2.1. Ethics and positionality statement

2.2. Analytical framework

To examine decision-making in the four CDI cases, we developed an analytical framework that combines the three dimensions of participation, knowledge sources, and knowledge types. In the following sections, we explain the rationale for selecting these concepts and how we define and apply them in our analysis.

2.3. Research area and case studies

2.4. Data collection

Data collection was inspired by the innovation biographies approach [87,88], which analyses innovation as a spatially embedded, knowledge-intensive learning process that emerges through interactions among diverse actors [cf. 89]. Methodologically, the approach reconstructs the “biography” of an innovation through narrative interviews, tracing how it evolves over time, across contexts, and through changing actor constellations [87]. While we did not conceptualise CDIs as innovations, we drew on this approach because it provides a process-oriented lens to examine how initiatives develop through actor interactions and knowledge dynamics. Like innovation processes, CDIs unfold over time and are shaped by how knowledge is mobilised and applied. Reconstructing the “biographies” of the CDIs therefore enabled us to trace their developmental trajectories and analyse how participation in decision-making, as well as the inclusion of different knowledge sources and types, evolved throughout each initiative.

For each case, we conducted interviews with CDI leaders and staff across organisational levels, including members of the initiatives’ board or steering committee, office staff based in Maroantsetra, and field agents working in the villages. Where applicable, we interviewed additional actors involved, such as beneficiaries and members of stakeholder platforms. In part one of the interviews, we asked openly about the development and main changes along the timeline of the CDI. In part two, we added structured questions on decision-making processes within the CDI, and inclusion of scientific and local knowledge. S2 Table provides an overview of interviewee characteristics, and S1 File the full interview guidelines, including questions not used for this article.

Between May 2023 and December 2024, we conducted 14–28 interviews per case and 74 interviews in total. Interviews were held in Malagasy, French, English, or Swiss German, depending on interviewee preferences and language skills. Most occurred at interviewees’ workplaces, with a few conducted online. They lasted 15–110 minutes, depending on interviewees’ level of involvement in the respective CDI. Audio recordings were transcribed, and interviews in Malagasy were additionally translated into French. To fill data gaps, we complemented the initial interview data with secondary data, literature, and follow-up interviews.

2.6. Data analysis

We analysed the data through a qualitative content analysis approach [90]. First, using MaxQDA [91], we extracted relevant parts of the interview transcripts by coding them into three main categories: “participation”, “knowledge inclusion”, and “CDI development”. From these categories, we transferred all references to decision-making in an Excel table (see S3 Table). These included, for example, the choice of the CDI’s overall approach, the selection of intervention villages, decisions to change or expand the approach, as well as day-to-day management decisions mentioned by interviewees. For each decision, we recorded who was involved – distinguishing between actors who directly participated in decision-making versus those who were only consulted or contributed inputs. We then ordered the decisions chronologically, using secondary data where necessary, and consolidated them into 13–15 key decisions for each case. Key decisions therefore encompass several related smaller decisions; for example, in the FCI case, multiple adjustments during implementation mentioned separately by interviewees were clustered into the key decision “adaptations in the process”.

To structure the CDI biographies, we grouped key decisions into four functional phases: (1) design, (2) implementation, (3) adaptation, and (4) operation. Though arranged chronologically, we defined the phases more by their function than by their timing. For example, adaptation decisions – aimed at revising the original strategy – sometimes occurred years or even decades apart, but were grouped together according to their shared purpose. We defined design and adaptation decisions as those decisions that shaped the overall direction of the CDI. In contrast, implementation and operation decisions concern how the strategy was carried out: implementation refers to the initial application of the original design, whereas operation refers to ongoing application of the strategy after adaptations.

To analyse knowledge use, we applied the two dimensions of our analytical framework “sources of knowledge” (i.e., local, scientific, practitioner), and “types of knowledge” (i.e., systems, target, transformation) presented in section 2.2. We revisited all instances of knowledge inclusion in MaxQDA, categorised them by source and type, and linked them to corresponding decisions (see S4 Table). During this process, we encountered cases where decision-making was shaped by knowledge that did not fit into the three defined sources. To capture this, we added a fourth category – “other knowledge” – mainly encompassing target knowledge from global discourses, debates, participation in international knowledge fora, etc.

3.1. Full Circle Initiative (FCI)

FCI was founded in 2019 (1) by Swiss actors drawing on prior research activities in North-eastern Madagascar. With participation from a Malagasy researcher, the initiative was designed around two principles: promoting innovative, non-forest-dependent livelihoods (2) and local ownership (3). These design decisions were primarily based on scientific and practitioner systems and target knowledge. During implementation, local participation and the use of local knowledge increased. Village residents proposed livelihood projects, which were selected (4) by the project team and a district-level stakeholder platform based on predefined criteria. Local project owners led implementation (5) with guidance from the project team and external experts. Following a pilot phase, the process was adapted (6) using feedback from local, district, and national actors. FCI also facilitated village visioning workshops (7). Overall, this phase relied mainly on local and practitioner target and transformation knowledge.

In 2022, FCI reached a turning point, when most of the individual projects of local residents were considered unsuccessful. Although projects were locally proposed and technically supported, they often insufficiently considered relevant systems knowledge. For example, a vegetable garden was established near a riverbed but was later destroyed during a cyclone, leading to project abandonment and frustration for the project owner. This outcome might have been avoided had flood risk and cyclone frequency been considered.

That same year, FCI was integrated into the Wyss Academy for Nature (8), a global research-action centre, and shifted its aim towards systems transformation (9). The initiative now explicitly aimed to combine local and external knowledge (10) and added a focus on value chain development (11) and a digital knowledge platform (12). As in the design phase, key decisions were mainly taken by the Swiss-Malagasy steering committee, but now in consultation with the district team, and were informed by scientific and practitioner knowledge across all three knowledge types.

To operationalise the adapted strategy, a shared valley-wide vision (13) was co-developed with local communities and district actors, including inputs from national and international researchers. Future interventions were selected (14) based on an analysis of system levers and the village- and valley-level visions. Operational authority (15) was increasingly transferred to the Malagasy project coordinator. Overall, in this phase, the three knowledge sources and types became more evenly represented.

3.2. SEPALI

SEPALI’s parent organisation, CPALI (Conservation through Poverty Alleviation International), was founded in 2002 by a US entomologist (1). After initial research, silkworm rearing (2) was identified as a promising approach to achieve CPALI’s goals. Although SEPALI’s current director brought expertise on butterfly species from other regions, the team relied heavily on local farmers’ knowledge to identify suitable silk-producing species and host plants (3).

The programme was introduced in villages (4) around the Makira protected area, where silkworm rearing was initially unfamiliar and met with fear. Implementation drew on the Malagasy team member’s community engagement expertise and scientific knowledge, including research to dispel concerns about crop damage. Village selection was initially guided by collaboration with the Wildlife Conservation Society, but growing interest led residents to request the programme. After forming farmer associations, activities began with tree-planting (5). While the Maroantsetra-based team made key decisions, local members contributed implementation knowledge and, following training, associations independently managed village-level activities (6). SEPALI simultaneously secured buyers, set cocoon prices, and registered as a company in Madagascar (7).

When the initial buyer of the raw silk withdrew, SEPALI shifted to producing artisanal silk products (8), first with women in the villages and later through establishing a manufacturing workshop in Maroantsetra. The Malagasy team received product design training from a US expert. Despite some sales, the organisation remained donor-dependent. The 2014 rosewood crisis made securing conservation-related funding difficult. In response, CPALI successfully applied for funding for an insect-for-food programme (9), which drew on local and scientific knowledge, but was discontinued after the funding period due to low farmer interest. Meanwhile, silk production continued, but fluctuations in vanilla and clove prices reduced farmer participation, prompting SEPALI to collaborate with farmers in other regions to stabilise supply (10).

At the time of data collection, product design decisions (11) were led by artisans in Maroantsetra with input from international collaborators, and most sales (12) occurred through CPALI’s wholesale brand, Tanana Silk. Under a purchase guarantee agreement, CPALI covered salaries and managed sales risk, while SEPALI supplied products. Although sales covered most costs, both organisations continued seeking external funding for further development (13). In both the adaptation and operation phases, key decisions were taken at the international and district levels, without local-level consultation. Nevertheless, local knowledge remained present throughout all phases.

3.3. Masoala National Park (Masoala NP)

Masoala NP was established as an ICDP (1) in collaboration between international conservation organisations, the National Association for the Management of Protected Areas (ANGAP), and Madagascar’s Ministry of Forests. Park boundaries (2) were defined based on extensive ecological and socio-economic studies and participatory mapping of village territories [82]. Although communities agreed with the initial design, conflict potential existed from the outset, notably regarding cultivated land, settlement relocation, and cultural values attached to forests [92,66]. The ICDP framework included plans to provide sustainable livelihoods alongside conservation measures (3).

Following the park’s establishment, some livelihood support activities were introduced (4), such as training in intensified rice cultivation, with interventions selected by park management and technical experts. Consistent with national policy, forest management rights in buffer zones were transferred to local communities (5) through VOI (Vondron’Olona Ifotony) associations. While communities were not involved in the decision on the management transfer, they participated in drafting the VOI regulations.

Soon after the park’s creation, declines in global support for ICDPs led to budget cuts and a shift towards strict conservation (6) [93]. Park boundaries were redrawn multiple times (7), moving closer to villages – actions perceived by authorities as technical corrections but by communities as breaches of agreement [66]. Combined with the withdrawal of development support, these changes intensified conflicts [66,93]. More recently, renewed emphasis on participatory conservation led Madagascar National Parks to establish advisory committees (8). In Masoala, the committee included local authorities, government agencies, and other district actors, and, until 2019, evaluated community development proposals (10). The latest five-year management plan (9), renewed in 2023, was developed through a participatory process and additionally informed by scientific knowledge including ecological monitoring and other SMART data.

Based on the management plan, annual planning (11) is developed under the park director’s leadership with staff input and headquarters’ approval. SMART data, alongside staff observations and experience, guide decision-making. While the director retains management authority, staff may propose initiatives, and routine operational decisions such as patrol organisation (12) are taken independently by office and village staff.

3.4. Fandroakando

Fandroakando was founded in 2020 (1) by two Canadian tourists in collaboration with a guide from Maroantsetra, motivated by concerns over forest conservation and the endangered Farankaraina forest. Although the forest was already classified and managed by local VOI associations, limited resources constrained effective management. Following its establishment, Fandroakando collaborated with the VOI while also launching its own initiatives. The founders defined the NGO’s main activities, including patrols, reforestation, and an ecolodge (2). Design-phase decisions were primarily guided by practitioner target and systems knowledge, while scientific knowledge was incorporated through a species inventory conducted prior to the NGO’s creation, and local target knowledge through the establishment of an environmental club (3) proposed by local students.

For implementation of the NGO, the founders secured funding (4), set up operational procedures (5), and created a website, while on-site administration (6), including staff recruitment, was managed by the local guide as executive director. Communication during this period was constrained, particularly due to Covid-19 restrictions, which may partly explain the overall limited use of knowledge in decision-making during the implementation phase. At that time, Fandroakando encountered several implementation challenges, including difficulties in collaborating with the VOI and other conservation actors, and unclear roles and responsibilities within the team.

In early 2023, fraud prompted the dismissal of several staff members and the appointment of a new executive director, followed by the recruitment of additional staff and the formation of a new board (7). The founders, executive director, and board jointly developed the NGO’s first comprehensive strategy (8), informed by staff feedback and community consultations. Based on these inputs, and followed-up by a household survey, a key decision was to place greater emphasis on community development (9). An additional strategic shift involved increased collaboration with other conservation NGOs (10).

At the time of data collection, the founders played a more active role in major management decisions (11), alongside the executive director. Operational changes, including monthly meetings with VOI associations, had improved collaboration and conflict resolution. Office staff implemented activities with the VOI (12), incorporating input from founders, village staff, and other community members, who could also propose development projects (13). Routine decisions (14) were handled by office staff and field agents, and, in addition to practitioner knowledge, were informed by data collected during patrols.

4.1. Patterns of actor participation and knowledge inclusion in decision-making

We compare the four CDIs across the three dimensions of participation, knowledge sources, and knowledge types, highlighting key patterns and notable differences, and situate them within the broader literature.

4.2. Key insights for the design and management of CDIs

Both actor participation and the inclusion or exclusion of knowledge sources are shaped by asymmetrical power structures [21,20]. As illustrated by our cases, in externally initiated and funded CDIs, ultimate authority over initiative design – including whether decision-making power is devolved to local actors – typically rests with founders and funding bodies [107,4]. Although our data do not allow for a systematic assessment of why local communities or their knowledge were only marginally involved in certain phases, existing literature suggests that structural and institutional constraints – including tight design timelines, budget limitations, and fixed donor agendas – play a major role [25,96]. By contrast, SEPALI and Fandroakando emerged from the ideas of a few individuals and evolved without a formal strategy. In these cases, observed patterns – such as the inclusion of local knowledge but not local knowledge holders in SEPALI – appeared to stem less from structural constraints than from limited reflection on the importance of such inclusion. Decisions to involve local actors in decision-making or to combine diverse knowledge sources were, in most instances, the result of explicit choices by those in positions of authority. This underscores the asymmetrical power relations participation and knowledge inclusion are embedded in, while also reflecting the agency of CDI managers and funders in influencing these dynamics [21,4].

While early and comprehensive inclusion of local communities and diverse knowledge systems would be ideal, our cases, and other studies [96], demonstrate that externally designed CDIs can still evolve into more inclusive configurations over time. Our findings suggest that the implications of externally driven design processes depend on whether early decisions result in fixed parameters, such as the spatial boundaries of a protected area, or in adaptive approaches, as is more common in non-area-based CDIs. An additional question, however, is whether externally designed CDIs are open to fundamentally different worldviews and values [see 108,78]. In Madagascar, for instance, the externally introduced concept of “environment” (tontolo iainana) – distinct from land (tany) – has been criticised for contradicting traditional Malagasy understandings of human–nature relationships and dismissing local epistemologies [109,66]. Fully valuing local knowledge therefore requires not only inclusive procedures, but also openness to plural ontologies and the willingness to adapt core conceptual foundations accordingly [110].

At the same time, FCI’s experience demonstrates that valuing local knowledge and devolving decision-making power, while important, is not enough if relevant knowledge types are not adequately considered. The discontinuation of locally proposed projects, due to insufficient attention to systems knowledge, ultimately undermined FCI’s objective of fostering local ownership. Overall, among the three analytical frameworks – participation, knowledge sources, and knowledge types – the framework of knowledge types proved most effective in explaining major setbacks observed in the CDI biographies. Literature highlights that the relative emphasis of systems, target, and transformation knowledge may shift across phases [48,111]. Our findings align with calls for early deliberation of target knowledge with affected actors [103,48] and further indicate that integrating transformation knowledge already during the design phase – by drawing on evidence about effective intervention strategies and inclusive co-design – may enhance strategic robustness and reduce the need for redesign. At the same time, systems knowledge remains critical during implementation, as context-specific knowledge needs may only emerge at that stage. Overall, our results support arguments for the iterative integration of all three knowledge types throughout an initiative’s trajectory [58,103,48].

While our study demonstrates the analytical value of distinguishing between participation, knowledge sources, and knowledge types, practical recommendations converge around processes of knowledge co-production. Rather than limiting co-production to a specific phase, it should be embedded iteratively across all phases — from agenda-setting and CDI design to implementation, evaluation, and adaptation. To strengthen co-design and co-production of knowledge before the launch of an initiative, scholars propose introducing a “Phase 0”, that is, an intentional preparatory period [112]. Although originally discussed for transdisciplinary research projects, its core elements – developing a deep contextual understanding (including local perceptions, expectations, power relations, and existing conflicts) and building trust and shared intent – are equally relevant for CDIs. Additional promising approaches include co-producing a theory of change [113], participatory drafting of management plans [114], and designing co-production processes that engage rather than suppress tensions and power dynamics, for instance through “co-productive agility” [115]. Drawing on the framework of systems, target, and transformation knowledge, Seiferth et al. [56] further demonstrate how these knowledge types can be co-produced through stepwise, dialogue-based workshops. Together, these approaches provide practical entry points for embedding iterative, inclusive knowledge co-production throughout CDI design and implementation.

4.3. Limitations and outlook

By tracing participation, knowledge sources, and knowledge types across the biographies of four CDIs, this study sheds light on how these dimensions interact over time. However, the depth of the data did not allow for an assessment of the quality of participation and knowledge inclusion. For example, even if local visions are considered, it remains unclear whether they genuinely reflect people’s aspirations or are shaped by assumptions about what external actors want to hear. Similarly, while we observed efforts to combine different knowledge sources, we could not determine whether these resulted in equitable collaborations [78,99]. Future research should thus explore these procedural aspects in greater depth. Additionally, examining CDI trajectories using a power-oriented analytical approach could deepen understanding of the dynamics at play, including the reasons behind the in- or exclusion of certain actors and knowledge sources [21,20].

As noted in the introduction, all knowledge systems are continually shaped through exchange and adaptation [78]. Thus, what we call “local knowledge” might not be purely local, but rather shaped through interactions with external influences. In FCI, for example, it was observed that ideas brought forward by local community members were often similar to activities that had been introduced by previous CDIs. In order to advance scientific debates and CDI practice, it may be necessary to pay greater attention to the discourses that shape knowledge [e.g., 116]; the underlying assumptions of local communities, practitioners, and researchers [e.g., 117]; and the question of how to achieve emancipation from decades of external dominance.