It is an open protocol that compiles corporate carbon disclosures from publicly available sources, cross-checks them against official records, scores them with an open formula, and cryptographically seals every record. The indicators it produces are Accuracy Estimate, Transparency Score, and Target Quality. It is not an audit or assurance service; it is a statistical plausibility analysis.
No. Our analyses do not include on-site verification and are not a study conducted under assurance standards such as ISAE 3410, nor do they substitute for one. Disclosures that have received independent assurance are additionally marked as "assured" on the profile, and this constitutes evidence separate from and superior to our analysis.
Only from publicly available sources: sustainability and annual reports published by companies; official registries — EPA GHGRP, Canada ECCC GHGRP, EU ETS and EU MRV, UK ETS, Australia NGER, Japan EEGS, Korea GIR; the official SBTi target list; and published grid factors. Values from company reports carry the source host, the document's SHA-256 hash and a page reference, and the link to the document opens for signed-in users with a company email address; registry records link to the regulator's dataset. No private or leaked data is used.
Profiles, scores, and the methodology are free and require no registration. Single records and the static shard files are open without an account; the full list with scores requires sign-in and is not yet generally available — see the Open API page. See the Terms of Use page for attribution requirements.
No. Scores are indicative/estimates; they are not investment, credit, or purchase advice, nor are they a letter-grade-style rating. The comparison universe consists only of companies that publish public reports — we explicitly disclose this structural selection bias.
It measures how defensible a disclosure is in light of the evidence presented: components such as sourcing, internal consistency, alignment with official records, and assurance status are combined into a single published formula. The result is given together with a 95% confidence band — if the evidence is weak, the band is wide. Accuracy Estimate does not mean "this company emits more/less"; it means "this is how well the evidence supports the reported figure."
Accuracy Estimate measures the defensibility of the disclosure through evidence, Transparency Score measures whether the calculation can be independently reconstructed (emission factor values, methodology references, scope disclosures), and Target Quality measures the consistency of commitments with official lists and the measured pace of reduction. The three answer separate questions and must not be used interchangeably.
Yes. Components, weights, and constants are in published configuration files; the Methodology page explains the entire calculation, with a worked example (the Methodology PDF is a print edition of that page as of September 29, 2026; the page itself remains the current reference). With the same inputs, any third party can reach the same score with pen and paper; the "why" panel next to every score shows component contributions.
Our disagreement rate on records resolved in the independent second-reading round is 6.6%, and this figure is publicly disclosed. Honesty caveat: this is a machine-to-machine reconciliation, it cannot capture the possibility that both readings make the same error; it is therefore a lower bound. The upper bound will be published once the human-labeled 200-report gold set is completed.
For three reasons: new evidence arrived (the band narrows), the formula version changed, or the comparison universe grew. Profiles with two or more scored dates show the Accuracy Estimate history and label the change between the first and latest value (formula version, universe size, or data update); no silent changes are made, and no old version is ever deleted.
The seal proves that a record has gone through a specific verification process and has not changed by a single character since that day — any later change — including one made by the operator — becomes detectable. Note the current limit: the seal chain runs on a local node with a single writer, so this is cryptographic tamper-evidence, not external notarization; public-chain anchoring is on the roadmap. It does not prove that the source report conforms to physical reality; that is the assurance auditor's job.
We are aware of this risk and have taken two precautions: the within-sector percentile compares only within the same sector group; and in heavy industry (steel, cement, glass, aluminum), a physical intensity-per-ton scale is additionally shown at the top of the profile. Physical intensity is featured in the comparison band but does not enter the score; the scale bias of the absolute percentile is explicitly disclosed in the methodology.
The three most common reasons: (1) the emission factor values are not numerically disclosed in the report — the disclosure cannot be independently recalculated; (2) there are internal inconsistencies or unit/scope flags; (3) the confidence band is wide because the evidence class is weak. The "reason" panel on the profile shows this component by component. The way to raise the score does not run through us but through the report: a report that discloses its values, is source-referenced, and is assured is reflected directly in the score at the next reading.
This label means that no processed report was found in the public sources we scanned — it is not a claim that "this company does not disclose," and it says nothing about the company's carbon performance. If there is no evidence, we do not invent a score; this is a deliberate design decision. If you have a report, submit the link via Support: the record is processed and the score is unlocked.
Submit the record's link, the value you are disputing, and your basis (report page, official record, etc.) through the dispute channel on the Support page. Our first-response commitment is 10 business days. If the error is in our reading, the record is corrected, sealed as a new version, and the company is not penalized; the correction appears in the profile's public version history (a consolidated public dispute log is planned).
No. A dispute never results in a penalty under any circumstances; the correction of our reading error is not used to negatively affect the company's score. Only if the review detects an internal inconsistency within the report itself is this separately flagged — but this is a consequence of the disclosure, not of the dispute, and it would have been detected even if you had not disputed it.
Use the Claim your profile page: find your company and verify with your corporate e-mail address (or a DNS TXT record). On claimed profiles, additional evidence (report link, assurance statement, correction request) can be submitted directly; every piece of evidence submitted goes through the normal verification pipeline and is reflected in the score only to the extent the formula allows. Claiming a profile is not buying the score — there is no score improvement in exchange for payment.
You are not in the universe because no public report of yours was found — that is not a judgment. Start with the guide for SME suppliers: create a free account, build your first inventory with the reporting tool, and share it with your customer. If you already have a report online, send us its link through the Support page.
Two different things are called badges. (1) The score badge (Get Your Badge) is an embeddable image of a record's published score with a link to its verification page; it is available to claimed profiles and to the largest disclosures, and it is never sold. (2) CB-B period badges (such as Sector Leader) are given by formula — not by application or payment — to records that meet periodic performance and transparency criteria published on the Badge Protocol (CB-B) page; none has been issued yet. Badge images may be used in corporate communications with attribution; if the criteria are no longer met, a period badge is dropped in the next period.
Notifying relevant companies and granting a 10-business-day dispute window before publishing major new findings is part of our process. Disputes received within this window are reviewed before publication; the process is explained on the Support page.
The open API requires no registration; profiles, scores, and statistics are served as JSON from read-only endpoints. Documentation and the endpoint list are on the API page. Attribution and terms of use are defined in the Terms of Use; write to us for commercial redistribution.
Reports are companies' copyrighted content; we link to the original source instead of copying it. If the link dies over time, the chain of evidence is not broken: every document's SHA-256 hash is sealed, and an archived copy is retained by us; it is provided on request via the dispute/contact channel.
Links to company reports are released from the record page to signed-in users with a company email address, up to 3 distinct documents per user per day, with further daily limits per email domain and for new accounts. This limits automated harvesting of the whole link set while keeping the evidence available to analysts, journalists and the companies themselves. Everything else on the record stays open without an account: the figures, the Accuracy Estimate, the Transparency Score, the source host, the document's SHA-256 hash and the page references. Links to official registries (EPA, EU MRV, Japan EEGS and similar public databases) are not restricted. The bit-for-bit seal check needs the sealed canonical text, which is released together with the document.
Yes — with attribution. We ask that you note that scores are indicative/estimates and that the universe consists of companies that publish public reports. Press-ready data and charts are in the Press Room. When citing the methodology, cite the Methodology page with the date you accessed it and the formula version shown on each record, for example: Carbonbench (2026). Corporate carbon disclosure records, accessed 2026-09-26; formula CB-P v1 · parameter table v1.3 · Transparency v0.2. https://carbonbench.org/methodology. The Methodology PDF is a print edition of that page as of September 29, 2026; for the version in force, cite the page and the version shown on each record.
It claims nothing. A company's absence from the universe only means that its report was not found within our scope in public sources; no positive or negative inference about carbon performance can or should be drawn.
Our scores are a statistical plausibility analysis; they are not an audit opinion, an assurance statement, or an expert report, and should not be used for such purposes. Seals only prove a record's integrity (that it has not changed). See the Legal Notice and Terms of Use pages for details.