Open protocol · CB-P v1

Everyone measures carbon.
No one measures how much these numbers can be trusted.

Carbonbench is the accuracy layer for corporate carbon disclosures: open formula, cryptographic seal, cross-check against official records — and, a first for the sector, our own published margin of error.
—Sealed records
—Companies
—Jurisdictions
—Scored disclosures
—Official-registry records
6.6%Published margin of error*
Why an accuracy layer

Comparability is an assumption, not a measurement

Most companies never disclose the emission-factor value behind their numbers, so their totals cannot be recalculated. Where they do, reports for the very same national grid use different factors — in Türkiye, anywhere from 0.42 to 0.48 kgCO₂e/kWh. And many listed companies publish no report at all. Every ranking built on top of these numbers rests on an unmeasured assumption of trust.
—

of records do not disclose the emission-factor value — these disclosures cannot be independently recalculated.

64%

of Borsa İstanbul companies have no report we could find — every scored universe consists only of those who chose to publish a report.

+9.9 pts

The measured effect of independent assurance — when compared within the same market, assured disclosures come out a median of ~10 points more defensible (method and stratum table).

Method

Four layers, one open formula

The score is computed with one published formula — the same input leads any third party, working it out by hand, to the same score. There is no black box.
1
Collect and seal Public reports are compiled by machine reading; every record is sealed with SHA-256, and no version is ever deleted. Every value carries a page reference in the source PDF.
2
Cross-checked against official records Official registries — EPA GHGRP (US), EU ETS and EU MRV (Europe), UK ETS, Australia NGER, Japan EEGS, Korea GIR — with independently verified emissions; the official SBTi list; grid factors. If a disclosure sits markedly below the official record, the measurable deviation feeds into the score.
3
Rule engine + second reading Unit errors, scope contradictions, missing dual reporting are caught deterministically. Every medium-confidence reading is confirmed by an independent second reading; contradictions go to human review.
4
Score it, and state the band Accuracy Estimate + 95% confidence band. No score is invented for those who provide no evidence; the band narrows as evidence arrives. The score history and the reason for any change are open on every profile.

Our own processing cost is about 3.5 US cents per report (only the second-reading step uses a paid model) — which is why the data can stay free to read.

Principles

What we deliberately do differently

Traditional ESG data provider

  • Produces a model estimate for a company that does not disclose
  • The methodology is summarized; the formula is not shared
  • The margin of error is never published
  • The source page is not shown
  • The reason for a score change is not explained

Carbonbench

  • No evidence, no score — we honestly write "no disclosure"
  • Formula, tables and constants are public
  • Our margin of error is published — lower bound 6.6%, the upper bound will arrive with the golden set
  • Every value carries a source PDF + page number
  • Score history + reason ("formula version changed / universe grew")

* The margin of error is the contradiction rate among records resolved in the independent second-reading pass — since it is a machine-to-machine agreement, this is a lower bound. The upper bound will be published once the human-labeled 200-report golden set is complete. Details: Press Room.

Cases

Numbers alone mislead

Don't weigh steel against a bank

A steel producer landed in the "bottom 14%" on absolute emissions. Per tonne of steel, it emits 0.63 tCO₂e — half the global average. In heavy industry, the physical-intensity gauge outranks the absolute percentile.

Saying "we're SBTi-aligned" isn't enough

An energy company's report says its targets are science-based, but its measured pace is −2.1%/yr — the sector pathway requires −7.4%/yr, and the company is not on the official SBTi list. We look at the official list and the measured pace, not the wording.

We catch our own mistakes too

A cross-check against an official registry showed one of our records to be 300,000 times too low — the error was ours (a misread unit), not the company's. The record was corrected, the company was not penalized, and the case was logged in our error matrix. The company never pays for our reading error.

Who it's for

For everyone pricing the trust behind a disclosure

Audit & assurance teams

Pre-audit screening: unit errors, scope contradictions, internal inconsistencies — automating what your team does by hand in the first week.

Banks & sustainable finance

Model card, score history and band narrowing: traceable KPI infrastructure for sustainability-linked loans.

Importers & exporters preparing for CBAM

The defensibility of a supplier's disclosure is the precondition for savings against the default-value penalty. The Accuracy Estimate is a company-level estimate, not product-level or CBAM evidence. Screen your suppliers →

SME suppliers

Your customer asked about Carbonbench? Get listed for free, in four steps and without jargon. Start here →