The eco–ultraprocessing paradox as evidence for a Tech–Nutri–Eco coherence framework in plant-based food sustainability governance

1 Introduction

Global pressure to reduce environmental footprints and to transition towards sustainable food systems has driven a rapid expansion of the market for plant-based foods (PBFs). These products are promoted as ethical and environmentally friendly solutions, with the potential to reduce greenhouse-gas emissions and the use of natural resources (Poore and Nemecek, 2018; Willett et al., 2019). Recent scholarship has examined how manufacturers of plant-based products navigate market pressures and consumer imaginaries in shaping their product strategies, with sustainability increasingly functioning as a central communicative device rather than a verified material property (Hansen, 2025). However, recent literature emphasizes that PBFs are not homogeneous: a substantial share falls within the category of ultra-processed foods (UPFs), produced through complex industrial technologies — such as protein isolation, extrusion, emulsification, and fortification — which raise distinct nutritional and environmental concerns (Alcorta et al., 2021; Monteiro et al., 2019).

This dissonance between public perception and technological reality corresponds to a phenomenon we term the “eco–ultraprocessing paradox”: products marketed as sustainable and healthy, yet obtained through energy-intensive processes and varying degrees of artificialization (Capozzi et al., 2021; Fardet and Rock, 2020; Estévez et al., 2024). In the absence of an integrated assessment, environmental performance (Eco-Score) and nutritional performance (Nutri-Score) may be simultaneously favourable, even when a product falls within NOVA group 4 (NOVA food processing classification system) — indicating a potential risk of greenwashing, particularly in commercial communication (Nugraha et al., 2024; Persakis et al., 2025; Feghali et al., 2025; Santos et al., 2024).

Current regulations further reinforce this fragmentation. The European framework addresses technological (European Parliament, 2008), nutritional (European Parliament and of the Council, 2011), and environmental dimensions (Directive (EU), 2024/825). Directive (EU), 2024/825 on environmental claims requires the verifiability of “green” claims but does not introduce explicit criteria for inter-dimensional coherence across technological, nutritional, and environmental performance. In the absence of such requirements, producers may selectively communicate a single favourable dimension — typically nutritional or environmental — thereby creating an illusion of sustainability (Directive (EU), 2024/825; Persakis et al., 2025).

At the same time, conceptual studies indicate that the transition towards sustainable food systems depends not only on reducing environmental impacts, but also on achieving coherent alignment among technological, nutritional, and environmental dimensions within a framework of systemic coherence (Corrado et al., 2017; Notarnicola et al., 2017). This approach integrates life-cycle thinking with public health and food safety criteria, to avoid unsustainable substitutions and unilateral optimizations.

The central proposition is that the sustainability of a processed food product cannot be inferred from strong performance on any single dimension, and that coherence among its technological, nutritional, and environmental dimensions is a distinct property, which current instruments neither measure nor require, and which must be read alongside the levels attained rather than in place of them.

First, the technological, nutritional, and environmental dimensions are assessed through instruments that operate independently and share no reconciliation criterion, so that favourable single-dimension scores can coexist with unresolved inter-dimensional conflict. Second, this fragmentation produces recurrent, classifiable incoherence patterns rather than isolated anomalies — most visibly the eco–ultraprocessing paradox, in which ultra-processed products attain favourable nutritional or environmental ratings. Third, coherence can be made explicit without introducing a new normative score: a diagnostic indicator, the Integrated COherence iNdex (ICON), renders inter-dimensional alignment measurable and exposes configurations that single-dimension assessment does not.

2 Methodology

The literature was examined through a critical interpretive synthesis (Dixon-Woods et al., 2006), an approach designed to critique the assumptions on which a body of literature rests rather than to summarise its findings. Coding is refined iteratively as the analysis proceeds, and new constructs are generated from the corpus rather than extracted from individual studies. The approach suits emerging fields in which concepts, tools, and theoretical frameworks partially overlap or are applied inconsistently. The aim of this approach is not the exhaustive accumulation of literature, but rather the identification of conceptual tensions and systemic incoherences that constrain the integrated assessment of food sustainability and that underpin the development of a triadic analytical framework. The design combines the literature synthesis with an illustrative quantitative component, in which Spearman rank correlation examines how coherence relates to each individual dimension, the Mann–Whitney U test compares coherence between ultra-processed and other products, and k-means clustering tests whether the products form distinct dimensional profiles.

2.1 Search strategy and selection criteria

The search targeted scientific literature published between 2015 and 2025 in major international databases (Scopus, Web of Science, ScienceDirect, and SpringerLink). This temporal window was selected to capture the period during which the Nutri-Score, Eco-Score, and the NOVA classification began to be used concurrently in research and public policy. The search keywords combined terms related to food processing, sustainability, and labelling, including: “plant-based foods,” “ultra-processed,” “NOVA classification,” “Nutri-Score,” “Eco-Score,” “greenwashing,” and “sustainability claims.”

The initial results were examined iteratively through title and abstract screening to exclude materials that did not meet the inclusion criteria (editorials, technical notes, commercial reports). Selection was guided by conceptual relevance and included only articles that:The objective of this process was not exhaustiveness, but the achievement of thematic saturation, defined as the point at which additional articles no longer contributed new concepts, tensions, or perspectives relevant to the relationships among the three axes. Once this conceptual threshold was reached, the set of articles documented in Supplementary Table S1 constituted the basis for the critical interpretive synthesis. The remaining references cited in the manuscript document the assessment instruments, the regulatory framework, and the methodological guidance, and were not part of the coded corpus.

• explicitly addressed at least one of the Tech, Nutri, or Eco dimensions in the context of food sustainability assessment;
• discussed assessment tools, policy instruments, or governance implications; and
• originated from peer-reviewed scientific journals.

2.2 Analysis and coding procedure

The content analysis followed an iterative, narrative, and thematic approach, informed by methodological guidelines (Xiao and Watson, 2019; Grant and Booth, 2009). Each selected article was examined in full and coded within an analytical matrix, subsequently synthesised as Supplementary Table S2, which included variables such as: study objective, instrument type (technological, nutritional, environmental), degree of integration across dimensions, and the main methodological limitations reported.

To ensure the traceability of analytical decisions and consistency of interpretation, coding was conducted in two successive stages designed to reduce the risk of interpretative inconsistency:

• an exploratory phase to identify recurrent concepts;
• a thematic refinement phase to verify internal consistency and to delineate categories coherently.

Consistent with this synthesis approach (Snyder, 2019; Gough et al., 2017), the analysis focused on identifying epistemic and methodological tensions among the technological, nutritional, and environmental dimensions of food sustainability, without quantifying their frequency or weighting their relative importance, with the aim of constructing an integrative framework capable of assessing internal coherence across these dimensions.

To ensure the traceability of inductive inference, each identified incoherence pattern (T1–T10) is documented in Supplementary Table S2 and accompanied by explicit references to the articles from the initial corpus (RowID from Supplementary Table S1) that underpinned its identification. The articles listed in the Supporting_articles column (identified by RowID from Supplementary Table S1) were selected through qualitative synthesis of descriptive fields and serve an exclusively illustrative and interpretative role, without aiming for statistical representativeness or quantitative assessment of pattern frequency.

2.3 Illustrative component

To exemplify the applicability of the Tech–Nutri–Eco framework, an illustrative proof-of-concept component was added, based on a set of 49 plant-based food products drawn from the Open Food Facts (OFF) database (Open Food Facts, n.d.), obtained after excluding nine items from an initial set of 54 due to missing indicators (Supplementary Tables S3–S5), to which four further products were added by API retrieval from the same database.

This component serves a strictly demonstrative, non-inferential purpose and aims to highlight how differences across the technological (NOVA), nutritional (Nutri-Score), and environmental (Eco-Score) dimensions can be analyzed in an integrated manner within a single coherence framework.

The analysis was conducted exclusively on products for which all three dimensions were available, without applying imputation procedures or statistical aggregation, to preserve transparency, traceability, and the strictly demonstrative nature of the illustrative exercise. The final product set and the normalization rules applied are documented in the Supplementary Tables S4–S6. Data processing, normalization, and ICON calculation were performed in Python 3.9 using the pandas (v. 2.3.3) and NumPy (v. 2.0.2) libraries. Spearman rank correlation and the Mann–Whitney U test were selected because they require no distributional assumption, which the ordinal origin of the three input scores (NOVA, Nutri-Score, and Eco-Score) makes necessary; they were computed with SciPy (v. 1.13.1). K-means clustering, with the number of clusters selected by silhouette score, was applied to the three normalised dimensions to establish whether the products form distinct dimensional profiles independently of the ICON formula; it was computed with scikit-learn (v. 1.6.1). Visualizations were generated with Matplotlib (v. 3.9.4) and Plotly (v. 6.6.0). The software tools, library versions, and the resulting statistical values are documented in the Supplementary material (Supplementary Table S7).

2.4 Justification of the methodological design

Existing frameworks address the technological, nutritional, and environmental dimensions separately without formally integrating them. Conceptual and normative sources are therefore treated as primary material rather than background, since they define the assessment logics whose incoherences are examined here. The synthesis maps these incoherences across the literature and grounds the operationalisation of the integrative Tech–Nutri–Eco framework, with the illustrative application demonstrating its diagnostic use.

3 Dimensions of sustainability assessment (Tech–Nutri–Eco)

This section examines separately the three main dimensions used in the assessment of food product sustainability — technological, nutritional, and environmental — with the aim of highlighting their internal logic, methodological limitations, and potential sources of divergence. The distinct treatment of these dimensions does not imply a normative hierarchy, but rather constitutes a necessary analytical step to understand how fragmented assessments may lead to incongruent perceptions of sustainability. This sequential analysis subsequently underpins the development of the integrative Tech–Nutri–Eco framework and the proposed coherence indicator.

3.1 Technological dimension: food processing and the NOVA classification

The technological dimension of food sustainability refers to the extent to which the structure and integrity of a food product are modified through technological and industrial processes. In the current literature, the most widely used classification framework is NOVA, which groups foods into four levels according to the extent and purpose of industrial processing, ranging from unprocessed and minimally processed foods (NOVA 1) to formulations of ingredients, mostly of exclusive industrial use, designed to be durable, convenient, and highly palatable (NOVA 4) (Monteiro et al., 2019). Although NOVA has become a central reference in public nutrition, its use as a proxy for the technological dimension of sustainability is limited. First, the scale was not designed to measure environmental impacts, but rather to characterise processing in relation to diet quality (Monteiro et al., 2019; Fardet and Rock, 2020; van Tulleken, 2025). Second, its interpretation is sometimes ambivalent: certain industrial processes, such as pasteurisation, controlled fermentation, or protein extrusion, may contribute to food safety and energy efficiency, depending on the technological and energy context, without necessarily implying a negative environmental impact and, in some cases, may even improve energy performance or food safety (Tompa et al., 2025; Michel et al., 2024). This ambivalence appears in the relationship between the degree of processing (NOVA) and perceived sustainability: the quadrant “NOVA 4 + high perceived sustainability” is the central tension zone of the eco–ultraprocessing paradox, characteristic of many contemporary plant-based products (Figure 1).

For plant-based alternatives, this limitation becomes critical: many products perceived as “sustainable” fall within NOVA group 4 due to extrusion, protein isolation, and industrial reformulation processes (Spiteri et al., 2024; Tachie et al., 2023; Fanzo et al., 2023). Within the Tech–Nutri–Eco framework, the role of the NOVA classification is not to assess technological sustainability per se, but rather to make visible the tensions between perceptions of naturalness and industrial reality — which constitute the starting point for the incoherence patterns.

3.2 Nutritional dimension: profiling and nutri-score

The nutritional dimension of food sustainability is currently the most regulated and visible, being dominated in Europe by the Nutri-Score system (Julia et al., 2018). Nutri-Score assigns food products a grade ranging from A (favourable) to E (unfavourable), calculated on the basis of energy density, saturated fat, sugars, and sodium content, as well as contributions from fibre, protein, and fruit components (Van Der Bend et al., 2022; Merz et al., 2024; Amberntsson et al., 2025; Julia et al., 2018). Owing to its simplicity, Nutri-Score has become a key instrument in consumer information policies and in product reformulation strategies. However, its usefulness for integrated sustainability analysis is limited. First, the Nutri-Score algorithm is compositional rather than structural: it evaluates nutrients, not processes (Van Der Bend et al., 2022; Srour et al., 2023). As a result, a product obtained through industrial reformulation — for example, a soy yoghurt — may receive an A score, despite falling within NOVA group 4. Second, the scheme completely ignores the environmental dimension, creating an asymmetry between nutrition and sustainability (Green et al., 2021; Norwood et al., 2025). These inconsistencies help explain the phenomenon recently referred to as the nutrition reformulation paradox: the optimisation of nutritional values on labels, alongside a loss of food integrity and increased reliance on industrial processes (Srour et al., 2023; Peters and Verhagen, 2024).

3.3 Environmental dimension: impact assessment and eco-score

The environmental dimension of food sustainability is currently assessed through tools derived from the Life Cycle Assessment (LCA) methodology, standardised under ISO 14040 and ISO 14044 (International Organization for Standardization, 2006b; International Organization for Standardization, 2006a). At the European level, this approach is operationalised through the Product Environmental Footprint (PEF) framework (Commission Recommendation, 2013), which integrates multiple categories of environmental impact, ranging from greenhouse gas emissions to eutrophication, water use, and pressures on biodiversity (European Commission, 2021). For consumers, LCA-based assessment results are frequently synthesised in the form of the Eco-Score, an indicator inspired by Nutri-Score, which relies on environmental impact data derived from standardised LCA databases (e.g., Agribalyse), adjusted through a bonus–malus system that accounts for packaging, origin, and selected certifications (Open Food Facts, n.d.).

Despite its operational usefulness, Eco-Score presents several methodological limitations. First, in applications derived from PEF/LCA, the lack of primary product-level data frequently leads to the use of secondary data and “representative product” profiles (benchmarks) based on market or category averages, which reduces the specificity of the score at the producer level (Bach et al., 2018). Second, the weighting and correction system introduces a high degree of arbitrariness, limiting comparability across food groups (Poore and Nemecek, 2018; European Commission. Joint Research Centre, 2018). Finally, the absence of correlation with the nutritional dimension allows an environmentally optimised product (e.g., a coconut cream) to receive an Eco-Score “A” despite having modest nutritional value — the inverse of the paradox observed for Nutri-Score. When analysed separately, each of these instruments responds to a legitimate logic; when considered together, they may generate divergent assessments of the same product, highlighting the concrete conditions under which fragmented evaluations lead to incoherences across technological, nutritional, and environmental dimensions.

4 Recurrent patterns of dimensional incoherence

The inductive analysis of the literature indicates that incoherences among the technological, nutritional, and environmental dimensions of food sustainability assessment do not emerge as accidental outcomes, but rather as recurrent patterns generated by the fragmented application of existing instruments. Based on the analysed corpus, 10 distinct incoherence patterns were identified (T1–T10; Supplementary Table S2), which can be conceptually aggregated into four main families. This structuring enables a coherent critical synthesis while preserving the granularity of the inductive analysis.

4.1 Metric reductionism and the implicit neutralisation of technological processing (patterns T1, T2, T8)

A first family of incoherences arises from metric reductionism, whereby the technological dimension of food products is either obscured or implicitly treated as neutral within nutritional and environmental assessments. The literature indicates that favourable nutritional scores can be achieved through compositional reformulation without reflecting the degree or nature of technological processing (T1) (Monteiro et al., 2019; Scrinis, 2020; Deschasaux et al., 2018). Nutrient-centric algorithms assess declared contents rather than food structure or the intensity of industrial transformations, allowing positive nutritional evaluations to coexist with products classified as ultra-processed. Baked beans in tomato sauce, for example, may carry a Nutri-Score of A while falling within NOVA group 4. Similarly, environmental assessments based on LCA/PEF frameworks capture quantifiable effects such as energy use or process-related emissions, but do not discriminate processing as a distinct qualitative dimension of the food product (T2) (Anastasiou et al., 2022; Zampori and Pant, 2019). As a result, processing intensity is treated as a methodological “black box,” limiting the capacity of assessments to reflect the technological implications for the multidimensional coherence of sustainability. These two mechanisms are further reinforced by the implicit assumption of technological neutrality in product comparisons (T8) (Sadler et al., 2021; Scrinis, 2020), where processing is rarely introduced explicitly as an evaluative variable. Taken together, these patterns lead to situations in which processing is neither penalised nor contextualised, but simply absent from inter-dimensional coherence criteria. This systematic omission of technological processing represents not merely an individual limitation of existing tools, but generates a structural incoherence when their results are interpreted jointly as indicators of product sustainability.

4.2 Fragmentation of communication and the validation of sustainability through isolated indicators (patterns T3, T4, T9)

A second family of incoherences concerns the fragmentation of sustainability communication and the selective use of indicators for communicational validation purposes. The literature shows that the parallel use of nutritional and environmental labels may generate contradictory messages for the same product, in the absence of a framework that explains the relationship between them (T3) (Bolhuis et al., 2025; De Bauw et al., 2021). A coconut cream may carry the highest environmental grade alongside one of the lowest nutritional grades, with no indication of how the two are to be reconciled. This dissociation does not necessarily reflect flaws in the instruments themselves, but rather a lack of institutional and conceptual coordination among evaluation frameworks. In this context, sustainability becomes validated through communicational mechanisms — labels, certifications, traceability — rather than through integrated material performance (T4) (Boiral et al., 2026; Grinberga-Zalite et al., 2025; Nugraha et al., 2024). The acceptance of procedural credibility as a substitute for multidimensional assessment encourages the partial communication of sustainability. These processes are amplified by the strategic use of indicators selected to support commercial claims rather than to enable analytical integration of performance dimensions (T9) (Van Amstel et al., 2008; Brécard, 2014; Sciortino et al., 2025). Evaluation thus becomes selective, and indicators are used as discursive instruments rather than as components of a holistic analysis. This family of patterns highlights the structural vulnerabilities of sustainability claims governance, particularly in the absence of an explicit coherence criterion across indicators.

4.3 Scale and category errors in sustainability assessment (patterns T5, T6)

A third set of incoherences arises from confusions of analytical scale, whereby arguments that are valid at the macro level are methodologically and uncritically extrapolated to the micro level (T5) (Tilman and Clark, 2014; Daas et al., 2024). The literature shows that systemic assessments — such as the benefits of plant-based diets or food system transitions — are frequently applied directly to individual products, without accounting for formulation specificity, processing characteristics, or consumption context. Complementarily, the “plant-based” category is often treated as an implicit proxy for sustainability (T6) (Van Der Weele et al., 2019; Smetana et al., 2023), overlooking the substantial heterogeneity of products within this class. This category error leads to overly favourable classifications for highly processed products based solely on their inclusion under a broad label, without explicit assessment of technological, nutritional, and environmental dimensions. Chickpeas in water and baked beans in tomato sauce may carry identical nutritional and environmental grades while sitting at opposite ends of the processing scale. Together, these patterns underscore that the lack of clear delineation between levels of analysis and between descriptive categories and evaluative criteria generates systematic incoherences in sustainability assessment.

4.4 Pseudo-integration and the absence of an operational coherence criterion (patterns T7, T10)

A final family of incoherences concerns the confusion between correlation and interpretative integration, as well as the absence of an explicit criterion for reconciling multiple assessments. The literature indicates a tendency to interpret formal correlations between scores (for example, between Nutri-Score and Eco-Score) as evidence of integrated sustainability, even though such relationships may conceal material conflicts at the level of individual products (T7) (Smetana et al., 2023; Fardet, 2024). A soy yoghurt may carry the top grade on both scales while being ultra-processed, so that agreement between the two indicators conceals the technological conflict rather than resolving it. This pseudo-integration is facilitated by the absence of an operational coherence criterion capable of reconciling technological, nutritional, and environmental assessments (T10) (Van Der Bend et al., 2022; De Bauw et al., 2021). Although existing instruments are robust within their respective paradigms, the lack of a reconciliation mechanism among them causes results to remain unaligned, thereby perpetuating the fragmentation of sustainability assessment.

The identification of this methodological gap indicates that current sustainability evaluations remain fragmented even when robust instruments are employed, in the absence of an explicit criterion for reconciling technological, nutritional, and environmental dimensions.

5 The Tech–Nutri–Eco triadic model

Building on the recurrent incoherence patterns identified in the analysed literature, we propose the Tech–Nutri–Eco model, which approaches food sustainability not as a sum of isolated performances, but as an emergent property of coherence across three fundamental dimensions: technological, nutritional, and environmental performance. Within this framework, the sustainability of a food product is conditioned not only by the level achieved in each individual dimension, but also by the internal compatibility among these dimensions. The technological dimension reflects the degree of processing and the technological efficiency associated with product manufacture, with direct implications for food structural integrity, ingredient use, and the intensity of industrial transformations. The nutritional dimension captures the quality of the compositional and functional profile, including nutrient density and suitability for public health. The environmental dimension addresses environmental impacts across the product life cycle, including resource use and pressures on ecosystems. Although each of these dimensions is currently assessed through established instruments, the relationships and tensions among them are rarely analysed explicitly within a common framework (Figure 2).

The Tech–Nutri–Eco triadic model is conceived as a heuristic framework rather than as a scoring system. Its primary role is to structure and render visible the tensions among the three axes, enabling the identification of situations in which high performance in one dimension is achieved at the expense of another. In this sense, the framework facilitates the recognition of “frontier” products — products that excel technologically, nutritionally, or environmentally in isolation, yet exhibit structural incoherences when assessed integratively. In the absence of such a representation, existing instruments may obscure relevant trade-offs and favour selective interpretations of sustainability. Although the triadic model is essential for the conceptual clarification of inter-dimensional tensions, it remains insufficient for comparative product assessment and for supporting decision-making, governance, or the verification of sustainability claims. This limitation highlights the need for a scalar instrument capable of operationally translating the conceptual coherence defined by the triadic model into a comparable, transparent, and reproducible measure. ICON is developed on this basis, as an operational extension of the Tech–Nutri–Eco framework.

6 The ICON indicator: definition and conceptual rationale

We propose the ICON indicator to quantify the degree of internal coherence among the three fundamental dimensions of food sustainability — technological (Tech), nutritional (Nutri), and environmental (Eco) — which are already assessed through established instruments. ICON does not measure absolute product performance and does not constitute a sustainability score; rather, it provides a diagnostic of the coherence among these performances when considered simultaneously. The rationale for introducing ICON derives from recurrent patterns identified in the analysed literature, according to which plant-based food products may achieve favourable scores on one or two dimensions (for example, nutritional and environmental) while exhibiting weak or contradictory performance on the technological dimension. In the absence of an integrative assessment, such imbalances often remain invisible, and sustainability is interpreted through the lens of isolated performances.

ICON is constructed as a conceptual measure of dispersion among the normalised values of the Tech, Nutri, and Eco dimensions. A product exhibiting similar performance levels across all three axes will display a high degree of internal coherence, whereas a product characterised by selective optimisation or structural trade-offs will register a low level of coherence. Through this approach, ICON enables a distinction between products with a coherent internal profile and products whose apparent sustainability results from compensation effects or the over-communication of a single dimension. ICON is designed as a complementary instrument to existing frameworks (NOVA, Nutri-Score, Eco-Score, PEF), rather than as a substitute for them. Its role is to connect these assessments within an integrative framework, providing an orientative methodological basis for analysing the coherence of sustainability claims and for identifying situations of selective optimisation, with relevance for governance and public policy processes.

7 Proof-of-concept: application of ICON to an illustrative set of plant-based products

7.1 Illustrative product set and data used

To demonstrate the operational functionality of the Tech–Nutri–Eco framework and of the ICON indicator as a diagnostic tool for internal coherence, the ICON was computed for an illustrative set of 49 plant-based food products available on the European market, drawn from six functional categories: simple legumes, meat alternatives, dairy alternatives, plant-based drinks, culinary ingredients, and other plant-based items. Of these, 45 products were manually screened from commercial databases and supplemented by 4 additional products retrieved from the Open Food Facts (n.d.) database to enhance category coverage. The environmental values are Eco-Score grades as published by Open Food Facts. Data were collected between October 2025 and March 2026: the initial 45-product set was compiled in October–November 2025, and the four additional products were retrieved from the Open Food Facts API on 25 March 2026. Product selection aimed to cover a range of technological and functional categories relevant for exploring configurations of coherence and incoherence across technological, nutritional, and environmental dimensions, rather than for quantitative comparison across categories. This exercise is intended as a demonstrative proof-of-concept rather than a statistically representative sample; accordingly, all findings are reported descriptively and should be interpreted as illustrative of the framework’s analytical capacity. For each product, three variables corresponding to the analysed dimensions were extracted and transformed onto the [ 0 , 1 ] interval.

TechNorm, representing the technological dimension, mapped onto a scale compatible with the NOVA classification (Monteiro et al., 2019) and normalised such that higher values indicate a lower degree of processing, to avoid sign inversions in the interpretation of ICON. The NOVA group of each product is the one recorded in Open Food Facts, which assigns it from the declared ingredient list, so that the presence of at least one ultra-processing marker places a product in group 4 (Supplementary Table S3). Within NOVA 4 no further differentiation is applied: all ultra-processed products receive the same value, irrespective of processing intensity or additive load.

NutriNorm, derived from the Nutri-Score (Chantal and Hercberg, 2017; European Commission. Joint Research Centre, 2022) or calculated on the basis of nutritional information declared on the product label (European Union, 2011);

EcoNorm, derived from the Eco-Score, used as an illustrative proxy of environmental performance (Open Food Facts, n.d.).

The normalised dataset allows the three dimensions to be examined jointly across the illustrative set, supporting the interpretation of coherence and incoherence profiles (Figure 3; Supplementary Tables S4–S6).

7.2 Definition and application of ICON

ICON quantifies the degree of relative internal alignment among the technological, nutritional, and environmental dimensions of a food product, evaluated on the same normalised scale.

Let T = TechNorm, N = NutriNorm and E = EcoNorm, with values normalised on the [ 0 , 1 ] interval. The pairwise absolute distances between the three dimensions are calculated as:For three normalised values T , N , E, the sum of pairwise absolute distances is given by:which allows a natural scaling of the measure to the [ 0 , 1 ] interval. On this basis, ICON is defined as:ICON values close to 1 indicate a high degree of structural alignment across dimensions (similar values of T , N , E), whereas values close to 0 reflect pronounced structural divergences associated with selective optimisation across dimensions (Figure 4). Alignment is not the same as level: a product may be aligned at a low level on all three dimensions. Nor does a value on its own identify the source of divergence, since the same value can arise from different profiles.

d TN = ∣ T − N ∣

d TE = ∣ T − E ∣

d NE = ∣ N − E ∣

d TN + d TE + d NE = 2 ( max ( T , N , E ) − min ( T , N , E ) ) ,

ICON = 1 − d TN + d TE + d NE 2

7.3 Demonstrative interpretation

The application of ICON to the analysed illustrative set of 49 products (mean ICON = 0.323 ± 0.246; median = 0.250) enables the descriptive identification of three general profile typologies:The eco–ultraprocessing paradox, defined operationally as TechNorm = 0.00 (NOVA 4, ultra-processed) combined with either EcoNorm ≥ 0.50 or NutriNorm ≥ 0.75, was identified in 28 of the 49 products in the illustrative set (57.1%). The paradox was most prevalent among meat alternatives (75.0% of category products), plant-based drinks (69.2%), and dairy alternatives (66.7%), all of which are characteristically ultra-processed yet frequently achieve favourable nutritional or ecological ratings. This finding illustratively quantifies a structural misalignment: products that score well on individual sustainability or nutritional metrics may nonetheless exhibit low dimensional coherence when assessed holistically. Six products in the illustrative set achieved ICON = 0.000 (the theoretical minimum), all of which were NOVA 4 products with EcoNorm = 1.00, among them the soy yoghurt and the coconut cream referred to above — representing the most extreme manifestation of the paradox wherein maximal ecological performance coexists with maximal processing intensity.

• products with balanced values across the three dimensions, characterised by high internal coherence;
• products that are internally coherent but exhibit uniformly low performance across all axes, illustrating that structural coherence does not automatically imply high absolute performance;
• incoherent products, in which favourable performance in one dimension coexists with weak performance in another, suggesting possible selective optimisation or inter-dimensional discrepancies.

Spearman rank-order correlations between individual dimension scores and ICON values provided further insight into the drivers of coherence across the illustrative set. TechNorm exhibited the strongest positive correlation with ICON (ρ = 0.729, p < 0.001). Because ICON derives from the dispersion among the three normalised dimensions, and TechNorm takes its extreme value (0) in the ultra-processed majority of the set while NutriNorm and EcoNorm remain concentrated at higher values (median 0.75 each), this correlation is partly structural rather than an independent empirical finding. Technological processing is therefore the dimension along which incoherence is most visible in this set, not a demonstrated causal determinant of it. By contrast, EcoNorm was negatively correlated with ICON (ρ = −0.571, p < 0.001), which follows from the same construction: high ecological scores, when occurring in ultra-processed products with TechNorm = 0, amplify the dimensional gap and thus reduce ICON. NutriNorm showed no statistically significant association with ICON (ρ = −0.195, p = 0.180). A Mann–Whitney U test showed that NOVA 4 products (n = 31) had significantly lower ICON values (mean = 0.192 ± 0.173) than non-NOVA 4 products (n = 18; mean = 0.548 ± 0.182; U = 43.0, p < 0.001; Supplementary Figure S4). These results, whilst illustrative and not generalisable beyond the set analysed, show that ICON separates the two groups, a result the construction of the index makes expected, since TechNorm takes its extreme value in every NOVA 4 product.

Tension profile analysis further revealed that Tech–Eco tension (mean = 0.648) was the dominant source of dimensional misalignment across the illustrative set, followed by Tech–Nutri tension (mean = 0.518), whilst Nutri–Eco tension was markedly lower (mean = 0.189). This asymmetric tension structure — in which nutritional and ecological performance are substantially more aligned with each other than either is with technological processing level — provides empirical grounding for the theoretical claim that the eco–ultraprocessing paradox reflects a specific structural fault line in the Tech–Nutri–Eco space (Supplementary Figure S3). The mean dimensional profiles per product category (Supplementary Figure S1) further illustrate this asymmetric structure: meat alternatives, plant-based drinks, and dairy alternatives exhibit a near-collapsed technological axis alongside elevated nutritional and ecological profiles, whilst simple legumes retain the highest technological values among the multi-product categories, since two of them are unprocessed (NOVA 1) and combine high values on all three dimensions (Supplementary Table S6). Sensitivity analysis showed that ICON is algebraically equivalent to the normalised range, whilst showing moderate divergence from coefficient-of-variation-based (ρ = 0.773) and Euclidean-distance-based (ρ = 0.683) measures, which also respond to the level of the values and not only to their spread (Supplementary Figure S2).

The eco–ultraprocessing paradox is not uniformly distributed across plant-based product categories but follows the share of ultra-processed products within each category. All six dairy alternatives are NOVA 4, of which four meet the paradox criteria. Meat alternatives and plant-based drinks show a similar concentration, with 3 of 4 and 9 of 13 products meeting them, reflecting the co-occurrence of high processing intensity and favourable label-mediated performance. Simple legumes are less affected: two of the six are NOVA 4, and both meet the criteria, and the single culinary ingredient is not NOVA 4. Across the set, 28 of the 31 NOVA 4 products meet the paradox criteria, 10 of them in the residual “Other” category, which comprises 19 products. The paradox therefore appears where processing is highest, rather than in individual formulations.

8 Discussion

The 10 patterns identified in the literature fall into four families, and the framework addresses them unevenly. It supplies the operational coherence criterion whose absence is described under T10, and it operates at product level, where the correlations misread as integration under T7 conceal conflicts. By including processing among the assessed dimensions it counters the technological neutralisation of T1, T2, and T8, and by resolving products within categories it exposes the category errors of T5 and T6. The fragmentation of communication described under T3, T4, and T9 concerns how assessments are reported rather than how they are reconciled, and is reached only through the governance application of the index.

8.1 Illustrative findings in relation to previous evidence

Within the plant-based class, the environmental evidence points in the opposite direction to the scores. Life-cycle assessments report that the advantage of plant-based substitutes narrows as processing intensifies, to the point where highly processed products may approach the impact of poultry meat (Smetana et al., 2023), and that sustainability gains become more uncertain the further a product is from its minimally processed form, legumes retaining the clearest advantage (Van Der Weele et al., 2019). In the illustrative set, the six products with ICON = 0.000 are all NOVA 4 and all carry the highest environmental grade, and Nutri–Eco tension is the lowest of the three pairwise tensions. The two consumer-facing scores agree with each other while both diverge from processing intensity.

The concentration of favourable environmental grades among the NOVA 4 products, from which the lower coherence of that group and the negative association between EcoNorm and ICON follow, confirms at product level what reviews of life-cycle assessments report: plant-based substitutes have substantially lower environmental impacts than the animal products they replace (Nájera Espinosa et al., 2025), across greenhouse gas emissions, land use, and water use (Bryant, 2022). A favourable environmental grade for an ultra-processed substitute is the expected outcome of that assessment rather than a feature of the set. The negative association between EcoNorm and ICON expresses the same concentration through the index and, being partly a property of its construction, is not compared with that literature as an independent result. What the index adds is that the favourable footprint and the ultra-processed status, assessed in that literature as separate outcomes, are registered as a single measurable misalignment. The divergence is with the evidence at dietary level: in a representative sample of French adults, ultra-processed foods represented 19% of the diet and contributed 24% of its greenhouse gas emissions, an association that vanished after adjustment for energy intake (Kesse-Guyot et al., 2023), and in a scoping review of sustainable-diet indicators NOVA was synergistic with greenhouse gas emissions in 8 of 13 studies but discordant with fresh-water use in 8 of 12 (Tompa et al., 2025). The two levels compare different objects — a substitute with the product it replaces, and a class of foods with the remainder of the diet. At dietary level the contribution of ultra-processed foods has been reported as secondary to the type of plant-based food consumed (Daas et al., 2024), and the distinction between the two levels is itself pattern T5.

The asymmetry of the tension profile is consistent with product-level evidence on the two consumer-facing scores. Across 57,000 products in the United Kingdom and Ireland, more nutritious products were often more environmentally sustainable, with exceptions, and foods viewed as substitutable could differ markedly in impact (Clark et al., 2022); at food-group level, the foods associated with improved health have among the lowest environmental impacts, red meat the highest, and processed foods high in sugars harm health but can have relatively low environmental impacts (Clark et al., 2019). The low Nutri–Eco tension in the illustrative set confirms a relationship documented at both levels. The predominance of Tech–Eco tension has no counterpart in that evidence, since environmental evaluations do not register the industrial processes and the additives that ultra-processing introduces (Seferidi et al., 2020); the tension profile extends this observation by locating the divergence between the processing classification and the environmental score at product level and by measuring it relative to the other two pairs of dimensions.

The distribution of the paradox across categories can be set against two market surveys that examined the same categories on one dimension each. In Portuguese retail, 84.3% of 407 pre-packaged plant-based products were ultra-processed (Maganinho et al., 2024); in Dutch supermarkets, 67% of meat alternatives and 95% of milk alternatives among 916 products carried a Nutri-Score of A or B (Huybers and Roodenburg, 2023). The concentration of the paradox in meat alternatives, plant-based drinks, and dairy alternatives confirms both surveys and extends them by showing that the two properties they report separately co-occur in the same products; the lower frequency among simple legumes follows from the same gradient. Within each category, nutritional composition differs considerably across brands and product types (Nájera Espinosa et al., 2025), and products viewed as substitutable can differ markedly in environmental impact (Clark et al., 2022), so the category counts describe the illustrative set and not the categories as such.

The illustrative analysis demonstrates that the misalignment among the three assessments can be measured within the set; it does not demonstrate that the index improves any decision, and its uses in claim verification and in reformulation are applications that the measure makes possible, not results of the analysis.

8.2 Implications for policy and governance

The incoherences observed in the illustrative set are consistent with the fragmented application of technological, nutritional, and environmental assessments, while public communication often reflects only partial aspects of product performance. In the absence of an explicit coherence criterion, sustainability claims may therefore capture only a partial representation of product performance. When used as an orientative screening tool, ICON could support the governance of sustainability claims without competing with existing schemes (NOVA, Nutri-Score, Eco-Score), by signalling the degree of internal alignment among technological, nutritional, and environmental dimensions. Applied as a screening step, the index requires the value and the dimensional profile to be read together. This perspective aligns with recent calls for integrated ecological regulation of food systems, which argue that governance frameworks must simultaneously address the health and environmental dimensions of ultra-processed foods rather than treating them as separate policy domains (Northcott et al., 2023).

From a public policy perspective, a coherence filter could be conceptually aligned with the logic of Directive (EU), 2024/825 on strengthening consumer protection in the green transition (Directive (EU), 2024/825), by identifying the claims for which the remaining dimensions diverge from the one communicated, and which therefore warrant closer verification. At the methodological level, such screening could complement the objectives of the Farm to Fork strategy (European Commission, 2020) and the direction of the proposed Directive on explicit green claims (European Union, 2023), without implying the introduction of a new mandatory label, but rather by strengthening the integrated interpretation of existing assessments. For industry and retail, ICON may function as an internal and voluntary tool for establishing whether a reformulation that improves one dimension leaves the others unaligned, which is the configuration that exposes a product to challenge once a claim is made on the improved dimension. Applying the index requires the three assessments to be available for the same product. In the illustrative set this was the binding constraint on selection, and it remains the principal condition for wider application.

8.3 Future research directions

Building on the Tech–Nutri–Eco framework and the illustrative exercise presented, strengthening and validating coherence assessment requires further development along five major research directions.

• Expansion of the database — replication of the analysis on larger samples drawn from multiple sources (OFF, commercial databases, verified LCA datasets) to assess the statistical robustness and distributional behaviour of ICON values.
• Integration of the social dimension — the current Tech–Nutri–Eco axes do not include social dimensions (working conditions, economic accessibility, supply-chain ethics), although these aspects are becoming increasingly relevant in defining food sustainability.
• Linkage with quantitative LCA indicators — ICON can be analysed in relation to midpoint and endpoint indicators (e.g., GHG emissions, water use, eutrophication) to assess the extent to which internal incoherence is associated with higher levels of environmental impact.
• Differentiation within the ultra-processed category — construction of a continuum inside NOVA 4, distinguishing products by the number and type of industrial processes and by additive load.
• Extension to animal-source products — application of the same three assessments to dairy, meat, and their plant-based substitutes, category by category, so that coherence can be compared across the two classes.

Through the development of these directions, the Tech–Nutri–Eco framework may conceptually inform the future evolution of cleaner food production models, offering a systemic and verifiable approach for analysing and governing sustainability claims in the food industry.

8.4 Limitations

The analysis presented is illustrative and exploratory in nature, and the results cannot be considered representative of the European market for plant-based products. The limited size of the analysed dataset (49 products), as well as the selection constrained by the simultaneous availability of all three dimensions (Tech, Nutri, Eco), restrict the generalisability of the conclusions and may introduce selection bias.

The use of secondary data from public sources (Open Food Facts and manufacturer-declared information) entails uncertainties regarding the accuracy of product classification and the consistency of environmental assessments. In this context, Eco-Score is used exclusively as an illustrative proxy of environmental performance and does not substitute detailed quantitative LCA or PEF assessments; nor does it correspond to the French official environmental cost method, whose food component was presented in May 2026, after the data used here were collected. Recomputing the illustrative set under that method would test whether the coherence configurations reported are specific to the scoring system used. The normalisation procedure applied to the Tech–Nutri–Eco dimensions involves a controlled simplification of the original methodologies (NOVA, Nutri-Score, Eco-Score), without fully reproducing their granularity or underlying assumptions. The technological dimension therefore cannot discriminate within NOVA 4, which accounts for 31 of the 49 products. In addition, the proposed framework does not incorporate social dimensions of sustainability, which may be relevant in more comprehensive evaluations of food systems. The statistical analyses (Spearman correlations, Mann–Whitney U test, tension profiles, and clustering) were conducted on the illustrative set of 49 products. Their inferential scope is confined to this set: the associations hold within the sample but cannot be extrapolated to the European plant-based food market, given the non-probabilistic selection and the constraint of simultaneous availability across all three dimensions. The correlation between TechNorm and ICON is further shaped by the construction of the index and by the concentration of nutritional and ecological values in the upper range of the sample, and does not by itself establish a causal hierarchy among dimensions.

9 Conclusion

Current assessments of the sustainability of plant-based products are insufficient to describe their actual profiles when technological, nutritional, and environmental dimensions are structurally misaligned. In the absence of an explicit coherence criterion, favourable performance along a single axis may mask significant discrepancies between processing intensity, nutritional quality, and environmental impact. Treating multidimensional coherence as an evaluative principle shifts the focus from isolated performance to internal consistency. The illustrative application supports its tractability: the eco–ultraprocessing paradox proved both identifiable and frequent, and ICON separated ultra-processed products from the rest along a single interpretable measure. Such diagnostic indicators can operationalise this shift without substituting for or hierarchically ranking existing frameworks. For the verification of environmental claims under EU Directive 2024/825, a coherence criterion offers a means of exposing selective performance that single-dimension assessment leaves intact.

Generative AI statement

The author(s) declared that Generative AI was used in the creation of this manuscript. During the preparation of this manuscript, generative artificial intelligence tools were used to assist with language editing and clarity improvement. The authors reviewed and edited the content as necessary and take full responsibility for the content of the publication.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fsufs.2026.1943562/full#supplementary-material

Mean normalised Tech–Nutri–Eco dimensional profiles by plant-based product category (n = 49). Each line connects the mean normalised values of one category on the three dimensions. On the technological axis, higher values indicate a lower degree of processing. Supplementary material at the source

Sensitivity of ICON to the choice of coherence formula (n = 49). Each panel compares ICON with one alternative measure computed on the same normalised values; the dashed line marks equality. ICON is algebraically equivalent to the normalised range, and diverges from measures based on dispersion around the mean or on distance from the balanced point. Supplementary material at the source

Mean pairwise dimensional tension profiles by plant-based product category (n = 49). Each tension is the absolute difference between two normalised dimensions, so higher bars indicate greater divergence between them. Categories are ordered by decreasing median ICON. Supplementary material at the source

Exploratory coherence analysis (n = 49). Panel (a) shows the position of each product in the Eco–Nutri space, with colour indicating the technological value; the dotted lines mark the two paradox thresholds (EcoNorm = 0.50 and NutriNorm = 0.75) and the circled points are products with ICON = 0.000. Panel (b) compares the ICON distribution of NOVA 4 products with that of the remaining products. Products sharing the same pair of scores overlap in panel (a). Supplementary material at the source

Supplementary material Supplementary material at the source

Footnotes

Edited by: Charles Odilichukwu R. Okpala, University of Georgia, United States

Reviewed by: Juliet Bennett, The University of Sydney, Australia

Francesco Vella, University of Palermo, Italy