Legal Notice

The English text of this page is the legally binding version.

This page explains what Carbonbench is and explicitly defines what it is not. Read it before using our scores.

What We Are Not

Not an audit, assurance, or rating service

Carbonbench is not an independent audit, an assurance service (assurance engagement), or a credit rating agency. The Accuracy Estimate, Transparency Score, Target Quality, and similar indicators published on the site are a statistical plausibility analysis produced by previously published deterministic rules applied to public disclosures (CB-P v1). The word "Score" here is an indicator/estimate, not a letter-grade-style rating — we deliberately do not assign letter grades.

Relationship to ISAE 3410

Does not substitute for on-site verification

Our analyses are not a study conducted under ISAE 3410 — the assurance standard for greenhouse gas statements — or similar on-site audit/assurance standards, and they do not substitute for it. ISAE 3410 assurance that a company obtains internally or from an independent assurance provider constitutes evidence separate from and superior to our analysis; it is additionally shown on the profile with an "Assured" label.

Coverage and Selection Bias

Universe = companies that publish public reports

The scored universe consists of companies that have chosen to publish a sustainability/carbon report publicly. Companies that do not publish a report are out of scope, and this creates a structural selection bias: more transparent companies already tend to enter the assessment. We do not hide this part either — we separately publish how many companies remain out of scope (see the Press Room, finding "The silent"). Official-registry records (EPA GHGRP, UK ETS, EU MRV and others) are the exception: they come from regulator data, not from a company's choice to publish.

A company's absence from the list says nothing about that company's carbon performance — it only means no report was found in public sources.

Our Own Margin of Error

Discrepancy in independent second reading: 6.6% — this is a lower bound

6.6%

Adjudicated records only, machine-to-machine agreement. Current counts are on the Press Room page.

We compare every medium-confidence extraction against a second reader who has never seen the record. The discrepancy rate on adjudicated records came out to 6.6%. Honesty note: this is a machine-to-machine agreement — a metric that cannot capture the chance of two readers making the same mistake — and therefore a lower bound on our margin of error, not the upper bound. The upper bound will be published separately once a human-labeled gold set is completed. Full matrix: error matrix (docs/error-matrix.md).

Asking a data provider "what is your margin of error?" usually gets no answer. Ours is published and kept up to date.

Reading Error Correction Policy

The company does not pay the price for our mistake

If a record is found to have been misread by us (the report is correct, our extraction is wrong), the record is corrected and re-sealed as a new version — the old version is not deleted, it remains as "superseded." Such corrections are never used to affect the score negatively; a correction merely reflects the accurate reading. In contrast, if an inconsistency within the report itself is detected (e.g., the same emissions being double-counted under two different line items), this is separately flagged and, if necessary, reflected in the Accuracy Estimate — because this is a problem of the disclosure itself, not of ours.

Dispute Channel

10-business-day SLA

You can submit your dispute regarding any record through the Support page or at support@carbonbench.org. Our commitment for the first response to disputes is 10 business days. Corrections that follow a dispute appear in the profile's public version history, and no version is ever deleted; a consolidated public dispute log is planned.

"No Disclosure" Label

Not a claim of certainty, a search result

The "no disclosure" or "n/a" (not assessed) label you see on a profile does not mean "this company definitely does not disclose emissions." It means: no measurable emissions inventory or sufficient evidence was found in the public sources we scanned. When there is no evidence, a score is never invented — this is a deliberate design decision. When a company submits evidence (via the claim flow), the record is updated and the score is unlocked.

Terms of Use Privacy Open API Methodology
Last updated: September 29, 2026 · version 1.1

Summary

Carbonbench is not an audit, assurance engagement, or credit rating agency, and does not substitute for ISAE 3410 or other field-verification assurance standards. Scores are a statistical plausibility indicator (CB-P v1), not a letter-grade rating. The scored universe is limited to companies that chose to publish a report — a structural selection bias we disclose and separately track ("silent" companies). Our own error rate, measured via independent double-reading, is 6.6% on adjudicated records — published as a floor, not a ceiling, pending a human-labeled gold set. Our own reading errors are corrected without penalizing the company; inconsistencies in the company's own disclosure are flagged instead. Disputes get a first response within 10 business days; resulting corrections appear in the profile's public version history. A "no disclosure" label means "not found in public sources," not a claim of certainty.