Jessica Man, Martin Kleppmann.
University of Cambridge. E-Energy ‘26 — 17th ACM International Conference
on Future and Sustainable Energy Systems, Banff, 22–25 June 2026.
doi:10.1145/3744255.3811720. CC BY 4.0.
Sources:
sources/whitepapers/2026-08-17-emission-impossible-verifiable-cloud-emissions.pdf
(immutable original) and its Markdown twin
sources/whitepapers/2026-08-17-emission-impossible-verifiable-cloud-emissions/2026-08-17-emission-impossible-verifiable-cloud-emissions.md
(supplied already converted by Martin Pompéry, 2026-08-17 — no Marker step
needed). Cited against the Markdown.
This is the first primary source in the wiki on Kleppmann’s own research output, and it is by the Co-PI of CCTI. Read together with Heiss et al. (2023) and the joint concept page Verifiable Carbon Accounting, which carries the criteria grid and the closing-constraint thesis. This page is the source summary only.
The claim
A data centre operator can prove to each customer that the emissions reported to that customer are accurate — “without the customer being able to learn sensitive information about the data centre operator or other customers.” Stated cost: ≈ $150/month for an operator with one million customers, plus $0.01 per customer who requests a report; proofs of 37 KiB, verified in under a second. The authors state this is the first realisation of such a system using zk-SNARKs.
The framing is explicitly regulatory: CSRD and UK SECR mandate reporting but do not require independent audit “because such manual audits are too costly”, while full transparency is not viable because the underlying data is commercially sensitive. ⟨Inference: that is the confidentiality gap, stated as the paper’s premise rather than discovered as a finding.⟩
The key insight
“The key insight of this paper is how to ensure that the cryptographically verified data is an accurate reflection of the physical world. We partly rely on trusted data sources (such as an electricity meter with a tamper-proof hardware element), and also couple carbon accounting to financial accounting. In order to lie about its emissions, the company would then also have to lie about its finances, which raises the stakes by making misreporting a crime.”
⟨Inference: this is the sentence with the most strategic weight in the paper for SINE. ICC argues that carbon accounting should be like financial accounting. Man & Kleppmann bind it to financial accounting cryptographically, and thereby import the existing legal apparatus for financial fraud instead of building a new enforcement regime for carbon. See the ledger page.⟩
The three proofs
- Total emissions. Inside the circuit: signed smart-meter readings and signed grid carbon-intensity values are combined minute by minute over a day, signatures and timestamp consistency checked, and the total returned with the first/last timestamps and the verifying public keys. Because the circuit is the method, “the verifier knows which circuit was used in a proof, [so] it can check the accounting method” — here, location-based accounting is enforced by construction.
- Customer shares. Proves the per-customer allocations sum to the company total, “following the Greenhouse Gas Protocol’s Completeness Principle … which requires that the entire emissions from producing a product or providing a service must be allocated to customers.” Completeness alone is not sufficient: a company could invent a fake customer, dump 90 % on them and divide everyone else’s by ten. So the circuit additionally takes each customer’s invoice amount, split into compute charges and non-energy charges (licences, domain registrations), sums revenue per category, and proves that every customer’s kWh-per-$ ratio is within configurable bounds — in the prototype, within a factor of 2 of the mean, with the check skipped for free-tier-scale usage. The ratio is itself commercially sensitive (it measures a customer’s profitability), so it is never disclosed; even the bounds can stay secret if a financial auditor signs that they are reasonable. To stop the operator showing different allocations to different customers, all customer data goes into a Merkle tree whose root is published once per period.
- Per-customer proof. Recursively verifies (1) and (2), checks they are for the same reporting period, and adds the customer’s Merkle inclusion path. Because the expensive parts are proved once and reused, the marginal cost per customer report is cents.
Technical choices
- o1js (the o1Labs library built for Mina): Kimchi/Pickles over the Pasta cycle of elliptic curves — no trusted setup, so no CRS “toxic waste” to dispose of. Poseidon hashing; Schnorr signatures over Pallas.
- Recursion is the scalability mechanism: “we can prove once per reporting period that the allocation of emissions across all customers is fair, and then reuse this proof every time we need to prove the correctness of one particular customer’s emissions.”
- No blockchain required. §6, verbatim: “Our proposal does not depend on a blockchain or distributed ledger, except for publishing a single hash commitment per accounting period – a task that can also be fulfilled by a simpler system such as a transparency log.” Certificate Transparency is the named model.
- Prototype and evaluation data:
github.com/psjm3/verifiable_cloud_carbon_emissions,doi:10.17863/CAM.130100.
Stated assumptions and limits
- Trusted signing devices. The smart meter, its manufacturer, the certificate authorities and the grid operator “need to be trusted to sign only accurate data” — but “no trusted party is involved in reviewing the company’s actual carbon accounting.”
- Financial honesty. “We assume, however, that the company will not commit financial fraud.” The only financial check needed is a single figure — total revenue — checkable by customers if the accounts are public, otherwise by a financial auditor.
- Scope was deliberately narrowed. Verbatim: “Along with the assumption of a single electricity meter, this allowed us to side-step the problem of ensuring that all sources of emissions have been accounted for. However, we believe that it is possible to prove that all relevant emissions are accounted for by extending the methodology in this paper.” Hardware manufacturing, building construction and employee travel are named as future work, as is finer-grained workload timing.
- Future direction, stated: “Cryptographic verifiability could be especially valuable in industries where cheating is rife, such as agricultural production, resource extraction, and manufacturing in parts of the world where government regulation and enforcement are weak. … Our long-term goal is to work towards accurately reporting and verifying the emissions of the entire world economy.”
Relation to Heiss et al.
Man & Kleppmann §6 (Related Work), verbatim:
“Heiss et al. propose a verifiable carbon accounting method for product carbon footprints. Like us, they rely on trusted sensors to generate digitally signed readings of electricity or fuel consumption. They sum the emissions from several sources in a zk-SNARK implemented in ZoKrates, similarly to our total emissions proof (albeit only for a single time period). However, they do not address the problem of fairly allocating the total emissions across several customers. They do not use proof recursion and rely on a blockchain smart contract to verify proofs.”
The Acknowledgments thank Jonathan Heiß (with Anil Madhavapeddy) for feedback on the paper. ⟨Inference: the two papers are one research strand with a stated line of advance, not two independent efforts that happen to overlap — which is what makes a single SINE position across both coherent.⟩
Relevance to SINE
⟨Inference.⟩
- It closes the “no mechanism” half of the confidentiality gap with numbers. $0.01 per report and sub-second verification is a different kind of claim from “permissioned access, aggregation, and data minimisation”.
- It is CCTI prior art by CCTI’s own Co-PI. The CCTI stack’s WP-A evidence-type table lists exactly sensor data → claims about the physical world and recursive proofs → claims supported by another instance of the same infrastructure. This paper is both, already implemented and measured, in a domain SINE already works in.
- It defuses the blockchain objection the wiki has been carrying against Heiss’s design: the requirement is a public, append-only, non-retractable commitment, and a transparency log satisfies it.
- It leaves the trusted-meter assumption standing, which is what CCTI’s M5 gate criterion is meant to remove. Tension recorded on the VCA concept page.
- The financial coupling is the sharpest thing SINE can bring to the ICC conversation — it is an argument about enforcement, not about architecture, and ICC’s document has no equivalent.