Self-reported nature gains cannot lift a project above neutral under this rubric. For a portfolio manager judging an impact claim, that ceiling is the paper's most useful choice.
What goes into RoEn?
Čekan/ová and Spearman argue that impact reports too often count inputs and outputs, such as "100 companies supported", instead of changes in ecosystems. The literature they cite gives the question scale. More than half of global GDP, at least $44 trillion to $58 trillion, depends moderately or highly on nature. For every $1 spent protecting nature, $30 is spent destroying it. Only 8% of organisations report a mature ESG strategy, though 77% say ESG significantly affects their reputation.
Their proposed Return on Environment (RoEn) is meant to sit beside ROI. It records what an investment returns to, or takes from, the ecosystem it touches. One part puts monetary values into a benefit-cost ratio, (V_ES + L_A − D_E)/C. Ecosystem service gains and avoided losses, each measured against a counterfactual, enter the numerator alongside ecological damage. The authors report the ratio as a range.
The other part rates carbon, water, biodiversity, resilience, natural capital and community from −2 to +2. Those six dimensions feed a five-level continuum: destructive, harmful, neutral, beneficial, regenerative. Each score receives a weight (1/6 by default), is multiplied by a verification factor and enters the sum. The continuum adapts the Footprint Evaluation typology of impacts on natural systems. For the term itself, the authors credit a 2010 Southeastern Pennsylvania open-space valuation.
There is no dataset behind the rubric. The paper is a desk review of purposively chosen cases, with one illustration at each continuum level. BP represents destructive; the Triodos Future Generations Fund, harmful. ČEZ's generation mix illustrates neutral, the highest level renewables approach. The 14,500-acre Pocosin peatland that AXA IM agreed to finance is beneficial, while mangroves are regenerative. The authors disclose that they drafted the scoring method with Anthropic's Claude and then revised it themselves. They show the composite score only for a hypothetical peatland and the ratio only for one published floodplain figure.
Verification decides the ceiling
Independently verified gains, ground-truthed with local communities, receive a verification factor of 1.0. Third-party or remote-sensing verification alone gets 0.5; self-reported gains get 0.25. The factor applies only to positive scores. As the authors put it, "weak MRV discounts claimed gains but never softens reported harm" (MRV meaning monitoring, reporting and verification).
Run the maximum score through the formula.
A self-reported project scoring +2 on all six dimensions finishes at 0.5. Neutral runs from −0.5 to +0.5, so even the strongest self-reported claim stays there. Satellite or third-party verification alone takes that maximum to 1.0, within beneficial. A regenerative rating, above +1.5, requires community ground-truthing as well.
The authors illustrate the difference with scores of +2, +2, +1, +1, +1 and 0. Community verification produces 1.17, beneficial; self-reporting produces 0.29, neutral. By our arithmetic, third-party verification alone would produce 0.58, just above the 0.5 cut-off.
Further rules constrain the rating. A destructive score on any one dimension caps the overall result at neutral, preventing carbon gains from offsetting a breached boundary elsewhere. An unassessable dimension is marked insufficient evidence rather than scored 0: "absence of evidence of harm is not evidence of no harm." Its weight passes to the dimensions that can be assessed. If the missing dimension is material to the site, the rating cannot exceed neutral.
That changes how familiar labels read. The continuum places renewables at best neutral and a fund that beats its index on GHG intensity in harmful. More than 60% of the $6.2 trillion labelled sustainable bond market in 2025 went to renewable energy and low-carbon transport. Under this rubric, the renewable-energy share of that 60% buys neutrality at most.
The ratio's missing-damage problem
For the composite, missing evidence never becomes a score of zero. The monetized ratio treats terms without credible values differently: they "should be omitted (that is, set to zero)." Omitting V_ES is conservative. Omitting D_E, ecological damage, makes the result look better.
The worked example uses The Nature Conservancy's $7 saved in future disaster costs per $1 of floodplain protection. The authors present a 7:1 RoEn ratio, setting V_ES and D_E to zero because the source reports only avoided losses.
Unestimated damage therefore enters the numerator as no damage.
The three-to-one mangrove return for the wider Caribbean is likewise a source-published benefit-cost ratio. The authors acknowledge that their cases rely on differing valuation methods.
Scale poses another problem. The weights sum to 1, and the score has no size term: a 14,500-acre peatland and a one-acre restoration with identical categorical scores receive identical ratings. The natural capital dimension, indicated by ecosystem asset appreciation, can also reprice gains already counted under carbon, water and biodiversity. Resilience, indicated by avoided disaster loss, echoes L_A in the ratio. Because the two figures appear side by side and are never summed, neither double counts. Readers should still avoid treating them as independent evidence.
The authors call the composite "a convenient summary rather than an exact quantity" and say it "should always be reported together with the individual dimension scores (each si)." Showing those scores lets readers see where overlap sits. It cannot supply the size term the dimensions themselves lack. The authors describe the 1.0/0.5/0.25 factors as indicative and leave calibration of weights, verification factors and thresholds to future research. We did not find a sensitivity analysis for those factors or the ±0.5 and ±1.5 cut-offs. A rating only 0.08 above a cut-off warrants one. Publishing thresholds before they are calibrated invites the borderline call produced by the authors' own peatland example.
Why we could not test the score
Each project needs community ground-truthing and a counterfactual trajectory for the score; its ROI side also needs project cash flows. Our market and company data cannot show what a restored peatland did relative to an untreated one nearby. The authors call the rubric preliminary and identify testing on real portfolios as future work.
Claims filter, rather than return forecast
The paper offers RoEn as a complement to ROI and makes no claim that its score predicts returns. The ratio measures priced value per dollar; the score records ecological direction. Its financial argument rests on cited work. In Agarwala and coauthors' worst case across 23 countries, annual interest bills rise by $162 billion, equal to 81% of the $200 billion yearly nature target. Those authors also estimate that $83 trillion of assets are mispriced. Agarwala's result concerns sovereign credit. It says nothing about whether a project scoring 1.17 out-earns one scoring 0.29.
A few dozen verified projects showing that higher composite scores track lower losses or cheaper financing would change my view. For now, the rubric earns its place as a filter on claims, chiefly because its verification ceilings have teeth. The Galápagos debt swap, financed through the $656 million Galápagos bond and already facing complaints about missing information for local communities, is the kind of deal to put through it first.