# What does "audit-grade carbon data" actually mean? Canonical: https://senseible.earth/climate-intelligence/audit-grade-carbon-data-what-it-actually-means Published 2026-05-21. "Audit-grade" is the most overused phrase in carbon software marketing. It has a precise meaning, and most products that claim it do not meet it. The phrase has five concrete properties. A dataset either has all five or it is not audit-grade. ## The five properties 1. **Traceability.** Every final number can be traced back to a specific source document or measurement, with no broken links in the chain. 2. **Methodology pinning.** Each record is tagged with the exact version of the methodology used to compute it (e.g. GHG Protocol Corporate Standard 2015 amendment, IEA grid factors 2023). 3. **Immutability.** Once recorded, the source document and its computed output cannot be silently altered. Changes leave an audit trail. 4. **Reproducibility.** Anyone with the same inputs and the same methodology version produces the same output. No black-box AI in the math. 5. **Explainability.** Any number can be decomposed into its inputs and the operations performed on them, in plain language a non-technical auditor can follow. ## Why each property matters ### Traceability Auditors work backward. They pick a final emission figure, then trace it back through the calculation chain to the source. If the chain breaks (a missing invoice number, an orphaned spreadsheet row), the auditor cannot sign. Modern MRV stores the source document, the extracted fields, the methodology applied, and the output as one linked record. ### Methodology pinning GHG Protocol guidance updates. IEA grid factors update annually. Emission factors for specific gases are revised. A number computed under the 2020 factors is not the same as a number computed under the 2024 factors. Audit-grade data carries the version it was computed under, so historical numbers remain reproducible even after the methodology has moved on. ### Immutability Spreadsheets are the canonical anti-pattern. Anyone with edit access can change a number and no one knows when or why. Audit-grade systems use append-only logs, SHA-256 hashing of source documents, and version control on every record. Senseible records the document stub with its hash *before* processing begins so the audit trail exists even if processing later fails. ### Reproducibility If a third party cannot rerun your calculation and get the same number, your number is not audit-grade. This is a major reason black-box AI-generated emission estimates fail audit: the same input on the same day can produce different outputs depending on the model state. Deterministic math (HSN-to-scope, factor multiplication, GHG Protocol conversion) is reproducible. AI-generated estimates are not. ### Explainability An auditor must be able to ask "how did you arrive at 12.4 tCO2e for this purchase?" and receive an answer like this hypothetical one: "Invoice INV-2026-0421, 8 tonnes of HSN 7208 (hot-rolled steel), classified as Scope 3 Category 1 (purchased goods) by the GHG Protocol mapping, multiplied by the supplier-specific factor of 1.55 tCO2e/tonne." Anything less is not explainable. ## What is *not* required for audit-grade - **Third-party assurance.** Useful and often legally required, but assurance is the *recognition* of audit-grade data, not its definition. Data can be audit-grade before any auditor sees it. - **Blockchain.** Achieves immutability but is not the only way to do so. A properly designed RLS-enforced append-only database with hashed records is equally valid. - **Real-time IoT sensors.** Useful for some categories. Most MSME emissions can be made audit-grade from existing invoices and bills. ## How to test whether a vendor's claim is real Ask these five questions. If any answer is unclear, "audit-grade" is marketing copy. 1. *Show me the methodology version pinned to a record from 18 months ago. What changes if I rerun it today with current factors?* 2. *Show me the source document for this number. Can you prove it has not been altered since ingestion?* 3. *Walk me through how a 12.4 tCO2e figure was computed, in plain language.* 4. *If your AI model is retrained tomorrow, will my historical numbers change?* 5. *Can a different engineer at your company re-derive the same number from the same inputs?* The acceptable answers are: full version history, SHA-256 hash on the source document, a complete derivation chain, "no: math is deterministic", and "yes, deterministically". ## Why this matters for MSMEs specifically MSME owners often think audit-grade is "for big companies". It is not. Three forces are pulling audit-grade into MSME territory: - **EU CBAM** requires EU importers of covered goods to declare embedded emissions, so they ask exporters for verifiable installation-level data (or fall back on default values). - **Green loans** increasingly require third-party-verified baselines for the preferential rate to hold. - **Tier-1 buyers** are pushing verifiable-data requirements into their supplier codes of conduct. An MSME that built its baseline on non-audit-grade data is likely to have to rebuild it at the worst possible time: when a buyer or banker asks. Building audit-grade from day one is the cheaper path. ## What Senseible enforces by default - SHA-256 hash on every ingested document. - Methodology version pinned per record (GHG Protocol, IEA factor vintage, CBAM regulation version). - Append-only audit ledger. - Deterministic math; AI restricted to OCR and field extraction. - Per-record explainability surfaced in the verification UI. These are not options. They are the default because anything less is not audit-grade. ## FAQ **Is my Excel spreadsheet audit-grade?** Almost never. Excel fails immutability and version pinning by design. It can be the *intermediate* tool feeding an audit-grade system, but not the system of record. **Does audit-grade cost more?** Not necessarily. Audit-grade software does not have to be rebuilt at the audit step. The expensive path is to discover at year-end that the data fails. **Can third-party verification fix non-audit-grade data after the fact?** No. A verifier signs only what is traceable. Non-traceable data gets rejected; that is the verifier's whole job. ## Related guides - [How can we trace every emission number back to raw sensor or invoice data?](https://senseible.earth/climate-intelligence/how-to-trace-emission-data-to-source) - [What is Senseible and How is It Different from Sensibull?](https://senseible.earth/climate-intelligence/what-is-senseible-different-from-sensibull) - [What Documents Do I Need for Carbon Verification?](https://senseible.earth/climate-intelligence/documents-needed-carbon-verification)