When does a model's estimate count as assurance evidence? Traceability, attributability and contestability, with a satellite worked example
An estimate that a model produces outside the reporting entity, such as a satellite methane rate, a machine-learning plume detection or a figure a language model extracts from a report, is not audit evidence just because it is accurate. This working paper proposes three conditions under which such an estimate becomes evidence: traceability (the chain from observation to figure can be re-walked), attributability (the figure can be tied to an identifiable facility, entity and accountable actor) and contestability (the reporting entity can rebut it on stated grounds). It sets out where such an estimate sits under ISSA 5000: as information from a source external to the entity (para 90), through a practitioner's expert (paras 56–58) or a management's expert (para 92), or not at all. A worked example traces published Tanager-1 and EMIT methane plumes near an underground coal mine in Queensland's Bowen Basin to the mine's Safeguard Mechanism figures. The provider documents and versions the quantification. Attribution, annualisation and the link to the reported figure are left to the user. The reported figure sets a unit-surrender obligation, so the gap reaches the financial statements. The example illustrates a method; it is not an estimate of any mine's emissions and does not suggest misreporting. The deposit includes a standard-library Python script that regenerates every figure in the worked example from public sources. Data sources: Carbon Mapper plume catalogue (used under Carbon Mapper's Terms of Use and cited, not redistributed); Climate TRACE (release 5.12.0); Clean Energy Regulator, 2024–25 baselines and emissions data. Retrieved 17 November 2026. Version 1.1 (18 November 2026) corrects the account of processing versions. Each detection is processed once by the provider's pipeline current at the time; at the worked-example site, 19 detections over 26 months span 14 pipeline versions (3.12.0 to 3.115.2). Version 1.0 implied that published figures had been revised, which the catalogue does not show.
Authors
- Maria Prokofieva (ORCID: https://orcid.org/0000-0003-1974-3827)
Institutions
- Victoria University (AU)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-17
- DOI
- https://doi.org/10.5281/zenodo.24304924
- Primary Topic
- Auditing, Earnings Management, Governance
- Type
- article
- Field-Weighted Citation Impact
- 0.00