How the METRICS congress shapes its global reference framework for anticipating, deciding, acting and rebuilding in response to complex risks

Foreword

The METRICS reference framework aims to provide comparable indicators and auditable methods that help businesses, sectors, local authorities and countries answer practical questions:

• What risks are we actually exposed to?
• At what threshold can the situation take a critical turn?
• How much time do we have to make a decision?
• Where are the weakest links in our decision chain?
• What knock-on effects can a crisis trigger?
• How long will it take to restore essential functions?
• Which decisions do the most to reduce our exposure?
• How does our level of preparedness compare with that of other organisations or regions?

Five pillars, one systemic risk

The METRICS reference framework covers five broad families of risk:

• Armed conflicts
• Terrorism
• Health crises
• Financial crises
• Natural disasters

The framework draws on:

• A shared vocabulary for forecasting, crisis management and recovery
• Measurement methods
• Key performance and resilience indicators
• Warning thresholds and tipping points
• Dependency maps
• Scenarios for how disruptions spread and trigger knock-on effects
• Comparisons by sector, region and type of organisation
• Protocols for assessment and decision-making
• Tools for assessing business continuity
• Indicators for a return to normal operations
• A regularly updated barometer
• Recommendations drawn from lessons learned

Organisations face risks that move across physical, digital, health, social and geopolitical domains. A cyber alert may disrupt a hospital, a drought may affect energy and food security, a hostile operation may exploit a disaster, while a health crisis may undermine supply chains and the continuity of government. In this context, sector-specific management creates blind spots. It often detects signals too late, separates forecasting from decision-making and treats post-crisis review as a documentation exercise.

The METRICS reference framework is an integrated risk architecture based on six continuous functions: observe, assess, anticipate, decide, act and learn. It brings together field data, sensors, Earth observation, open sources, digital telemetry, health data and expert knowledge. It turns this information into probabilistic scenarios, decision thresholds and verifiable actions. The framework treats forecasts as estimates. Forecasting becomes a calibrated tool that makes assumptions, confidence and the cost of errors explicit.

The main contribution of the METRICS reference framework lies in connecting four time horizons that are often treated separately: strategic foresight, operational forecasting, risk management during events and post-crisis reconstruction. The model applies to military and hybrid threats, health emergencies, terrorism, cybersecurity, natural disasters, environmental crime and critical infrastructure disruptions. It also defines an evaluation protocol based on probabilistic quality, actionable lead time, decision value, equity, traceability and recovery speed.

Keywords systemic risk, foresight, forecasting, resilience, intelligence, early warning, crisis management, reconstruction, artificial intelligence, space-based observation

1 Introduction

Contemporary risk is less a list of hazards than a network of dependencies. Power infrastructure depends on telecommunications, telecommunications depend on energy, hospitals depend on both, and civilian and military operations depend on shared supply chains. The United Nations Office for Disaster Risk Reduction describes this phenomenon as systemic risk whose effects spread across sectors and territories.^[3] Its 2025 global report estimates that the annual cost of disasters exceeds USD 2.3 trillion when indirect effects and ecosystem damage are included.^[4]

An organisation may have excellent sensors and still remain vulnerable if it cannot turn a signal into a decision. The scientific and operational challenge is therefore to connect three distinct elements. The first is threat knowledge, which describes what may happen. The second is forecasting, which estimates what may happen next and with what confidence. The third is action, which selects a measure despite uncertainty, costs and constraints. Post-crisis review completes this chain by measuring damage, restoring essential functions and correcting models.

How does the METRICS reference framework improve foresight and forecasting without erasing the differences between a conflict, an epidemic, a terrorist attack, a computer intrusion or a natural disaster? The framework relies on a common architecture for data, confidence, scenarios and decisions, supplemented by modules specific to each domain.

2 Method and scope

The study follows a decision-oriented conceptual synthesis method. It brings together risk science, early warning systems, civil resilience, international health, cybersecurity, threat analysis and space-based observation. The documentary foundation combines peer-reviewed research and recognised institutional frameworks. Risk is approached from a perspective that combines events, consequences, uncertainties and the quality of knowledge, in line with recent developments in risk science.^[1]

This approach is neither an experimental trial nor evidence of effectiveness in the field. It provides a testable model. Each component can be validated using archives, exercises, simulations and then controlled deployments. This distinction is essential. A design article may propose mechanisms, indicators and hypotheses, but must not present future performance as an already observed result.

The scope covers intentional, accidental and natural risks. It also includes compound situations, such as a disinformation campaign during an epidemic, a cyberattack against infrastructure during a storm, sabotage concealed by an industrial accident or a military operation that disrupts medicine distribution. The method therefore prioritises dependencies, cascading effects and decisions under constraints.

3 Scientific foundations of the framework

3 1 Risk as a dynamic system

Risk cannot be reduced to the product of a probability and a consequence. This formula remains useful for comparing simple options, but conceals variable data quality, deep uncertainties, interactions and adaptive behaviour. In a complex system, an intervention sometimes changes the threat itself. An adversary changes tactics, a population responds to a warning, a virus evolves, a market redirects its flows and a computer network changes its traffic.

The analysis must represent at least five components: threat, exposure, vulnerability, consequence and response capacity. Resilience does not mean resistance alone. It covers preparedness, shock absorption, continuity of critical functions, recovery and adaptation. The Atlantic Alliance's resilience requirements reflect this continuity between preparedness, resistance, response and rapid recovery, with particular attention to the civilian dependencies that also support military operations.^[8]

3 2 Weak signals and critical transitions

Some systems produce signs before a major transition. An increase in variance, slower recovery after a disturbance, an unusual geographical concentration or a breakdown in correlation may indicate that a regime is approaching a critical threshold. The literature shows that such signals may appear in ecosystems, markets, physiological systems and other complex dynamics, while noting that the exact tipping point remains difficult to predict.^[2]

In the METRICS reference framework, a weak signal is therefore not mysterious or isolated information. It is a measurable deviation from a baseline, whose value depends on its recurrence, consistency with other sources and closeness to a plausible causal mechanism. The proposed system must retain rejected as well as confirmed signals, because false positives and false negatives provide information about model quality.

3 3 Foresight, forecasting, decision-making and review

Foresight explores several plausible futures over a broad horizon. It looks for disruptions, vulnerabilities and options that remain useful across several scenarios. Forecasting estimates an event or state over a defined horizon. It must provide a probability, a time frame, a geographical area, a confidence level and the variables that could invalidate the estimate. Risk management then selects a response. The post-crisis phase measures effects, organises recovery, establishes responsibilities where possible and turns experience into capability.

Table 1 Time horizons of the METRICS reference framework

4 Integrated architecture of the METRICS reference framework

The proposed architecture rests on a continuous chain that preserves the link between the initial signal, assessment, decision and outcome. This traceability would prevent two common failures. The first occurs when a model issues a warning without explaining its basis. The second arises when an organisation acts but does not measure whether the action reduced risk.

4 1 Observe

The observation function would collect heterogeneous data without assuming that any single source holds the truth. It could integrate optical and radar imagery, atmospheric measurements, satellite navigation, local sensors, security logs, epidemiological data, logistics flows, open sources and human reports. Source diversity improves coverage but also increases the risk of contradiction, manipulation and misleading correlation.

Each data item must be accompanied by its provenance, a time stamp, a location where relevant, a sensitivity level and a quality indication. Space-based observation is particularly useful when field access is dangerous, impossible or contested. It enables comparison of a current situation with a historical baseline, including for changes in terrain, fires, floods, ships, industrial sites and ecosystem damage.^[9]

4 2 Assess

Assessment turns a raw stream into knowledge. It looks for duplicates, temporal inconsistencies, sensor faults, manipulation and coverage biases. It assigns a confidence level to both source and information, then distinguishes observed fact, inference and hypothesis. This distinction must remain visible all the way to the decision-maker.

In hostile environments, assessment includes countering deception. Images may be old, digital accounts coordinated, positions falsified and technical logs incomplete. A sensitive warning must therefore rest on sufficient corroboration or trigger additional collection. A lack of confirmation does not always mean an absence of threat, especially when the adversary controls some of the traces.

4 3 Anticipate

Foresight builds scenarios from mechanisms, trends and disruptions. It can use dependency graphs, propagation models, agent-based simulations, scenario analysis and challenge workshops. The aim is not to select a single future. It is to identify decisions that retain their value while several futures remain possible.

A useful scenario describes a trigger, a chain of effects, leading indicators, the actors concerned and the available options. It also specifies what would change the analysis. This discipline reduces hindsight bias and makes revision easier when new information emerges.

4 4 Decide

The decision module would turn a forecast into a choice. Two warnings with the same probability may justify different responses if consequences, the cost of delay, action reversibility or equity differ. The system must therefore display not only probability but also expected losses, options, side effects and escalation rules.

Models can improve some judgements, but the value of prediction depends on its integration into an explicit decision function. Research on combining algorithmic predictions and human decisions shows that evaluation must account for earlier choices that produced the observed data and for outcomes that remain unseen.^[11] In the METRICS reference framework, the final decision must therefore remain attributable, open to challenge and documented.

4 5 Act

A warning has value only if an organisation can act in time. Each threshold must be linked to a measure, a responsible person, a resource, a communication channel and a stopping condition. Measures may include protecting a site, switching a service to an alternative system, increasing surveillance, mobilising medical resources, evacuation, network segmentation or preserving evidence.

Global early warning systems identify four inseparable capabilities: risk knowledge, observation and forecasting, warning dissemination and preparedness to respond.^[5] The METRICS reference framework adopts this approach and adds an operational register tracking implementation, deviations and actual effects.

4 6 Learn

Post-crisis learning begins during the crisis. Decisions, assumptions, threshold changes, unavailable data and unexpected effects must be recorded as action unfolds. After stabilisation, analysis compares forecasts with outcomes, identifies exposed vulnerabilities and measures the time needed to restore essential functions.

Learning from experience produces three concrete changes. It corrects data and models. It changes procedures and responsibilities. It guides investment in resilience. Without this budgetary and organisational link, lessons learned remain a record without effect.

5 Applications in priority domains

The core architecture would remain common, but signals, models and decision rules would vary. The following table summarises the main categories of information. The subsequent sections specify uses and limitations.

Table 2 Signals and uses by domain

5 1 Defence, security and hybrid threats

In the military domain, foresight must combine likely intentions, observed capabilities, logistics, terrain, civilian infrastructure and the information environment. A concentration of assets does not have the same meaning depending on fuel availability, routes, weather, doctrine and political constraints. The model must therefore separate capability, intent and opportunity.

Hybrid threats exploit the interfaces between peace, crisis and conflict. A disinformation campaign may prepare sabotage, amplify a shortage or reduce acceptance of a health measure. The METRICS reference framework uses a graph linking narratives, physical events, digital disruptions and logistics movements. The aim is not to automate political attribution, but to indicate corroborating evidence, alternatives and the strength of evidence.

After a crisis, priorities focus on continuity of authority, restoring essential services, repairing infrastructure, preserving evidence and assessing residual capacity. Civil resilience frameworks emphasise continuity of government, services to the population and civilian support to military operations.^[8]

5 2 Health crises

Health warnings benefit from combining data with different time lags and biases. Consultations provide a clinical picture, laboratories confirm with a delay, wastewater may detect circulation before hospital admissions, mobility indicates potential spread and absenteeism sometimes signals a wider disruption. Careful fusion reduces dependence on a single indicator.

Forecasting must provide trajectories rather than a single curve. It estimates admissions, critical care requirements, stocks, available personnel and the effects of possible measures. Biological and behavioural assumptions are explicit. Confidence intervals widen with the forecast horizon. The International Health Regulations amended in 2024 strengthen the global preparedness and response framework while retaining States' responsibility for national measures.^[7]

After a crisis, the METRICS reference framework helps structure the analysis of excess mortality, delayed care, medicine shortages, staff fatigue and territorial inequalities. Recovery is not limited to a decline in cases. It includes restoring the health system, monitoring delayed effects and preparing for subsequent waves.

5 3 Terrorism and organised violence

Terrorist risk requires particular caution because events are rare, data are incomplete and the consequences of misclassification can be serious. The proposed system must prioritise analysis of relevant behaviours and networks rather than identity categories. It must also distinguish threat indicators, target vulnerabilities and operational feasibility.

Financial data, public communications, travel, purchases and site reconnaissance can contribute to an assessment, but none of these traces is sufficient on its own. Methods for mapping terrorist financing recommend multiple quantitative and qualitative sources from competent services, the private sector, statistics and open information.^[12]

Action must remain proportionate and subject to oversight. The METRICS reference framework helps prioritise a verification, temporarily strengthen site protection or recommend human review. It must not turn an opaque score into evidence. After an event, the proposed platform could support timeline reconstruction, network analysis, victim protection, service continuity and identification of signals that were missed or misinterpreted.

5 4 Cybersecurity

Cybersecurity generates large volumes of data, but the noise is considerable. Effective detection combines identities, assets, vulnerabilities, events, dependencies and business context. Unusual behaviour becomes a priority when it concerns a critical asset, a privileged identity or a coherent attack chain.

The NIST cybersecurity framework organises management around governance, identification, protection, detection, response and recovery.^[6] The METRICS reference framework connects these functions with other domains. A computer compromise could thus be assessed in terms of its potential effects on a hospital, a transport chain, a security operation or a public warning.

After an incident, the proposed platform would retain the timeline, containment decisions, forensic images, evidence and affected dependencies. It would track restoration in order of criticality and verify that corrected controls actually reduce exposure. Recovery must include secret rotation, backup validation, enhanced monitoring and notification of the parties concerned.

5 5 Natural disasters and climate change

Natural disasters result from the interaction between a hazard, exposure and vulnerability. The same flood has different effects depending on urbanisation, building quality, transport access, the time of day and evacuation capacity. Climate projections indicate that risks increase with warming and that some changes may become abrupt or irreversible.^[10]

The METRICS reference framework sets out how to combine meteorological, hydrological, seismic and space-based observations with population, infrastructure and vulnerability data. Forecasting would become impact forecasting. It would not merely indicate where water will rise or fire will spread. It would estimate which roads, hospitals, networks and populations are likely to be affected.

After the event, comparative imagery and field reports accelerate damage assessment. The proposed system would prioritise access, search operations, care, power restoration and re-establishing communications. Reconstruction must reduce future vulnerability; otherwise, it reproduces the risk that has just materialised.

5 6 Environmental crime and industrial risks

Environmental crime lends itself to independent monitoring because declarations can be compared with observable changes. Mine expansion, deforestation, a thermal discharge, a plume, a waste dump, an oil slick or a ship's activity may leave spatial and temporal traces. Radar data remain useful at night and through clouds, while optical and spectral data characterise land cover, vegetation or certain atmospheric markers.^[9]

Operational value comes from fusion. An isolated image shows an anomaly, but combining it with permits, routes, declared volumes, weather data, local sensors and navigation can establish a stronger hypothesis. In the maritime domain, fusion of radar and cooperative tracking systems helps identify ships that deliberately stop transmitting.

After an incident, the METRICS reference framework helps delineate the affected area, estimate the drift of pollution, document the timeline and support remediation. Evidence must retain its provenance and integrity if intended for administrative or judicial scrutiny.

5 7 Critical infrastructure and supply chains

Critical infrastructure forms the shared fabric of crises. An energy outage can interrupt telecommunications, transport, water, payments and healthcare. The model must represent technical, contractual, geographical and human dependencies. It must also identify scarce resources that determine recovery, such as transformers, components, specialist teams or alternative routes.

Forecasting here relies on graphs and propagation simulations. A failure is injected into the model to estimate affected services, likely duration and workaround options. Exercises enable comparison of these results with operators' actual responses. After a crisis, the priority is the sequenced restoration of functions that reduce human and economic consequences most quickly.

6 Analytical methods

6 1 Data fusion and knowledge graphs

A knowledge graph connects authorised individuals, organisations, assets, places, events, vulnerabilities and sources. It enables a relationship to be traced back to its evidence and prevents people with the same name or poorly identified objects from being merged. Quality rules must preserve contradictions rather than erase them.

Fusion can combine probabilistic and symbolic methods. Probabilistic models estimate the likelihood of a state from imperfect data. Symbolic rules apply known constraints, such as a temporal impossibility or geographical incompatibility. Human expertise intervenes in ambiguous cases, new modes of action and sensitive consequences.

6 2 Probabilistic forecasting

An actionable forecast must be probabilistic and calibrated. If one hundred events are assigned a probability of sixty per cent, about sixty should occur over a sufficiently long period and under comparable conditions. Calibration protects against overconfidence. Discrimination measures the ability to distinguish high-risk and low-risk situations. Both qualities are necessary.

Possible methods include time series, Bayesian models, trees, neural networks, survival models, graphs and simulations. The choice depends on volume, mechanism, event frequency and the need for explanation. Rare events require appropriate validation, controlled synthetic scenarios and close attention to changes in data over time.

6 3 Simulation and operational twins

An operational twin represents essential functions, resources, dependencies and constraints. It does not seek to reproduce every detail of the world. It enables decisions to be tested, such as closing a road, losing a data centre, a mass influx of patients or the unavailability of a team. Results are used to prepare plans and choose redundancies.

Simulations become dangerous when confused with predictions. They explore the consequences of rules and assumptions. Their value depends on parameter transparency, sensitivity analysis and comparison with observed events or exercises.

6 4 Human intelligence and artificial intelligence

Artificial intelligence can speed up sorting, detect patterns, summarise large volumes and suggest scenarios. It does not automatically possess the institutional context, legitimacy or accountability needed for a sensitive decision. The METRICS reference framework must specify which decisions the system may recommend, which it may execute under supervision and which require human authority.

The proposed system must also protect human judgement against passive automation. The decision-maker should see the main reasons, sources, uncertainties, alternatives and consequences of an error. For decisions with serious consequences, an adversarial review or dual approval reduces the risk of unjustified confidence in a score.

7 Decision-making under uncertainty

Decision-making under uncertainty compares the cost of acting too early with the cost of acting too late. A low threshold increases warnings and false alarms. A high threshold reduces noise but may miss a serious event. The appropriate threshold varies with severity, reversibility, action lead time and the people affected. It must be defined before the crisis, then adjusted according to known rules.

In the METRICS reference framework, a decision matrix links each confidence level to a set of proportionate measures. Preliminary information may trigger enhanced collection. Moderate corroboration may justify discreet pre-positioning. Strong corroboration and major consequences may activate a full response. This progression preserves the ability to act without presenting a hypothesis as fact.

Decisions must be revisable. Each action is assigned a duration, a responsible person, criteria for continuation and stopping criteria. Reversibility protects against responses becoming permanent after the risk has changed. Measures affecting freedoms, health or access to a service require stronger justification and oversight.

8 Post-crisis actions

8 1 Rapid assessment and continuity

In the first few hours, the METRICS reference framework helps build a shared picture of damage, affected people, interrupted services and available access routes. Space-based, aerial, local and digital sources would be compared. Priority would go to functions whose interruption causes the most serious consequences or blocks other sectors' response.

Continuity relies on service objectives, alternative solutions and resources identified in advance. The proposed system would track the switch to a backup site, restoration of a backup, delivery of stock, redeployment of a team and the gradual return to normal operations.

8 2 Evidence, attribution and accountability

Intentional crises and some accidents require a chain of evidence. Data must retain their source, date, integrity, transformations and the identities of those who accessed them. Attribution must distinguish technical certainty, intelligence assessment and political or judicial decision. The METRICS reference framework supports this discipline without claiming to replace the competent authorities.

8 3 Reconstruction and reduction of future risk

Reconstruction must correct the vulnerabilities revealed. Rebuilding exposed infrastructure identically restores service but preserves risk. Reconstruction choices compare speed, cost, sustainability, redundancy and territorial equity. The global risk report emphasises that investments incorporating resilience can prevent a substantial share of future losses.^[4]

8 4 Verifiable learning from experience

Learning from experience compares the expected and actual timelines. It examines the signals available, those that were seen, those that were ignored, the decisions taken, delays and effects. Its purpose is not merely to identify fault. It measures the conditions that made error likely, such as inaccessible data, ambiguous responsibility, a poorly defined threshold or insufficient capacity.

9 Scientific evaluation and indicators

A risk system must not be evaluated solely by its detection rate. A model that warns about everything detects all events but becomes unusable. Evaluation must cover probability quality, actionable lead time, decision value, impact on people, robustness and recovery.

Table 3 Proposed performance indicators

Validation would follow four levels. The first uses archives and withholds some data to test forecasting. The second replays crises in exercises. The third deploys the system in silent observation without influencing decisions. The fourth measures its effect under controlled operational conditions. This progression reduces the risk of introducing an unevaluated tool into a critical function.

Thresholds and models must be monitored after deployment. Declining calibration, behavioural change, a new sensor or a regulatory development may make a model less reliable. Performance is therefore a property monitored over time, rather than a certification obtained once and for all.

10 Governance, security and ethics

10 1 Governance of responsibilities

The METRICS reference framework must specify who owns the data, who validates the model, who sets thresholds, who receives warnings and who authorises action. Roles vary by domain, but the chain of accountability must remain explicit. A common risk committee could arbitrate cross-sector dependencies and maintain a coherent doctrine.

10 2 Data security and sovereignty

Risk data may expose the vulnerabilities of an organisation or territory. The proposed architecture would apply minimisation, access control, encryption, separation of environments, logging and retention rules. Sensitive models would themselves be protected, because their structure may reveal monitoring criteria or facilitate evasion.

10 3 Rights, equity and proportionality

Systems applied to health, security or terrorism may affect people who pose no threat. Indirect data may reproduce historical inequalities. Governance must prohibit unjustified variables, measure performance gaps, provide recourse where possible and subject sensitive uses to human validation.

Proportionality requires the intensity of the response to match the evidence and consequences. A low score may justify further observation, but not an intrusive measure. This rule protects people and also improves operational quality by preventing resources from being consumed by uninformative warnings.

11 Roadmap for the METRICS reference framework

Implementation could begin with a limited scope that demonstrates the complete chain from signal to decision to outcome. The first case must have accessible data, a frequently recurring decision, an identified responsible person and a way to measure the effect. A recurring natural disaster, a cyber incident affecting a defined service or a logistics disruption often provide suitable starting conditions.

Table 4 Implementation pathway

12 Current boundaries of the METRICS reference framework

The proposed framework offers a common language across domains without claiming that their models are interchangeable. The same architecture could preserve evidence, assess a source, produce a scenario and record a decision. Biological, military, hydrological, criminal and digital mechanisms, however, require their own data and validation.

The main difficulty lies in data governance, responsibilities, trust between partners, the ability to act and measurement discipline. A highly accurate system remains useless if a warning arrives after the window for action, if no one has the means to respond or if teams do not understand why the threshold was crossed.

The second limitation concerns adversaries and changing environments. Hostile actors learn detection criteria and adapt their behaviour. Health and climate crises also alter observed relationships. The METRICS reference framework must account for data change, deception, uncertainty and the possibility that the model itself becomes a target.

The third limitation is ethical. The more domains the proposed platform connects, the more intrusive it could become. The value of an integrated view does not justify unlimited collection. Purpose, minimisation, controlled access and proportionality must form part of the technical design and evaluation criteria.

13 Conclusion

The METRICS reference framework connects foresight, forecasting, risk management and post-crisis action in a single architecture for decision-making and resilience. Its value comes from a chain of evidence that shortens the time from a credible signal to a proportionate response and then measures whether the response had the intended effect.

The framework addresses complex risks using data from multiple sources, calibrated models, human expertise, explicit thresholds and verifiable feedback. It supports work on defence, health, terrorism, cybersecurity, natural disasters, environmental damage and critical infrastructure across organisational boundaries.

The next step is to choose an operational case, identify the decision to improve, set a measurable baseline and run a phased pilot. The credibility of the METRICS reference framework rests on its ability to demonstrate better foresight, faster decisions, safer action and shorter recovery times.

References

1 Aven T. Risk assessment and risk management. Review of recent advances on their foundation. European Journal of Operational Research. Volume 253. 2016. Pages 1 to 13. DOI 10.1016 j.ejor.2015.12.023.

2 Scheffer M. Bascompte J. Brock W. et al. Early warning signals for critical transitions. Nature. Volume 461. 2009. Pages 53 to 59. DOI 10.1038 nature08227.

3 United Nations Office for Disaster Risk Reduction. Our World at Risk. Transforming Governance for a Resilient Future. Global Assessment Report. 2022.

4 United Nations Office for Disaster Risk Reduction. Resilience Pays. Investing and Financing for Our Future. Global Assessment Report. 2025.

5 United Nations Office for Disaster Risk Reduction. Early Warnings for All. Executive plan and global early warning chain. 2023 to 2027.

6 National Institute of Standards and Technology. The NIST Cybersecurity Framework 2.0. NIST CSWP 29. 2024. DOI 10.6028 NIST.CSWP.29.

7 World Health Organization. International Health Regulations of 2005, amended in 2014, 2022 and 2024. Version in force since September 2025.

8 North Atlantic Treaty Organization. 2026 Resilience Baseline Requirements. 2026.

9 European Space Agency. Environmental Crimes Use Cases. Reference ESA CICA USECASE 2026 3422. 2026.

10 Intergovernmental Panel on Climate Change. Climate Change 2023. Synthesis Report. Sixth Assessment Report. 2023.

11 Kleinberg J. Lakkaraju H. Leskovec J. Ludwig J. Mullainathan S. Human Decisions and Machine Predictions. Quarterly Journal of Economics. Volume 133. 2018. Pages 237 to 293. DOI 10.1093 qje qjx032.

12 United Nations Office on Drugs and Crime. Guidance Manual for Terrorist Financing Risk Assessments. 2025.

13 International Organization for Standardization. ISO 31000 Risk management Guidelines. 2018.