, and a CONFIRMED verdict does not mean the company is in breach of law — it means the evidence record is inconsistent with the claim at a level that warrants further scrutiny.

AspectDetail
ModelReports are generated by a multi-agent pipeline using Claude (Anthropic). The model synthesises and scores evidence — it does not generate or fabricate data. Every source cited in a report was fetched and retrieved at run time.
CitationsEvery evidence item includes a source URL, a confidence value (0.0–1.0), and a supports/contradicts/context classification. Evidence items with no retrievable URL are flagged as unverifiable and do not increase the score.
Data gapsEvery assessment discloses where regulatory ground truth is absent — when a company is not in a mandatory database, when data is paywalled, or when disclosure is incomplete. Data gaps are listed explicitly, not silently omitted.
Trace IDEach report includes a trace ID: a hash of the pipeline run. The same company and claim URL run through the same pipeline version will produce an identical trace ID, confirming reproducibility.
ScopeCurrent coverage is limited to environmental (E) claims. Social (S) and governance (G) claims are outside scope. CSRD ESRS E1 (climate change) is the primary lens.
Paywalled sourcesPrasine does not access paywalled databases. Where a relevant source requires registration or payment (GOGEL full dataset, some regulatory portals), this is disclosed in the data gaps section.

Questions about the methodology, a specific scoring decision, or how Prasine could support your work? Get in touch.