Explanation
Double materiality, or what a model can evidence
Most double materiality assessments are strong on the financial side and thin on the impact one. A walk through the Deck's Double materiality and CSRD pages on Nutris, what a valuation can evidence, and where the reporter still decides.
Valued before November 2026, when each pathway carried an additionality percentage. Results now value each pathway against its alternative where it has one, and show an absolute and an additional impact, as the net impact bridge explains.
Most double materiality assessments are strong on one axis and thin on the other. Finance and risk teams have years of practice estimating how sustainability matters hit the business. The other direction, how the business hits people and nature, usually comes down to a workshop, a survey and a heat map, and when an auditor asks why one topic scored higher than another the honest answer is that a room agreed it did.
An impact valuation can put numbers behind that axis. It will not settle the assessment, and it repays careful reading. This walks through the Double materiality and CSRD pages of the Deck on Nutris, an early-stage fava-bean protein producer with about 10.1 million dollars of revenue. Its model has 50 pathways and a net societal value of minus 6.1 million dollars, split across natural capital at minus 7.0 million, human capital at plus 224,000 and social capital at plus 705,000.
What the Deck gives you
Open the company in the Deck. The pages beside the chat end with a group headed Report, and two of its pages matter here. Double materiality places the ten ESRS topics on two sides and scores them. CSRD re-presents the same model as the datapoints the standards ask for, in their physical units, with what the model cannot prefill marked as company data.

Neither page runs a new valuation. Both read the published release the Deck shows, and the year bar above them sets the year the impact side reads, the analysis year by default or a year of a scenario, which the page then stamps as projected and not for filing. The Deck prepares the Double materiality page when a release is published, with one AI pass that changes nothing in the model, and an editor can regenerate it with the button beside the heading. Nothing on the page moves a score or the threshold. The Deck uses one fixed threshold, 3 on either side, and the company's own assessment sets its own.
The framing at the foot of the page is worth reading rather than skipping. It calls the page a screen to support the company's own assessment, not a compliance statement, and says the reporter sets its own thresholds, consults its stakeholders and documents the process. Everything below depends on that distinction holding.
Reading the matrix
Two sides, two kinds of evidence
The matrix looks like one output. It is two, and they come from different places.

The impact side, the vertical axis, is computed from the model, with no judgement in the starting position, just the model sorted by size. Each topic's gross monetised value is compared with the largest topic's. The largest scores 5, and each half order of magnitude below it costs a point, down to 1. The release places its environmental value on the topics itself, while one AI pass places each human and social pathway on its topic. For a company valued against an alternative, each topic is scored twice, with the company and against its alternative, each reading against its own largest topic, and the topic takes the higher score; the two readings are never added. The financial side, the horizontal axis, is a judgement. One AI pass scores it from the release's risks, opportunities and financials, names the channel and the horizon, and writes the basis under each topic.
That difference decides how much weight each side can carry. A topic sitting high has a large modelled impact and the pathways to show for it. A topic sitting far right has a judgement, informed by the model but not produced by it, and a regeneration can move it. Presenting the matrix as a single model output overstates the second side and invites a challenge that is hard to answer.
One Nutris topic is material on the financial side alone. Consumers and end-users scores 1 on impact against 3 on financial, with 15,000 dollars of modelled value, most of it improved protein intake and plant-based dietary choice. The financial score rests on customer qualification and repeat whey-replacement orders, a commercial channel the model does not value. The topic is reportable. It did not get there through the model.
The impact score is also relative. It ranks a company's topics against its own largest one, here pollution at about 4.9 million dollars of gross value, so a 3 at Nutris and a 3 at a multinational are not the same size. It ranks topics inside one company, and it does not compare companies.
What a low score can mean
Each topic also carries an evidence note explaining what sits behind its score, and that note is worth reading alongside the number.

Own workforce scores 1 on both sides. The model holds 14,000 dollars of employee wage income against 11 dollars of injury risk, and nothing else on the topic. So the score is 1 because the model has little on its own workforce, not because a processing company has nothing there. The summary at the top of the page says as much when it lists own-workforce pay, safety and training data among what the reporter must still supply, and a refinement of the model on those points would be worth planning in the Model.
Affected communities scores 1 for a different reason. Its note says the environmental public-health burdens stay in the environmental topics rather than being counted again here. A 1 there is a decision not to double count, not an absence of effect.
The same habit applies across the matrix. A score reflects what the model currently covers as much as the size of the effect, and the note is where that shows.
Reading the ESRS datapoints
The CSRD page reorganises the same model by ESRS topic, in the physical units the standards ask for, computed from the release rather than generated. Material topics come first, and the rest follow marked as ones that may be omitted. Each datapoint carries its disclosure code, the prefill, the whole value chain beside it where the datapoint is bounded to own operations, and a source. For a company valued against an alternative, the quantities are those with the company, as an inventory reports them.
The value chain decides the answer
The Overview page shows it by value-chain step. Nutris's supply chain contributes minus 5.37 million dollars, own operations minus 894,000, products and services plus 176,000, and end of life minus 9,300, producing the net minus 6.09 million.
Almost the entire burden sits upstream, in the fava crop Nutris buys.
The physical datapoints say the same thing in their own units, topic by topic. Land use is 52,036 square-metre-years of crop-equivalent land in own operations against 21.6 million across the whole value chain, so more than 99 per cent of it sits outside the factory gate. More than 93 per cent of terrestrial biodiversity pressure does too. Water consumption runs the other way, with 31,429 of 40,046 cubic metres used in own operations.
That has a direct consequence for scoping. An assessment drawn around own operations would see a fraction of a per cent of the land Nutris occupies and little of its biodiversity pressure, and would rank the topics differently from one drawn around the value chain. The standards require the wider boundary, but the practical point is sharper than compliance. The boundary decides the answer before any scoring begins.
Two sign conventions on one screen
The datapoints and the topic values use the word positive to mean opposite things.

Physical datapoints are quantities. Positive means emitted, consumed or occupied, and avoided quantities are never netted against them. Total greenhouse gas emissions read 5,399 tonnes of CO2-equivalent, and the 247 tonnes Nutris helps avoid sit on their own row, marked as evidence and kept out of the inventory. Bigger is worse and nothing is ever negative.
Monetised values are values. Negative means societal value destroyed and positive means societal value created. The Double materiality page shows both inside a single topic. Biodiversity and ecosystems nets to minus 1.3 million dollars, with the purchased fava crop at minus 1.4 million and avoided whey production at plus 88,000.
One number in that list is good news. Anyone scanning for the largest figures rather than reading the signs will get the topic backwards.
A related subtlety sits in the scoring. Topic exposure is gross, so a positive pathway adds to a topic's score exactly as a negative one does. Workers in the value chain scores 3 on impact with 194,000 dollars of net value, nearly all of it beneficial, led by 143,000 dollars of benefit to workers from the purchased fava crop and 21,000 of extra income for the farmers who supply it. High impact materiality does not mean harm. It means the model has something large to say. The same topic carries 18 labour-rights risk cases on the CSRD page, sector estimates of child and forced labour, injuries and overtime that point due diligence at suppliers rather than find cases. A positive score and a due-diligence flag sit on one topic, and neither cancels the other.
The company data rows are the work plan
Each datapoint carries a source, and the label is the signal. Proxy means the model prefilled a life-cycle quantity that approximates the datapoint, on another boundary or in characterised equivalents rather than masses. Evidence means no ESRS metric, but support for the topic's assessment. Company data means the model never holds it.
Twenty datapoints come prefilled for Nutris. Most of the rest are company data, and they are specific. Climate needs the energy mix, the transition plan and the anticipated financial effects. Biodiversity needs the site locations to say whether any sit in or near sensitive areas. Own workforce needs headcount, injuries and the gender pay gap, and the page notes that the model's job figures are supply-chain estimates, not the company's own workforce.
Read together, those rows are a data-collection plan built from the company's own impact profile rather than from a generic checklist. They say which system each missing datapoint has to come from, and they distinguish what the model cannot know from what nobody has collected yet. The summary on the Double materiality page makes the same division at the top level, listing what the reporter must still supply, from policies, targets and actions to verified farming and energy performance, evidence of customer substitution, and own-workforce pay, safety and training data.
The proxy label deserves the same attention. Pollution datapoints come as PM2.5 and NOx equivalents over the life cycle, while E2-4 asks for the masses of pollutants above the E-PRTR thresholds. A proxy is evidence for the assessment and a starting point for the inventory, not the inventory.
For most teams that is one of the more useful things on the screen, and it is the part that looks least like a result.
What a heat map cannot show
The standards ask for material impacts in both directions, positive and negative. Most assessments document only the negative, because a workshop scoring exercise has no unit in which to express a benefit and no way to weigh one against a harm.
A valuation does. Nutris's capital split runs natural capital at minus 7.0 million dollars against human capital at plus 224,000 and social capital at plus 705,000, and the positives are attached to named pathways rather than asserted. The purchased fava crop and the extra farmer income carry 194,000 dollars to workers in the value chain. Improved protein intake and plant-based dietary choice carry 14,000 to consumers and end-users. Those are reportable positive impacts with evidence behind them, in a format that sits beside the harms rather than in a separate goodwill section. The page also says where a positive does not belong. Payroll taxes, at 57,000 dollars, stay outside the ten topics, because ordinary tax contributions are not an ESRS topical impact.
The ranking is also defensible in a way a show of hands is not. The impact side is derived from 50 pathways, so when someone asks why pollution outranks climate the answer is about 4.9 million dollars of gross value against about 930,000, with the contributing pathways listed underneath. That is an evidence trail rather than a recollection of who spoke loudest in the room.
And a good deal of reporting work arrives already done. Greenhouse gas emissions by scope and their intensity, particulate and ozone-forming emissions, phosphorus and nitrogen reaching water, land use, water consumption and fossil and mineral resource use all come through as prefills, and where a datapoint is bounded to own operations the whole value chain sits beside it. That split is a requirement, and it is usually one of the more laborious parts of a first reporting cycle.
None of this settles the assessment. It changes what the workshop starts from, which is a ranked and sourced set of impacts to argue with rather than an empty matrix.
Working with the result
Three things are worth doing before this reaches a workshop.
1. Test the threshold and the close calls in your own assessment. The Deck holds the threshold at 3 and lets nobody move a score, so the sensitivity work happens where it gets logged for the IRO-1 process description. Nutris has five of ten topics material. Consumers and end-users is material on its financial score alone, at exactly 3, so one point less would drop it, and resource use and circular economy sits one point below on the financial side. Whether that is the right cut is a scoping decision, and it is better made deliberately than inherited.
2. Ask the chat why a topic scored what it did. The Deck's chat reads the same pages and links to them, and the Double materiality skill in its + menu sets out the table, the close calls and the reporting readiness in one answer. The reasoning behind a score is more useful than the score, and this is the fastest route to telling a topic with a genuinely small impact apart from one whose pathways are still being built.
3. Publish again after changing the model. The Deck reads the published release, so a calibration in the Model reaches these pages once it is published, and the Deck prepares the Double materiality page again for the new release. Download Excel in the Deck bar takes out the model behind the figures, with its pathways, physical indicators and sources, so a reader can see where the evidence stops rather than discovering it during assurance. The two report pages themselves stay in the Deck.
Where it stops
The page produces evidence for an assessment. It does not produce the assessment.
It does not establish financial materiality, since impact pathways describe effects on people and nature while financial materiality concerns effects on the business, and the financial side is a judgement prepared from the release's risks, opportunities and financials rather than derived from the valuation. It does not decide materiality, which also turns on scale, scope, likelihood and remediability. It does not replace stakeholder engagement, because a model can estimate where impacts are largest but not how affected people experience them or whose perspective is missing. And it does not make anyone compliant, since legal scoping, the formal assessment, policies, targets, controls, governance approval and assurance all sit outside it.
Above all it is only as complete as the assessment behind it. Anything outside the boundary is absent from the output, which is what the evidence note on each topic is there to surface.
Background
The Corporate Sustainability Reporting Directive is the EU framework for corporate sustainability reporting, and double materiality is its organising idea. Companies in scope report both how their activities affect people and the environment and how sustainability matters affect their own development, performance and position. The European Sustainability Reporting Standards define what gets reported, across twelve standards covering general requirements, general disclosures, five environmental topics, four social topics and business conduct. Ten of the twelve are topical, which is why the matrix carries ten topics.
Two changes matter for scoping. Directive (EU) 2026/470 entered into force on 18 March 2026 and narrowed the scope to undertakings exceeding both 1,000 employees and 450 million euros in net turnover, with both thresholds required. First application is for financial years beginning on or after 1 January 2027, with first reports expected in 2028, and only limited assurance is now required.
The standards themselves are being simplified. The Commission adopted the revising delegated act on 3 July 2026, cutting mandatory datapoints by roughly 60 per cent. It sits under a scrutiny period during which Parliament and Council can accept or reject it but not amend it, and it enters into force once published in the Official Journal. The revised standards apply for financial years beginning on or after 1 January 2027, with voluntary early application for 2026. The Deck's pages still follow ESRS Set 1, Delegated Regulation (EU) 2023/2772, and name it on the page.
Confirm which version applies before relying on any of it. Scope, reporting year and standard version all move independently.