Impact statement
Decentralised climate solutions have the potential to benefit millions of rural and underserved communities, yet they are often excluded from carbon removal markets because existing monitoring, reporting, and verification (MRV) systems are primarily designed for large industrial facilities. This paper demonstrates a practical pathway to benefit these communities by developing and deploying a robust digital MRV framework – FICRA (Fleet-Integrated Carbon Removal/Reporting Architecture) – across a fleet of small-scale pyrolysis reactors in India, Kenya, the United States, and Canada. By combining low-cost sensors, human-centred mobile applications, automated data validation, and transparent reporting, FICRA shows that high-integrity carbon removal is achievable even in locations with limited infrastructure, intermittent connectivity, and operators with varying levels of technical skills. This expands the reach of carbon markets to regions where biomass residues are abundant, but technical barriers have historically prevented participation. The resulting social impact includes new income streams for farmers, improved waste management, and support for Indigenous and rural communities seeking to adopt climate-positive technologies. Beyond biochar, the architecture introduced here offers a blueprint for scalable digital MRV applicable to other distributed carbon removal pathways such as enhanced rock weathering, modular bioenergy with carbon capture and storage, and small-scale gasification. The insights generated also contribute to evolving governance discussions, demonstrating how verifiers, registries, and policymakers can reliably assess projects composed of many small, mobile, and geographically dispersed units, while broadening the carbon market access to the underserved communities most poised to benefit from climate-related livelihood opportunities.
Introduction
As the carbon dioxide removal (CDR) sector matures, credits must transition from niche to credible, bankable assets; this demands rigorous, transparent, auditable methods to monitor, report, and verify CDR removal (Jaffer et al., Reference Jaffer, Smith and Zhou2024).
Monitoring, reporting, and verification (MRV) practices developed to date, however, are optimised primarily for centralised facilities with stable power, dense instrumentation, and proximity to accredited laboratories. Yet such paradigms exclude a large spectrum of CDR opportunities located in rural, low-infrastructure regions, where technical resources are scarce. This creates an equity gap: communities with significant mitigation potential are structurally disadvantaged by MRV systems not designed for their operational realities.
Biochar represents a particularly compelling case study: It currently constitutes the majority of delivered CDR, owing to technological maturity (Thengane et al., Reference Thengane, Kung, Gomez-Barea and Ghoniem2022; Thengane et al., Reference Thengane, Burek, Kung, Ghoniem and Sanchez2020a), applicability across diverse geographies (CDR.fyi, 2025; Puro.earth Registry, 2025), and the potential co-benefits (Thengane et al., Reference Thengane, Kung, Gupta, Ateia, Sanchez, Mahajani, Jim Lim, Sokhansanj and Ghoniem2020b). On the supply side, a handful of industrial-scale biochar projects currently dominate market issuance. These industrial biochar facilities exhibit high process stability and MRV readiness, whereas distributed systems – often based on artisanal kiln designs – face more variable conditions, limited documentation, and discounted credit prices (Quantum Commodity Intelligence, 2025). Historically, small-scale/distributed biochar efforts have often relied on artisanal, low-tech kiln designs (Kung et al., Reference Kung, Rigu, Karau, Gachigi and McDonald2015), manual measurement, and infrequent/manual verification, with high uncertainty in emission abatement. However, with the advent of small-scale, portable biochar technologies that are nonetheless engineered to be high-tech, this difference in the MRV treatment is no longer necessarily inherent to the distributed model itself, but rather arises from the absence of robust, scalable, and context-appropriate MRV architectures that can support high-quality data reporting from distributed but industrial-grade projects.
Recent advances in digital MRV (dMRV) suggest a pathway to overcome these constraints. For example, low-cost internet-of-things devices, encrypted telemetry pipelines, automated quality-control layers, and cloud-based audit trails have been proposed as mechanisms to increase transparency while reducing verification costs (Jaffer et al., Reference Jaffer, Smith and Zhou2024; World Bank, 2023). When designed around human-centred constraints – including semi-literate users, offline functionality, and culturally adaptable interfaces – digital monitoring, reporting, and verification (dMRV) offers the possibility of industrial-grade verification rigour across decentralised fleets, potentially enhanced by the application of artificial intelligence (AI) and machine learning techniques (Watts et al., Reference Watts, Potter, Mohan, Kumari, Thengane, Sokhansanj, Cao and Kung2023; Zou et al., Reference Zou, Garg, Sharma, Kung, Thengane and Caounder review). This paper constitutes an initial exploration of coupling these newer dMRV tools with a specific fleet of portable, continuous biochar systems known as the Takavators, with the intention to elevate distributed biochar to “high-tech, high-integrity” status even in remote, underserved communities.
Specifically, we present Fleet-Integrated Carbon Removal/Reporting Architecture (FICRA), a dMRV-pyrolyzer combination that occupies a distinct design space between capital-intensive, high-tech biochar MRV systems and the low-tech, artisanal techniques (Supplementary Table S1). The fleet-wide approach enables cross-validation, anomaly detection across heterogeneous contexts, and the potential for pooled MRV infrastructure (e.g. cloud analytics) that amortises the instrumentation cost across multiple sites. In the following sections, we describe the architecture, deployment, results, and potential pathways to scale such a decentralised dMRV-pyrolyzer framework. As it stands, FICRA is currently an implementation architecture designed to operate on top of existing registry-approved MRV frameworks. Currently, it does not introduce new crediting rules, permanence definitions, or systems of record. All components described in this paper are intended to operate in parallel (aka shadow accounting) with conservative, registry-aligned MRV and are evaluated solely for their potential to reduce operational burden and verification cost without compromising integrity, until such a day that sufficient evidence supports relying on FICRA exclusively for crediting.
Although this paper focuses on biochar, the underlying MRV architecture has relevance to other distributed CDR modalities – such as enhanced rock weathering, modular bioenergy with carbon capture and storage (BECCS) units, and certain ocean alkalinity enhancement strategies.
Materials and methods
Deployment context
We selected Takachar’s portable continuous biochar reactors (“Takavators”) as a case study because the Takavators already have a track record in diverse biomass conditions, ranging from rice husks, coffee husks, groundnut shells, wood chips, bagasse, coconut shells, and pelletised biomass (Kung, Reference Kung2017; Kung et al., Reference Kung, Shanbhogue, Slocum and Ghoniem2019; Kung and Ghoniem, Reference Kung and Ghoniem2019a) under different moisture contents (Kung and Ghoniem, Reference Kung and Ghoniem2019b), particle sizes (Kung et al., Reference Kung, Thengane, Ghoniem, Lim and Sokhansanj2021), and reactor setpoints (Kung et al., Reference Kung, Thengane, Shanbhogue and Ghoniem2019; Kung et al., Reference Kung, Thengane and Ghoniem2020a; Barr et al., Reference Barr, Kung, Thengane, Mohan, Sweeney and Ghoniem2020). This gives us a wide aperture of biomass use contexts and datasets to stress-test FICRA. We analysed a fleet of 17 Takavators deployed across India, Kenya, the US, and Canada between 2022 and 2026 (Figure 1a), some of which are already Puro.earth-certified. The units operated often in collaboration with farmer cooperatives or Indigenous communities, under rural, variable conditions. Reactors were moved between sites, so the fleet-level analysis treats each physical unit as a node in a dynamic, geographically distributed production network.
(a) Parade of Takavators deployed in different contexts. (b) Map of sensors on a representative Takavator. (c) Infrastructure of the controller input/output connected to the cloud.
Figure 1. Long description
Figure 1 introduces the Takavator system and its digital instrumentation in three panels. Panel (a) shows photos of portable Takavator biochar reactors deployed across diverse operating environments, illustrating use in different countries and feedstock contexts. The reactors are compact, trailer- or skid-mounted continuous pyrolysis units designed for decentralized operation. Panel (b) presents a schematic of a representative Takavator, identifying the locations of key process instrumentation, including thermocouples positioned along the reactor and flue gas path, load cells beneath the biomass feed hopper and biochar collection bin, blower controls, and the screw conveyor. Panel (c) illustrates the digital control architecture, showing sensors and actuators connected to a microcontroller-based control board that records telemetry, buffers data locally, and securely uploads operational data to cloud services for remote monitoring, validation, and reporting.
The Takavators are designed for off-grid use, requiring no external energy/fuel to run the thermochemical reaction; to power the electronics, they run on single-phase electrical power, or using solar panels or diesel generators. The reactors are operated by locally hired staff with limited training. These constraints shaped FICRA, pushing us towards robust sensors, simple interfaces, and automation that can tolerate intermittent connectivity and human errors.
Project baselines
Some of the deployments have already undergone successful Puro.earth certification, and FICRA is layered on top of these using shadow accounting that currently does not influence the underlying registry-approved processes. In decentralised biomass contexts, feedstock disposition pathways and counterfactuals may vary geographically and seasonally, and such is already accounted for in the existing Puro.earth verification reports. While this manuscript does not quantify such baseline heterogeneity extensively, the system architecture is structured to retain the evidentiary granularity (e.g. AI-assisted biomass purchase receipts) required for registry-level additionality review.
Reactor design and sensor instrumentation
Key process control parameters include screw conveyor speed (dictating the residence time) and blower speed/positions (dictating the temperature profile). Reactors were instrumented with thermocouples at multiple points along the reactor and flue gas path (Kung et al., Reference Kung, Thengane, Lim, Sokhansanj and Ghoniem2020b), electronic scales under biomass feed hoppers and biochar collection bins, and power meters. Figure 1b illustrates the thermocouple and blower layout.
FICRA ensures each Takavator is connected, via the controller, to the cloud, such that the sensor/actuator signals are uploaded/monitored near-real-time. To enable cloud connection, a modular microcontroller-based control board was designed for integrated datalogging, local circular buffers sufficient for 1 month of time-series data, and standardised interfaces for thermocouples, load cells, and digital inputs (e.g., safety interlocks) (Figure 1c).
Many Takavators routinely operated 24 hours/day, 6 days/week, with the 7th day for preventive maintenance. We recognise that when a reactor starts from a cold state, there is a ~ 1-hour transitory period during which the reactor heats up to steady state (Kung et al., Reference Kung, Thengane, Ghoniem, Lim, Cao and Sokhansanj2022); likewise, whenever the reactor shuts down, there is a cooling period. Accordingly, we have trained FICRA to parse out the transitory events from steady state as discrete production intervals. Steady-state operation – reflected in consistent temperature and residence-time profiles – is used as a predictive indicator of both output quality and emission behaviours.
Digital MRV data architecture
FICRA’s digital architecture (Figure 2a) is designed around three goals: ensuring tamper-evident, auditable records of production and end-use; reducing verification cost by automating routine data quality checks; and providing near-real-time visibility to external stakeholders. From the perspective of carbon registries, FICRA could ensure a transparent/auditable data trail; include redundant proofs for every data point; and produce registry-ready reports in standardised formats that can be automatically ingested. For clients/project operators, the design emphasises data privacy (especially amongst independent operators sharing the same production facility), sovereignty, and security, while enabling full-cycle project management via dashboards and scheduled analytics that surface actionable insights.
(a) FICRA’s data architecture. (b) Corresponding data operations flow diagram.
Figure 2. Long description
Figure 2 illustrates the digital architecture and data workflow underpinning FICRA. Panel (a) presents the system architecture, beginning with instrumented Takavator reactors that generate telemetry from sensors and controllers. Data are transmitted through edge hardware and secure cloud infrastructure, where they are ingested, stored, validated, and processed. Automated rule-based checks, AI-assisted anomaly detection, and human review ensure data quality before records are promoted to validated datasets. The validated information feeds dashboards, analytics, registry-ready reports, and external stakeholders such as project operators, buyers, and carbon registries. Panel (b) focuses on operational workflows, showing how machine telemetry is combined with human-reported information (e.g. feedstock records, lab results, production events, receipts, end-use documentation) to establish an auditable chain of custody.
At the edge, the reactor controller publishes telemetry packets containing timestamped sensor readings and machine state variables to a gateway running a lightweight, encrypted messaging protocol. Whenever connectivity is available, the gateway forwards these packets to a cloud ingestion layer. Local caching logic ensures that measurements are retained during network outages and uploaded automatically once connectivity resumes, preventing data loss in low-infrastructure contexts. In parallel, human operators submit non-machine data such as production and end-use information via an Android-based mobile application, and supervisors may edit or approve records through a web dashboard.
In the cloud, these datasets pass through a multi-layer validation pipeline that combines deterministic rule-based checks (e.g., sensor range, continuity, and physical constraints) with probabilistic AI-driven anomaly detection. When required – for calibration events, experimental runs, or configuration updates – human-in-the-loop review adds a final oversight layer. Once validated, data are transformed into registry-compliant summaries, thereby reducing the need for manual field reporting/summation. For industrial technologies such as the Takavator, some carbon registries are exploring or beginning to implement direct application programming interface (API) integration, which may help accelerate verification cycles, reduce physical audit needs, and support automated estimation of proxy variables otherwise difficult to measure directly, such as gas composition or biochar stability indices.
For auditability/traceability, we found the medallion (bronze-silver-gold) architecture to be a suitable framework. The bronze layer stores immutably archived raw telemetry, mobile app submissions, and receipt images. The silver layer contains cleaned/validated datasets, produced by rule−/model-based quality checks that flag anomalies, enforce internal consistency (e.g. enforcing moisture between 0 and 100%), and reconcile biomass input, gas emissions sampling, and biochar outputs. The gold layer aggregates these validated records into registry-facing reporting views, such as lot-based tonnes of CDR, traceable by reactor, date range, and site (Supplementary Table S2). Because of this architecture, data is never deleted, only transformed, and every transformation is traceable through Delta Lake’s time-travel features.
Human-reported interfaces
Machine data alone are insufficient for registry-compliant MRV; verifying chain-of-custody and biochar end-use requires human-reported information as well, as summarised in Figure 2b. Thus, we worked with our community partners and Valorize Systems to co-develop mobile/web applications called CarbonPilot (Supplementary Figure S1) allowing field operators to record each production event, including feedstock type, batch weight, measured moisture, operating issues, and end-use details such as the farm/plot where biochar was applied.
The mobile interface is co-iterated with semi-literate field operators. To engage with stakeholders meaningfully, we have adopted a series of human-centric discovery techniques as described in Al Haddad et al. (Reference Al Haddad, O’Neal, Batarseh and Martoncik2021), the National Science Foundation I-Corps program (NSF, 2025), and Gender-Based Analysis Plus (GBA+) to engage, co-design, test, and iterate with various operators in India, Kenya, the US, and Canada. After the initial discovery phase, structured co-design sessions were held during early deployment, where field teams, supervisors, and registry verifiers jointly reviewed the workflows. Feedback loops from these workshops led to iterative refinements.
We designed a guided, step-by-step workflow with built-in validations minimising cognitive load while enforcing procedural accuracy. Each submission is automatically geo-tagged, time-stamped, and associated with device metadata (Supplementary Figure S2), creating verifiable digital proofs at every stage. These data points collectively establish a complete, traceable chain-of-custody for each batch, supported by proxy methods such as FIFO (first-in-first-out) accounting to reduce field burden while maintaining traceability.
On the other hand, the web application used by project supervisors/owners consolidates financial records, lab reports, and compliance documents into a unified workspace. Standardised templates, structured forms, and workflow automation facilitate accurate record-keeping. AI-assisted text generation and validation tools further simplify the interpretation of complex registry rules, enabling non-technical managers to track progress and compliance in real time.
As an illustration (Figure 3), feedstock sourcing is documented through digital receipts that capture biomass weights and transport distances. AI-assisted data entry, inventory cross-checks, and approval flows help reduce transcription errors. Moisture content measurements can be recorded using portable sensors or oven-drying protocols to support consistency and potential future model refinement.
(a) Screenshot of a representative biochar sales receipt uploaded from the mobile app and results from automated AI parsing, illustrating automated extraction of structured information from receipt images (e.g. parsing out 8 × 50 kg bags) and presentation of extracted information for human review. Sensitive buyer information has been redacted. (b) A human-in-the-loop pipeline reconciles three readings per field: human entry, GPT-4.1, Claude Sonnet 4.6 – surfacing only disagreements for a final reviewer. The second model runs only on conflicted entries to limit cost. Error against the adjudicated final falls at each layer: (a) the surviving-error rate drops from 10.5% (human-only) to 0.8% after GPT, to 0.4% after Sonnet, leaving final review; (b) most human data-entry errors are caught by GPT, with Sonnet and the final reviewer mopping up the residual.
Figure 3 Long description
Figure 3 demonstrates how AI-assisted validation improves the accuracy of human-reported MRV records. Panel (a) shows a representative biochar sales receipt uploaded through the mobile application alongside automatically extracted fields including invoice number, delivery date, quantity, units, and total price. Confidence scores are displayed for each extracted field, while a reviewer interface compares values from the receipt, AI outputs, and the original spreadsheet entry so disagreements can be resolved before approval. Sensitive buyer information has been removed. Panel (b) quantifies error reduction across successive validation stages. The proportion of incorrect field entries decreases from 10.5% for human data entry alone to 1.4% after GPT review, 0.4% after a second AI model (Claude Sonnet), and 0% following final human review. A second chart shows most transcription errors are corrected by the first AI model, with progressively fewer requiring subsequent intervention.
Likewise, another illustration is through biochar end-use and sequestration. End-use tracking is achieved through digital documentation embedded in the MRV workflow. Field operators record and photograph evidence of biochar mixing or blending, and each sale is supported by AI-assisted receipt entry. Chain-of-custody checks within the cloud backend prevent double counting of carbon credits across facilities or buyers. Periodic stakeholder engagements confirm that biochar is applied in eligible end uses, such as soil amendment (Thengane et al., Reference Thengane, Kung, York, Sokhansanj, Lim and Sanchez2020c) and fertiliser substitution (Anshu et al., Reference Anshu, Lee, Mohan, Kung, Sokhansanj, Cao and Thengane2024; Thengane et al., Reference Thengane, Kung, Hunt, Gilani, Lim, Sokhansanj and Sanchez2021) as consistent with registry guidelines.
Sampling and measurements
Biochar quality and permanence claims in this study follow the Puro.earth biochar methodology as a standard guideline, which requires characterisation of total organic carbon, H/Corg ratio, ash, and other properties linked to long-term stability (Puro.earth, 2022). Because the reactor fleet operates in low-infrastructure settings where accredited laboratory services may be far away or expensive to access, we evolved a tiered gold-silver-bronze laboratory strategy (Table 1).
A tiered laboratory testing strategy for biochar and exhaust gas
Table 1. Long description
Table 1 compares three tiers of laboratory testing for biochar elemental composition and exhaust gas. Columns show the testing tier, elemental analysis method, flue stack analysis method, and relative price
The gold, or first, tier uses accredited laboratories. Elemental analysis is conducted by registry-accredited labs such as Eurofins or Soil Control Laboratories. Flue stack testing uses approved, calibrated methods recognized by the Central Pollution Control Board. Cost: $$$.
The silver, or second, tier uses other third-party laboratories. Elemental analysis may be performed by research universities, such as the University of British Columbia’s Biomass and Bioenergy Research Group. Gas is collected in Tedlar bags and sent to external laboratories. Cost: $$.
The bronze, or third, tier uses rural on-site infrastructure. Elemental analysis uses an on-site thermogravimetric analyser and optical measurements, while exhaust gas is tested with in-line portable analysers. Cost: $.
On a daily basis, biochar samples are collected. At some field production sites, we have built local instrumentation (such as a thermogravimetric analyser styled from a small muffle furnace) to conduct daily tests. These local tests are deemed to be “bronze tier”: While they give us some rapid, real-time indication of the biochar quality, they by themselves alone are insufficient for the purpose of carbon credit certification. On a weekly to monthly basis (depending on the specific site), aggregated biochar samples were dispatched to a “silver tier” testing facility, usually a local collaborating university such as the Biomass and Bioenergy Research Group at the University of British Columbia in Canada, the Thermal, Electronic, Elemental, Magnetic, Porosity and Optical (TEMPO) Facility at the University of California, Santa Barbara in California, or Indian Institute of Technology Roorkee in India, for elemental analysis and to cross-validate/correlate with the data points measured locally in third-tier analyses for accuracy. Finally, once every quarter, the aggregated biochar samples are sent to a “gold standard” registry-accredited laboratory such as Eurofins and Soil Control Laboratories for verification against second-tier and third-tier tests to ensure conformity with the relevant International Organisation for Standardisation (ISO) or Deutsches Institut für Normung (DIN) standards. Biochar sample selection follows a clustered design that prioritised new reactor – feedstock combinations and changes in operating conditions, while avoiding redundant sampling of long-stable configurations, principles which are consistent with Puro.earth’s 2025 biochar methodology.
Flue gas emissions during reactor operation likewise follow a tiered testing approach (Table 1). Each Takavator is equipped with gas sampling ports in the chimney region (Supplementary Figure S3). We have invested in a handheld in-line gas analyser (Chemist 600) for near-real-time monitoring of the flue gas for CO, NOx, O2, CO2, and CH4 locally. To obtain external validation of these local test reports, we periodically collect flue gases in Tedlar bags to send to third-party laboratories (e.g. Eurofins) for analysis on a wider spectrum of criteria pollutants (CO, NOx, SOx, CH4, CO2, and particulates). Finally, we have set up a gold-standard laboratory approved/calibrated by the Central Pollution Control Board following U.S. Environmental Protection Agency (EPA) standards, where all above-mentioned species are quantified.
Validation and quality control
FICRA’s data validation pipeline combines rule-based checks, statistical models, and human-in-the-loop review. At ingestion, basic format checks ensure all required fields are present and values fall within physically plausible bounds (e.g. non-negative weights, realistic temperature ranges). These rules catch common data entry errors such as misplaced decimal points, unit confusion, or inadvertent duplication of records.
For time-series telemetry, additional algorithms detect anomalies in weighing and temperature data. In weighing data, patterns such as sudden large jumps, prolonged periods at a fixed non-zero value, or step changes coinciding with operator activity are flagged as potential human interference (for example, a person stepping on the scale or forgetting to tare the scale). For temperature data, the system monitors the coherence of different thermocouple channels and flags implausible combinations (such as an exhaust temperature exceeding combustion-zone temperature by a large margin) that may indicate sensor failure or wiring issues.
Likewise, receipts/records uploaded by a human via the mobile or web applications are also first checked by algorithms to parse out key information such as the date, location, weight, etc. (Figure 3).
Records failing automated checks are routed to a human review queue. MRV staff compare human-reported data with machine telemetry, consult receipt images, and contact field staff for clarification. Only after passing both automated and human checks are records promoted to the gold layer and included in CDR calculations. This multi-layered validation approach is designed to reduce workload on verifiers/registries while maintaining transparency, as all raw and corrected values, as well as the reason for each correction, are preserved in an immutable audit log.
Compilation
Once telemetry, human-reported observations, and validation outputs are consolidated into the gold layer, the curated dataset is routed to a Streamlit-based application that serves as the primary interface for internal users. This application generates a suite of interactive dashboards and reports that allow operations and R&D teams to track reactor performance, monitor ongoing experiments, and review key operational indicators in near real time. By drawing exclusively from the validated gold dataset, the dashboards ensure that all displayed information reflects the most reliable and audit-ready version of the underlying records.
A parallel objective of the compilation layer is to produce reporting artifacts aligned with registry-facing life-cycle assessment (LCA) requirements. To this end, the reporting views in Streamlit structure outputs to match variables typically requested by registries and LCA practitioners, including electricity/water/fuel consumption, input/output mass flows, and indicators of operational stability. Because these views are preformatted according to standard templates, users can export registry-ready packets without extensive manual reprocessing or spreadsheet manipulation, reducing opportunities for transcription errors and ensuring consistency across reporting cycles.
In addition to serving as a repository for validated batch-level data, the Streamlit layer also supports real-time or near-real-time visualisation by ingesting live sensor streams through the Azure IoT Hub-Databricks-Streamlit pipeline. This enables operators and engineers to monitor active production runs, identify abnormal process behaviour, and validate experimental conditions as they unfold. The combination of real-time visibility with gold-layer reporting thereby links day-to-day operations with longer-term MRV obligations, providing a unified environment for both internal decision-making and external documentation.
To make this process transparent to registries and buyers, the data are published daily in a real-time dashboard shown in the Supplementary Figure S6.
Preliminary results
FICRA is an ongoing experiment, including benchmarking against more manual, registry-approved MRV protocols for precision/accuracy. This section presents the initial lessons learned.
Co-design outcomes
Decentralised dMRV systems can only succeed with end-user buy-in. The Takachar fleet was deployed across regions with vastly different linguistic, cultural, and technological contexts – from tribal communities in Northern Canada to smallholder farmers in India. These human-centred co-design sessions surfaced critical issues such as unit confusion (kilograms versus local units), sequence errors in data entry, difficulty in interpreting error messages, and the need for photographic prompts for key steps (for example, taking a picture of a weigh scale reading or the biochar application site). Iteratively, the app has evolved toward minimal text, the use of icons and colour cues, and constrained workflows that prevent skipping critical fields. A complementary web dashboard allows supervisors and MRV staff to review submissions, compare them with machine telemetry, and approve, reject, or request clarifications before records are promoted to the validated dataset. As we noticed that some users lack data access at the field site and only reconnect after work, we have implemented offline data queues for intermittent connectivity. This participatory process is not a one-time activity but an evolving partnership: field users are treated as co-developers rather than mere data collectors, ensuring that the resulting system is intuitive, resilient, and grounded in lived experience.
Training across a distributed fleet of multiple sites, each with their unique contexts/setpoints, posed another challenge. To minimise onboarding friction, the MRV system has been engineered to be process-agnostic but parameter-aware – that is, able to automatically adapt to local reactor configurations, feedstock types, and LCA-relevant processes (e.g. solar power versus diesel generator) while maintaining a standardised data model for reporting. This enables operators from different geographies to use a consistent workflow with minimal training, while still capturing project-specific nuances needed for lifecycle assessments and registry submissions.
Qualitative feedback from operators indicates that the revised interface felt more “for them” rather than “for auditors.” As an example, Supplementary Figure S4 illustrates how interface elements evolved in response to these engagements.
Mitigating human error
Despite technological automation and best human intentions, human factors remain critical to data quality. Several instructive field incidents underscored the importance of designing systems that anticipate and gracefully handle user errors by implementing the above-mentioned automation with human-in-the-loop.
One classic example of a type of human error resides in hand-operated moisture content measurement. Early pilots revealed that manual moisture calculations – performed with analogue scales and oven-dried samples – were prone to arithmetic mistakes, misplaced decimal points, or confusion over wet-basis versus dry-basis conventions. This led to physically implausible results (>100% moisture). In one particularly instructive instance, a field operator, accustomed to adding the biochar weight cumulatively, mistakenly added 10 randomly sampled moisture results rather than taking the average. To address this, we implemented automated moisture computation logic within the mobile app, complemented by affordable digital moisture sensors and AI-driven calibration based on reference data. This has eliminated most manual arithmetic and improved measurement reliability, resulting in our “no math” policy.
Another example of human error is with respect to biomass and biochar scale readouts. We have been accounting for unexpected incidents such as field workers accidentally stepping on the scale (Figure 4b), forgetting to remove the bucket from the scale for a long time (Figure 4d), and forgetting to tare the machine (Figure 4e). Based on these observations of actual human usage, we designed algorithms that can account for these abnormalities automatically, based on duration thresholds, step detection, and consistency checks with reported batch weights. The cleaned series (Figure 4f) provides a much clearer basis for computing biomass throughput and validating human-entered weights.
Representative electronic scale signals in time showing discrete “weighing” events, as well as abnormalities notated by green arrows: (a) normal weighing signals; (b) a human has inadvertently stepped onto the weighing scale; (c) people cleaning the reactor up after shutting the reactor down; (d) workers leaving a filled bucket on the weighing scale for extended periods of time; (e) workers forgetting to tare the machine; (f) an automatic rule-based cleaning of errors of weigh data; (g) new method with acknowledgement pressed by field worker as source of truth; and (h) raw input weights, acknowledged button press events and cumulative weights over time.
Figure 4 Long description
Figure 4 illustrates how FICRA mitigates common weighing errors during field operation. Panels (a–f) show representative electronic scale signals over time. Panel (a) presents normal weighing events as discrete increases in measured weight. Panels (b–e) show common anomalies, including an operator accidentally stepping on the scale, reactor clean-up activities after shutdown, a filled bucket left on the scale for an extended period, and failure to tare. Green arrows identify these events. Panel (f) shows the corresponding weight record after automated rule-based filtering corrects these anomalies to produce a cleaner estimate of biomass throughput. Panels (g) and (h) present an improved workflow in which operators press an acknowledgement button to confirm valid weight measurements. The resulting dataset combines raw scale readings, acknowledgement events, and cumulative biomass weight, reducing manual recording while preserving the operator as the authoritative source of measurements.
We realise that even the best models contain errors and hence are better suited as validation. There is a need to enable field workers to be the source of truth for field data. Hence, we built an automated weight acknowledgement hardware setup, which allows workers to simply press a button to record the weight – eliminating all manual data collection and allowing real-time view of cumulative weight and feed rate.
Sensor reliability
Sensor failures posed another class of challenges, given the variability of field environments (dust, heat, power fluctuations). First, we have evolved sensor selections such as thermocouples that are designed to withstand these unforgiving environments. Second, we have been creating some redundancies in the sensors (e.g. a network of thermocouples measuring self-correlated temperatures) such that if one sensor went offline, we would still have partial visibility to the system. The system therefore incorporates multiple layers of redundancy and intelligent fallback.
Automatic detection routines continuously monitor telemetry for signs of sensor malfunction, such as implausible flatlines, erratic oscillations, or divergence from correlated metrics (e.g., a temperature sensor reading steady at 25°C during active pyrolysis). When anomalies are detected, the system triggers alerts for human intervention involving sensor checks, leading to possible sensor replacement or recalibration, while preserving recent valid data for continuity.
Supplementary Figure S5 illustrates a diagnostic dashboard that summarises channel-level statistics and automatically suggests which sensors were likely malfunctioning. Once flagged, these issues could be addressed by field technicians at the next maintenance visit. Importantly, data from failed sensors were excluded from MRV calculations during the affected intervals, and conservative reconstruction methods were used only where justified by corroborating signals, such as redundancy between sensors or strong correlations with control variables.
Dashboards
A key FICRA design goal is to make MRV data not only accurate but also intelligible to external stakeholders in near-real-time, including registries, buyers, and rating agencies. To this end, we developed aggregated dashboards that expose both production metrics and MRV status. The real-time dashboard (Supplementary Figure S6a) provides per-reactor views of current operating conditions, recent production events, and the status of data synchronisation. The aggregated dashboard (Supplementary Figure S6b) presents fleet-level summaries of biomass processed, tonnes of biochar produced, and tonnes of CDR, disaggregated by country, reactor, and time period.
These dashboards serve several functions. Internally, they allow operations and MRV staff to identify underperforming sites, reactors with recurring data issues, and gaps in sampling coverage. Externally, they provide buyers and registries with an at-a-glance understanding of project activity and data completeness. The transparent linkage from dashboards back to underlying records and raw telemetry helps support third-party verification and independent ratings.
Registry engagement
The FICRA project has contributed to broader conversations regarding verification approaches for decentralised reactor fleets. Working collaboratively with registry partners such as Puro.earth, Isometric, and Rainbow, we participated in discussions around new audit configurations that treat each mobile reactor as a distinct production node within a single certified facility, with each new added reactor in the region as a capacity expansion rather than requiring a de novo audit. This adaptation allows the same rigorous standards applied to large, stationary biochar plants to be extended to distributed microreactors, without compromising traceability or data quality. Some of these changes have been reflected in Puro.earth’s 2025 Biochar Methodology, Isometric’s “Distributed and Small-Scale Projects v1.0” (2025), and Rainbow’s “Distributed Closed-Kiln Biochar Methodology v1.0” (2026).
Discussion and conclusion
Implications for scalability
The technical contributions of the FICRA framework can be summarised along three axes. First, it implements a multi-tier evidentiary classification architecture that preserves raw telemetry while progressively refining validated datasets. Second, it couples anomaly detection with human-in-the-loop escalation mechanisms to preserve auditability without sacrificing operational autonomy. Third, it demonstrates that such enforcement logic can be deployed across geographically distributed industrial units under different biomass/LCA contexts without a prescriptive centralised plant infrastructure. Together, these elements have the potential to differentiate FICRA from conventional batch reporting or purely documentation-based MRV systems.
What is important in the short term is that our preliminary results suggest that digital MRV has the potential to materially simplify and therefore reduce the marginal cost of verifying additional reactors and sites once the core architecture is in place, which we hope to quantify more deeply in a subsequent study. By automating data collection, basic quality control, and report generation, FICRA shifts verifier effort toward targeted review of anomalous cases rather than routine checks of every record. This is especially important for scaling to hundreds or thousands of small units, where per-site verification costs would otherwise render projects uneconomical. In the Takachar context, dMRV made it feasible to monitor a fleet spanning four countries without a corresponding linear increase in MRV staffing, and it allowed underserved communities to participate in high-integrity markets traditionally dominated by industrial facilities, resulting in rural institutional capacity building for carbon removal.
In addition, the combination of time-series telemetry and structured human-reported data supports more granular risk assessment by buyers and rating agencies. Rather than relying solely on method-level assumptions, stakeholders can examine project-specific evidence on reactor uptime, data completeness, sensor reliability, and operator behaviour. This aligns with emerging expectations in the CDR ecosystem for more transparent, data-rich disclosures and provides a template for how decentralised projects can demonstrate comparable integrity to centralised plants.
Generalisability
More fundamentally, the architectural abstraction underpinning FICRA is defined not by a specific CDR pathway but by the enforcement of bounded operating envelopes for distributed material transformation units. Any process in which physical variables (temperature, residence time, flow rate, chemical composition, or application dosage) can be constrained within predefined stability regimes may be governed by the same interval-based certification logic. Biochar production serves as a case study. The framework is equally applicable to mineral amendment deployment, alkalinity flux management, modular chemical synthesis systems, or other physicochemical transformation processes.
Furthermore, the portability of FICRA lies in its separation of concerns: the edge layer handles local sensing/control; the data ingestion layer standardises/timestamps records; the validation layer enforces physical/logical consistency; and the reporting layer aligns outputs with registry requirements. Only the specific variables, sampling strategies, and domain-specific rules need to be adapted when applying the framework to a new CDR pathway.
Governance and standards
The FICRA experience also raises questions for the governance of CDR standards: Many methodologies implicitly assume static facilities with fixed boundaries and stable configurations. Reactor fleets roaming amongst heterogeneous operating conditions challenge these assumptions. Our experience suggests that registries may need to more explicitly recognise “fleet-based” project definitions, where a project boundary consists of a set of standardised, sensorised units (a software-hardware hybrid) that may move between sites over time.
FICRA also enables more continuous forms of oversight: Rather than treating verification as a sparse, episodic exercise tied to specific issuance events, registries could require regular submission of machine-readable summary statistics. This could support ongoing risk monitoring and more dynamic adjustment of discount factors or buffer contributions in response to observed performance. Simultaneously, greater data availability creates obligations around privacy, data protection, and the representation of local stakeholders’ interests, particularly when projects operate in communities with limited bargaining power.
Limitations and future work
The work presented here represents a conservative first iteration of FICRA, with future efforts focused on benchmarking, cost forensics, and further structured registry engagement to evaluate which components may be suitable for broader adoption. Simultaneously, future work will benefit from evaluating FICRA across larger fleets and additional geographies across different CDR pathways. Second, although the current validation pipeline effectively detects and corrects many classes of errors, some failure modes – such as subtle sensor drift or systematic under-reporting of certain activities – remain challenging and may require more sophisticated statistical/physics-informed models. Third, we have not attempted a comprehensive economic analysis of MRV cost savings compared to traditional approaches; such analysis would be valuable for informing policy and investment decisions.
Furthermore, FICRA primarily addresses digital traceability, operational monitoring, and data integrity. However, it does not independently guarantee that biochar remains within approved end-use pathways after production. Risks such as combustion, disposal, stockpile loss, or diversion to non-eligible uses require complementary controls including chain-of-custody procedures, contractual requirements, stakeholder engagement, registry oversight, and periodic field verification. Consequently, FICRA should be viewed as one component of a broader assurance framework rather than a complete solution to permanence and reversal risk.
Future technical development should focus on deepening integration between reactor control systems and MRV logic, expanding the use of machine learning for anomaly detection, and improving interoperability with external systems such as registries, ratings providers, and insurers. Over time, we envision a potentially more autonomous MRV ecosystem in which edge devices not only collect data but also run lightweight checks/parses and generate cryptographic attestations, with humans in the loop primarily for governance and exception handling.
In the future, as we gain more data/knowledge, we can also envision compliance evaluation being more grounded in physical and thermodynamic consistency rather than solely in threshold validation of individual sensor values. Operational intervals that fail such conservation-based or physics-informed consistency checks are excluded irrespective of whether individual sensors appear nominal. In this way, interval eligibility reflects the integrity of the underlying physical material transformation process rather than merely the presence of structured data. This means that FICRA potentially can condition carbon credit eligibility and reporting on compliance with certified operating envelopes (based on physical parameters) rather than on isolated sensor thresholds.
Conclusion
We propose the FICRA framework for decentralised, fleet-based CDR projects using biochar as a case study. The results suggest high-integrity, digitally verifiable CDR is potentially achievable in decentralised fleets operating under real-world constraints. By combining robust edge instrumentation, human-centred data collection tools, and multi-layered validation, we are beginning to show that rural/small-scale actors can support alignment with evidentiary standards of leading registries beyond the artisanal classification. More broadly, the work offers a template for how digital MRV can support scalable/equitable deployment of CDR, helping to ensure that the transition to a carbon-neutral economy does not leave behind the vast majority of the communities that manage much of the world’s biomass resources.
Open peer review
To view the open peer review materials for this article, please visit http://doi.org/10.1017/cat.2026.10013.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/cat.2026.10013.
Acknowledgements
The authors acknowledge the XPRIZE Carbon Removal team for providing this publication opportunity – it has been an honour working with them. In Canada, we acknowledge the Silvachar team (Terry Smith and Kevin Smith) for site provision and field management, members of the Ulkatcho, Esk’meteic, and Lil’wat First Nations, as well as Tommy Krowiak, Nathan Yang, Jayden Li, Matthew Lin, Haroon Moammer, Idara Nkono, Jose Nuez, Jake Oberst, Katherine Palleson, Kaedan Rosman, Samuel Sjoden, and Boris Yip for on-the-ground operational support. In the United States, we acknowledge Carlo Berg and Jonathan Sanchez from Petaluma River Ranch for site provision, and Rod Kux, Tristan Schmid, Garrett Brown, Munira Dhuliyawalla, Justin Agenmonmen, Nicole Winn, and Maximiliam Pedicelli for operational support. In India, we acknowledge Tanmay Telang, Deoyani Sarkhot, Amit Sharma, Rajesh Singh, Lovely Aulakh, and others for on-the-ground operational support. In Kenya, we acknowledge the Africa Climate Ventures (ACV) team for on-the-ground support and strategic thinking. We acknowledge Professor Daniel Sanchez from the University of California, Berkeley, Michael Leitch from XPRIZE Carbon Removal, Robert Höglund from Milkywire, and others for their technical inputs. Finally, the authors acknowledge Melinda Szarics and Shashwat Misra for their technical support.
AI statement
AI tools were used to polish and revise the manuscript for errors.
Author contribution
Conceptualisation: S.G., V.M., J.K., S.R., S.H., T.S., R.H., S.T., Y.C., K.K.; Data curation: S.G., P.G., V.V., J.K.; Formal analysis: S.G., P.G., V.V., V.M., K.K.; Funding acquisition: V.M., J.K., S.R., S.H., S.T., Y.C., K.K.; Investigation: S.G., P.G., V.V., M.I., J.K., S.R., T.S.; Methodology: S.G., J.K., S.R., S.H., T.S., R.H., K.K.; Project administration: S.G., V.M., S.T., Y.C., K.K.; Resources: R.H., S.T., Y.C., K.K.; Software: S.G., P.G., V.V., M.I., T.S.; Supervision: S.G., P.G., V.V., M.I., V.M., J.K., S.R., K.K.; Validation: R.H., S.T., Y.C., K.K.; Visualisation: S.G., P.G., T.S.; Writing – original draft: S.G.; Writing – review/editing: R.H., S.T., Y.C., K.K.
Competing interests
Several authors are affiliated with Takachar Limited, the developer and operator of the Takavator systems described in this study. One author affiliated with Valorize Systems contributed to the development of digital MRV components described in this study. These affiliations represent potential competing interests. All analyses and interpretations presented in this manuscript were conducted in accordance with standard scientific and registry-aligned methodologies and externally reviewed.
Data availability statement
The datasets generated and analysed during this study include operational telemetry, laboratory analyses, and registry-aligned reporting outputs associated with active biochar production facilities. Due to commercial sensitivity and third-party confidentiality agreements, raw operational datasets are not publicly available. Aggregated data supporting the conclusions of this study are available from the corresponding author upon reasonable request.
Ethics and approvals
This study did not involve human subjects research requiring institutional ethics review. Field engagement and system co-design activities were conducted as part of operational deployment processes, and no identifiable personal data are reported in this manuscript.
Funding statement
XPRIZE Carbon Removal and Milkywire Climate Transformation Fund provided general funding to kickstart the FICRA implementation. In Canada, we acknowledge NorthX, NFRF, and NRC for funding. In the United States, we acknowledge Pacific Gas and Electric Company support through EPIC as well as CalSEED. In India, we acknowledge the Seghal Foundation, Ambuja Foundation, Aga Khan Foundation (and its funder GIZ), Innovate UK, Earthshot Prize, and Breakthrough Energy Fellowship for funding.
Inclusivity statement
This research included Takavator deployments on lands traditionally stewarded by Indigenous communities. We acknowledge the historical and ongoing relationships between Indigenous peoples and these territories and recognise their enduring stewardship. Specifically, the authors are grateful for the participation, comments, and feedback from the Ulkatcho, Lil’wat, and Esk’etemc First Nations, which informed aspects of project implementation and system design.
Open access
Comments
Dear Prof. de Arquer and Dinh,
We are pleased to submit our manuscript entitled “FICRA: A Digital MRV Architecture for High-Integrity Decentralised Biochar Carbon Removal” for consideration as a Research Article in the special issue on Innovations in Carbon Dioxide Removal.
This manuscript presents the development and deployment of FICRA (Fleet-Integrated Carbon Removal/Reporting Architecture), a digital monitoring, reporting, and verification (dMRV) framework designed for distributed, small-scale biochar carbon removal systems operating under real-world constraints. The work addresses a central challenge in the evolution of carbon dioxide removal markets: how to maintain high-integrity, auditable verification standards while expanding access to decentralised projects operating in rural and low-infrastructure contexts.
Existing MRV paradigms have largely been optimised for centralised, industrial facilities. In contrast, FICRA demonstrates how edge instrumentation, cloud-based validation, anomaly detection, and human-centred data interfaces can be integrated across a geographically distributed fleet of portable pyrolysis reactors operating in India, Kenya, the United States, and Canada. The manuscript details the system architecture, deployment context, validation logic, and early lessons from implementation. Importantly, FICRA is presented as a shadow-layer implementation architecture operating in parallel with registry-approved methodologies, without introducing new crediting rules. This conservative framing ensures alignment with existing standards while exploring pathways to reduce verification costs and operational burdens.
We believe this work is well aligned with the aims of Cambridge Prisms: Carbon Technologies, particularly the call for contributions on innovations in carbon dioxide removal that integrate technical, governance, and implementation dimensions. The manuscript has a four-fold contribution: (a) a practical architecture for scalable digital MRV in distributed CDR systems; (b) empirical insights from multi-country deployment under variable infrastructure conditions; (c) a governance-relevant discussion of how registries may adapt to fleet-based project definitions; and (d) a framework potentially generalizable to other distributed CDR modalities beyond biochar.
This manuscript has not been published previously and is not under consideration for publication elsewhere. All authors have reviewed and approved the submitted version and agree with its submission to Cambridge Prisms: Carbon Technologies.
We appreciate your consideration and look forward to the opportunity for peer review.
Sincerely,
Kevin S. Kung, Ph.D.
Corresponding Author
Department of Chemical and Biological Engineering
University of British Columbia
kevin.kung@ubc.ca