Auditing Scope 3: Why Field-Level Evidence Beats Industry Averages under CSRD

Traditional agriculture practices in BR Gunda village, Devadurga taluk, Raichur district, Karnataka, India. The image shows a farmer using bullocks (oxen) to till the land with a plough, while women follow behind manually sowing seeds of g. Photograph by Nanditha Gogate, WELL Labs, CC BY-SA 4.0, via Wikimedia Commons.

01The macro challenge: a reporting duty built on someone else’s numbers

Directive (EU) 2022/2464 requires in-scope undertakings to report sustainability information according to European Sustainability Reporting Standards [16]. Those standards sit in Commission Delegated Regulation (EU) 2023/2772, now itself under a simplification amendment [14][2]. Climate is not a footnote in that architecture. For an agrifood operator, the material climate figure is almost never the boiler house. It is cattle, fertiliser, land-use change, freight and the farms that never appear on the corporate ledger — the same food-system load Crippa and colleagues put at roughly a third of anthropogenic greenhouse gases [17]. CSRD is the entry point. The subject is whether that load is observed at origin or approximated in a way that erases regenerative practice and transfers the climate claim upstream.

That is Scope 3: indirect emissions produced along a firm’s supply chain. Stenzel and Waichman put the scale without embroidery: Scope 3 emissions “represent the majority of most firms’ carbon footprint,” and current estimations “are largely based on industry averages and other approximations,” whereas primary data shared along the chain would give “precise measures of Scope 3 emissions” [15]. The assurance layer sitting on those averages is now under the same integrity pressure as the voluntary credit market CSRD filers still lean on for “removals” colour.

"Traceability and farmer livelihoods are the foundation of a truly sustainable supply chain — enabling environmental protection, ensuring consistent quality, and building the resilience our business needs to remain competitive in an era of climate change and increasingly volatile markets." Miriam Trinker, Sustainability and Supply Chain Manager, HACOFCO, on the Honduras EUDR-ready pilot, July 2026 [5]

An emission factor assigned to “cocoa”, “soy” or “beef” cannot distinguish a shaded agroforestry plot from a recently cleared one, or a feedlot from a silvopastoral system. It also cannot see whether the same physical tonne has already been claimed by the trader, the processor and the brand. Caro, Corbett, Tan and Zuidwijk showed the allocation problem more than a decade ago: greenhouse-gas emissions in a supply chain are the result of joint production, and footprinting that ignores that jointness mis-assigns the tonne [19]. CSRD made the mis-assignment a limited-assurance artefact sitting on a food-system emissions load averages were never built to adjudicate plot by plot [17] — and Trinker’s July 2026 framing is the operator warning: without a farmer-held primary record, an assurer’s signature cannot repair a missing plot file [5]. Background integrity findings on third-party carbon auditing remain load-bearing method evidence [9][11]; they are not a substitute for current origin voices inside the expert window.

Wave one is already inside that artefact. The Commission’s own chronology is unambiguous: the first companies subject to CSRD applied the new rules for financial year 2024, for reports published in 2025 [2]. The freeze on additional datapoints for 2025 and 2026 is a filing convenience. It is not a licence to treat an industry-average Scope 3 line as observed fact, and it is not climate stewardship. Assurers who sign that line are signing a model, not a measurement of the land.

02The transitional opportunity: primary data is already the cheaper error

The firms that will come out of this cycle with a transferable asset are not the ones who bought a more granular emission-factor library. They are the ones who can show, plot by plot, who produced the goods, under what regenerative or extractive practice, and which downstream claimant is allowed to carry the tonne. That is the same inversion already visible in EUDR geolocation: a walked boundary is a primary observation; a map class is an inference. Scope 3 is the climate analogue — and the stewardship analogue. Stenzel and Waichman note that missing interoperability — no common measurement standard and no common technical exchange — is one of the three binding obstacles to that analogue, alongside contract limits on re-use and the fear of reverse-engineering process secrets [15]. A factor library solves none of the three.

Stenzel and Waichman are explicit about why the inversion is hard. Firms need data “not under their control.” Obstacles are legal and contractual limits on re-sharing, missing interoperability, and the risk that product-level emission data can be reverse-engineered into process secrets [15]. Those are real constraints. They are not an argument for averages. They are an argument for a narrow, purpose-limited exchange: activity data and practice flags from the origin, licensed without surrendering the primary record — not a full bill of materials harvested as a condition of offtake.

The Greenhouse Gas Protocol’s Land Sector and Removals guidance exists precisely because agricultural land use and CO₂ removals do not fit a stack-emissions template. It is organised as a companion to a land-sector standard, with worked examples and corporate case studies, and it treats land-based emissions and removals as inventory items that have to be accounted rather than as marketing claims [20]. That is the correct posture for a CSRD file. Credits sitting in a voluntary registry are a different instrument, and the integrity literature on those instruments is now hostile. A 2024 assessment of 2,346 carbon-crediting projects found that more than four-fifths of issued credits did not reflect real emission reductions [10]; a 2025 review of Verra-registered projects found systematic weaknesses in the auditing process that is supposed to police those claims [9][11].

Voluntary-market credits are not a workaround for that gap, and they are not regenerative agriculture. They are a second claimant on the same biophysical change. The live integrity load is recent and hostile: overstated issuance in the 2024 multi-project assessment [10], and 2025 reporting on weak third-party audits of credit projects [9][11]. Communities asked to accept a price on a forest they were already conserving remain the capture pattern that older market commentary already flagged [13][12]. A CSRD inventory line that quietly leans on the same projects inherits that record — and launders capture into the disclosure file.

For a cooperative, the transitional opening is unglamorous and valuable. Record the practice and the plot once. Licence the same dataset to several buyers who each need a Scope 3 primary-data point. That is cheaper than filling a different spreadsheet for every brand, and it is the only way the producer retains the option to sell a removal later without discovering that the brand already booked it as an inventory reduction. That is climate stewardship as infrastructure: the ecological work stays visible, and the margin stays nearer the land.

03The Institute analyst take: who owns the tonne — and who captures the claim

Investors have been pricing carbon opacity for years. Using Carbon Disclosure Project data for S&P 500 firms in 2006–2008, Matsumura, Prakash and Vera-Muñoz found that markets penalised emissions and then penalised non-disclosure on top: on average, firm value fell by $212,000 for every additional thousand metric tons of carbon emissions among disclosing firms in their sample [18]. CSRD moves that penalty from a voluntary CDP response into a statutory report with an assurance opinion. The residual risk is no longer “did they disclose”. It is “did they disclose a number that another claimant also holds” — and whether the farmer who changed the practice still holds the primary observation.

In agrifood, that collision is structural. The same farm reduction can appear as the cooperative’s outcome metric, the trader’s Scope 3 category 1, the brand’s insetting claim, and a credit sold into the voluntary market. Caro et al. described the joint-production mechanics; the market now prices the governance failure [19]. Corporate “regenerative” and transition-agriculture programmes that harvest plot data as a condition of offtake, then book the climate story as the buyer’s inventory progress, are the same mechanism under a friendlier label: capture of language and of the record.

"Farmers are having to adopt new digital tools and map farm plots, which is an expensive burden in order to continue to be viable suppliers to the European market." Brenda Mariana Huerta García, Senior Advisor, Climate & Environment, Fairtrade International, on Fairtrade Plot Insights, June 2026 [6]

That is the producer cost of the primary-data path Stenzel and Waichman prescribe [15][6]. Industry-average Scope 3 is an unallocated residual — and often an enclosure of origin ecological work into someone else’s filing — the same capture pattern older market commentary flagged when conservation value is booked upstream [12]. When two listed counterparties both report progress against the same physical flow, limited assurance on each file can still be internally consistent and collectively false. The first restatement cycle will be a reallocation of tonnes between buyers who thought they had exclusive claim, while the producer who did the work may own neither the data nor the premium.

Exporters already know the operational version. A European customer that cannot defend its Scope 3 line will push the data request down the chain, in whatever format its software vendor ships. Producers who wait for that format will build the file twice and lose the system of record. Producers who hold a plot-level activity record can answer once — and keep the licence terms under their own control.

04The proactive resolution: inventory the plot, licence the tonne, keep the stewardship claim

The correction is institutional, not statistical. Keep satellite and factor-based screening for coverage. Move the evidence that enters the assured Scope 3 total to origin: geolocated production, a practice flag that a named person can attest, and a chain-of-custody identifier that prevents the same tonne from being both inventoried and credited. The GHG Protocol land-sector work is the methods layer for land and removals [20]. CSRD and the ESRS are the disclosure layer [16][14]. Neither layer is a substitute for the observation — and neither layer is regenerative agriculture until origin retains data, seed and margin while the ecological work is measured in place rather than converted into offset theatre [10][12][17].

That is the legitimacy test for any stack sold into this gap: origin ownership of the primary record; verification the farmer can inspect; a shorter path from field to buyer rather than another paid gate; ecology measured on the plot, not claimed twice in a registry. Licence the tonne. Do not surrender the file.

Omnibus I bought time for companies that had not yet entered the regime, and it capped new datapoints for those already in. It did not make an emission factor into an observation, and it did not convert corporate inventory theatre into climate stewardship. The files that will survive the first restatement are the ones that can show the plot — and leave the producer holding it.