Civilizational Systems Outlook: 2026–2035

1Executive summary

This report is the first edition of the Civilizational Systems Outlook, the Department’s annual model-based assessment of systemic stress across critical global systems over a ten-year horizon. It was prepared by the Office of Strategic Risk Assessment (OSRA) under the Integrated Systems Modeling Program, with contributions from the Office of Advanced Artificial Intelligence (OAAI), the Office of Long-Term Systems Research (OLSR) and the Continuity Infrastructure Administration (CIA-DSC). The Outlook series was directed by the 2026 Research Priorities Statement¹ and is intended to give the Department, its advisory bodies and the public a consistent, documented reference point for long-horizon planning.

The assessment uses version 4.2 of the Integrated Civilizational Systems Model (ICSM), a coupled simulation of eleven interacting modules covering energy, food and agriculture, water, climate and environment, finance and trade, supply chains and manufacturing, information and communications, computational systems and artificial intelligence, public health, governance and social cohesion, and demography. The model was run as a 40,000-member ensemble across 1,280 scenario configurations on the ALPHA-1 system at Continuity Research Facility Alpha. Results are summarized through the Composite Systemic Stress Index (CSSI), a 0–100 measure of concurrent stress across the modeled systems.

Central finding. In 71 percent of weighted ensemble runs (90 percent interval 58–82 percent), the CSSI enters the systemic breakdown regime, defined as a CSSI of 70 or above sustained for 24 consecutive months, before 1 January 2035. The median entry year across the ensemble is 2033. The Office assesses that entry into the breakdown regime before 2035 is likely, with moderate confidence².

The breakdown regime is the model’s operational counterpart to what this report calls severe civilizational breakdown: a sustained, multi-region loss of capacity to maintain critical infrastructure, food distribution and governance at 2020 levels. It does not denote human extinction, the simultaneous failure of every system in every region, or an outcome from which recovery is excluded³.

Confidence and probability language

This report distinguishes between the probability of an outcome, expressed as the weighted fraction of ensemble runs in which it occurs, and the Office’s confidence in that probability, which reflects the quality of the evidence, the degree of validation and the extent of agreement among model variants and reviewers. Probability terms are used only with the meanings in Table 1. Confidence is stated as high, moderate or low. A finding stated as “likely, with moderate confidence” therefore means that the outcome occurs in between 66 and 90 percent of weighted runs, and that the Office regards the model’s representation of the relevant processes as adequate for the purpose but subject to material structural uncertainty.

Readers should keep two limits in view. First, the ICSM represents systems at the level of 24 world regions and monthly coupling intervals; it does not forecast individual events, and it cannot anticipate initiating shocks whose character is outside its scenario space. Second, the probability that the model assigns to an outcome is not the probability that the outcome will occur. The gap between the two is discussed in the section on uncertainty and limitations.

2Research background

The Department’s systemic risk work descends from the Interagency Task Force on Civilizational Continuity (ITFCC), convened in March 2024. The Task Force’s final report, Report on Long-Horizon Systemic Risk (ITFCC-2025-07, November 2025), concluded that federal risk assessment was organized by sector and by hazard, and that no standing capability existed to assess how stresses in one system propagate into others over periods longer than a budget cycle⁴. The Department was established on 12 January 2026 in part to provide that capability. OSRA assumed responsibility for the Task Force’s prototype model, which became ICSM version 4.0 when it was baselined in April 2026.

Two strands of earlier work inform the Outlook’s approach. The first is the study of interdependent infrastructure: failures in tightly coupled systems, in which processes are time-dependent and slack is limited, can propagate faster than operators can diagnose them, and failures in networks that depend on one another can cascade in ways not visible when each network is examined alone. The second is the tradition of global systems modeling, which treats population, resource use, production and environmental capacity as a coupled system and examines how its trajectory responds to different assumptions. The Office draws on this literature for general orientation only; the background works listed at the end of this report do not address, and should not be read as supporting, the specific findings presented here.

The Outlook builds directly on two Departmental reports. Integrated Civilizational Systems Model (ICSM) Version 4: Model Description and Validation (DSC-TR-2026-003) documents the model architecture, the definition of the CSSI and the initial hindcast program. Long-Term Systems Assessment: Interdependencies Among Critical Global Systems (DSC-TR-2026-010) mapped the dependency structure among critical systems and identified the coupling pathways that this report quantifies⁵. Methods for planning under deep uncertainty, which shape the ensemble design and the treatment of scenario weights, are set out in DSC-WP-2026-012.

Purpose and scope of the Outlook

The Outlook has three purposes: to estimate the current level of systemic stress; to characterize the distribution of trajectories that the model considers consistent with present conditions and documented assumptions over 2026–2035; and to identify the conditions under which adverse trajectories are avoided, so that research and investment can be directed toward them. The ten-year horizon was chosen because it is long enough for slow variables such as institutional capacity, demographic structure and infrastructure stock to change materially, and short enough that the model’s structure can still be tested against comparable historical periods.

The Outlook is global in scope and reports results by region type rather than by jurisdiction. It does not assess any single country, does not address national-security contingencies, and does not evaluate specific policies of the United States or any other government. Its recommendations are addressed to the Department’s own research program and to the federal science community.

Interpretation of the results requires attention to what kind of statement they are. The ICSM does not extrapolate a trend. It simulates the interaction of eleven systems under sampled conditions, and reports how often, and through which pathways, those interactions produce sustained, high, concurrent stress. Where this report states that an outcome occurs in a given fraction of runs, the statement is about the behavior of the model under its assumptions. The Office’s judgment about how far that behavior corresponds to the world is expressed separately, through the confidence language described above.

3Methodology

This section summarizes the model, the stress index, the ensemble design, calibration and validation. Full documentation of the version 4 architecture is given in DSC-TR-2026-003⁶; this section records the changes made for version 4.2 and the procedures specific to the Outlook.

4Modeling assumptions

Every run of the ICSM rests on assumptions about quantities that the model does not determine for itself or that cannot be estimated precisely from data. Table 4 lists those to which the headline result is most sensitive. Central values are the medians of the sampling distributions; ranges are the bounds within which values were sampled. Ranges were set by structured elicitation, informed by published sources and by the calibration described above, and were fixed before the production ensemble was run¹⁴.

Three structural assumptions are not captured in Table 4 but shape the results. First, the model assumes that the 24 regions retain the broad political and economic relationships of 2026, modified only within the five levels of the trade and geopolitical regime; it does not represent the dissolution of states or large-scale armed conflict between major powers, which are treated as outside the scenario space. Second, technological change outside the Computational Systems and AI module follows exogenous trajectories and cannot accelerate in response to stress. Third, the model contains no deliberate coordinated global response beyond the adaptive behavior calibrated into the Governance module. The first assumption likely biases results toward lower stress; the second and third likely bias them toward higher stress. The Office has not attempted to net these effects.

Assumption ranges are deliberately wide. The purpose of the ensemble is not to find the most probable future but to determine how much of the plausible range of conditions leads to the breakdown regime, and which conditions do not. Results that hold across wide ranges are more robust than results that depend on a narrow choice; where a result depends strongly on a single assumption, that dependence is reported.

5Infrastructure dependencies

Critical infrastructure is represented in the ICSM as a network of capacities with dependencies between them. The dependency mapping in DSC-TR-2026-010 found that the structure of these dependencies has changed in two ways over the past two decades. The number of dependencies has increased, chiefly through the extension of digital communications and automated control into energy, water, transport and finance. And the buffers between systems, such as fuel stocks, spare equipment, manual procedures and staff trained to use them, have diminished in most regions as efficiency gains were realized by removing slack.

Table 5 summarizes the principal dependency classes as represented in ICSM v4.2. The coupling coefficient expresses the fraction of a supplying system’s capacity loss that is transmitted to the dependent system after its buffer is exhausted; the buffer duration is the period for which the dependent system can operate without the supplying system; the recovery time constant is the characteristic time for restoration once supply returns.

Three features of the dependency structure account for most of its contribution to the headline result. The first is the electricity–communications–control loop: electricity supports communications, communications supports the control of electricity, and the loop has become tighter as control has been automated. The second is the long-recovery tail: some dependencies, notably those on specialized grid equipment and semiconductors, combine short buffers with recovery times measured in months or years, so that a disruption of a few weeks can leave a capacity deficit lasting much longer. The third is the role of governance capacity in restoration: almost every recovery process in the model draws on coordination, financing and authorization, so that erosion of institutional capacity lengthens recovery everywhere at once.

A useful summary statistic is the share of critical dependency paths, weighted by consequence, for which no non-automated fallback is represented. In the model’s 2026 nowcast this share is 41 percent globally, compared with an estimated 18 percent in 2010. In the ensemble median it rises to 57 percent by 2035. Runs in which it remains below 45 percent enter the breakdown regime before 2035 at roughly half the rate of the ensemble as a whole.

6AI-related systemic risks

The Computational Systems and AI module represents the role of advanced computational systems in operating other systems. It does not attempt to model the capabilities of particular systems, nor does it represent scenarios in which such systems act with independent objectives. Its concern is narrower and, in the Office’s judgment, better supported by evidence: the progressive transfer of control functions in critical infrastructure to automated systems, and the ways in which that transfer changes the failure behavior of the infrastructure as a whole.

Automation of control functions

In the model’s 2026 nowcast, automated systems perform a large share of routine decisions in grid dispatch, freight scheduling, payments and network management, and the model assumes that the share continues to grow in all configurations, at rates that differ by configuration family. Table 6 gives the assumed automation shares. Automation in this sense means that the decision is executed without real-time human confirmation; it does not imply that humans are absent from oversight¹⁵.

Automation reduces the frequency of small failures. Automated dispatch responds faster than human operators to routine fluctuations, and automated logistics scheduling reduces idle capacity. The model represents these benefits, and in most runs they lower stress in the first years of the projection. The difficulty arises in the tail. Automated systems that are reliable within the range of their training and testing conditions may behave unpredictably outside it, and the conditions under which they are most needed, large disturbances, are by definition rare in that experience. At the same time, the human capacity to take over, in the form of trained staff, manual procedures and the equipment to execute them, diminishes as it is used less.

Correlated failure modes

The most consequential property represented in the module is correlation. When many operators in many regions deploy systems derived from a small number of underlying models, built by a small number of vendors on overlapping data, the failures of those systems are no longer independent. A condition that causes one deployment to fail is more likely to cause others to fail at the same time, including in regions and sectors that are otherwise unconnected. The module represents this through a common-mode coupling term, parameterized from the correlated-failure dimension of the OAAI evaluation framework¹⁶, which introduces simultaneous degradation across deployments that share an underlying system.

Correlated failure converts geographic diversity, traditionally one of the most effective defenses against systemic failure, into a weaker defense. In runs with high common-mode exposure, the probability that a stress originating in one region produces simultaneous control degradation in three or more regions is approximately four times that in runs with low exposure. This is the principal mechanism through which the automation trajectory affects the headline result.

Concentration of compute

The module also tracks the concentration of the computational capacity on which automated control depends. The model assumes that the share of large-scale training and inference capacity held by the five largest operators remains between 70 and 90 percent through 2035, and that this capacity is itself concentrated in a limited number of locations with high electricity demand¹⁷. Concentration creates a second correlation channel: disruption to a small number of facilities, whether through electricity shortfall, cooling failure, supply-chain interruption or other cause, degrades many dependent services at once. It also couples the Computational Systems and AI module tightly to the Energy module, since growth in compute demand is among the principal sources of new electricity load in the model.

Evaluation gaps

The model’s representation of these risks is constrained by the evidence available. Systematic evaluation of advanced systems in control roles is recent, and coverage is incomplete. In the four sectors in initial scope of the OAAI framework, the Office estimates that systems representing less than a quarter of consequence-weighted automated control decisions had undergone evaluation at the depth that the framework associates with their deployment role by mid-2026. Where evaluation evidence is absent, the module uses wide priors on failure behavior; this uncertainty propagates to the headline result and is one reason that the automation trajectory has a large total-order sensitivity index. Improved evaluation coverage is, in the Office’s judgment, among the most cost-effective means of reducing uncertainty in future editions.

7Energy and resource vulnerabilities

The Energy module occupies a central position in the model’s dependency network: in DSC-TR-2026-010, electricity was the supplying system for more material dependencies than any other. Three developments in the projection period increase stress on it in most configurations. Demand is growing after a long period of near-stability in many advanced-economy regions, driven by electrification of transport and heating and by computational load. Weather variability is increasing the frequency of conditions under which both demand peaks and supply is derated. And the equipment required to expand and repair transmission and distribution networks is supplied by concentrated industries with long lead times.

The model does not assume that the transition from fossil fuels is itself a source of stress; in most configurations, additional renewable generation and storage improve resilience by diversifying supply. Stress arises where the pace of demand growth exceeds the pace of network expansion, and where the materials and equipment required for expansion are constrained by the supply-chain concentration discussed in the next section. Configurations in which transmission expansion keeps pace with demand show markedly lower Energy sub-index values in 2035, and correspondingly lower probabilities of entry into the breakdown regime. The Department’s own facilities are included in regional load data at their contracted values; their contribution to any regional stress metric is below the reporting threshold¹⁸.

Beyond electricity, the model represents three resource constraints with material influence on the result. Phosphate and potash fertilizer supply is concentrated and links the Food and Agriculture module to the Finance and Trade module through price. Groundwater depletion in several major irrigated production regions reduces the buffering capacity of agriculture against drought. And liquid-fuel supply remains the principal energy input to freight transport throughout the projection period, so that fuel disruptions propagate rapidly to food distribution. None of these constraints is sufficient on its own to produce breakdown-regime conditions in the model; each lengthens the recovery from shocks that originate elsewhere.

8Global supply chain instability

Global supply chains are represented in the Supply Chain and Manufacturing module as flows of 46 product classes between the 24 regions, with production capacities, inventories and substitution options. The module was extended for v4.2 with a sub-model of semiconductor manufacturing and critical-mineral processing, because these supply chains combine three properties that the model associates with systemic consequence: high concentration, long substitution times and a position upstream of other critical systems.

Concentration, measured where possible by the Herfindahl–Hirschman index¹⁹, matters in the model because it couples otherwise independent shocks. A disruption in a region that hosts a large share of a critical input, whether caused by natural hazard, infrastructure failure, a change in trade regime or a public-health emergency, propagates to every region that depends on that input. Where inventories are thin and substitution is slow, the effect persists after the original disruption has ended. In the ensemble, a disruption removing 30 percent of leading-edge semiconductor output for six months reduces global compute and communications replacement capacity for a median of 26 months, and raises the CSSI by a median of 4.1 points at its peak.

Inventory depth is among the most influential supply-chain parameters. The long-term reduction of inventories through just-in-time practices lowered costs during stable periods; in the model it also shortened the buffer between disruption and shortage. Runs in the upper third of the sampled inventory-cover range enter the breakdown regime before 2035 at a rate 11 percentage points lower than runs in the lower third, other conditions being equal. Maritime chokepoints and their associated insurance and freight markets add a further channel, through which regional disruptions affect trade costs globally within weeks.

The model treats the trade and geopolitical regime as an external driver with five levels, ranging from continued integration to fragmentation into competing trade blocs. Fragmentation raises stress in the model during the transition, as supply chains are duplicated and costs rise, and lowers some concentration risks thereafter. In the ensemble, the net effect of fragmentation on the probability of entry before 2035 is adverse, because the transition period coincides with the years in which other stresses accumulate.

9Environmental risks

The ICSM does not simulate the physical climate system. The Climate and Environment module takes the frequency, intensity and spatial correlation of climate-related hazards as boundary conditions, conditioned on four warming pathways drawn from the range of published scenarios, and translates those hazards into effects on other modules²⁰. The report therefore makes no independent finding about climate science; its findings concern how climate-related hazards, as characterized in the published literature, interact with the other systems represented in the model.

In the model, climate acts principally as a stress multiplier. It rarely initiates the cascades that lead to the breakdown regime. Instead it raises the baseline stress of food, water, energy and health systems, increases the probability that shocks in different regions coincide, and lengthens recovery. Table 9 lists the hazard channels represented.

The most consequential channel in the model is the increased correlation of agricultural shocks across producing regions. Global food supply has historically been buffered by the fact that poor harvests in one region tend to coincide with adequate harvests elsewhere. Where hazards become more spatially correlated, the probability of simultaneous shortfalls rises; such shortfalls in turn interact with fertilizer prices, trade restrictions and financial conditions. In the model, synchronized food-system shocks are among the most frequent proximate triggers of the transition from elevated stress to the breakdown regime.

Environmental pressures other than climate are represented more coarsely. Soil degradation, biodiversity loss and the condition of fisheries enter as slowly varying modifiers of food-system productivity; they have little effect within the ten-year horizon of this edition, but would be expected to matter more over longer horizons. The model does not represent tipping points in the physical climate system or ecosystems; the Office notes this as a limitation in the section on uncertainty, since such changes, if they occurred within the horizon, would not be reflected in the results.

10Projected outcomes

This section presents the projected trajectory of the CSSI, the probability and timing of entry into the systemic breakdown regime, the attribution of the result to drivers, the pathways by which runs enter the regime, and differentiation across types of regional system.

Trajectory of the Composite Systemic Stress Index

The ensemble median rises at an average of about 2.2 index points per year over 2026–2035, compared with about 0.75 points per year over 2000–2020 and about 1.3 points per year over 2020–2026. The spread of the ensemble widens markedly after 2029, reflecting the divergence between runs in which cascading interactions take hold and runs in which they are contained. The distribution in 2035 is asymmetric. The 5th percentile lies 16 points below the median and the 95th percentile 13 points above it: a minority of runs stabilize at levels only a few points above the 2026 value, while most continue to rise. Runs in the upper tail rise more slowly after entering the breakdown regime than before it, because several module sub-indices approach the top of their scales.

The model’s projected increase does not arise from any single assumed trend. In a diagnostic set of 2,000 runs in which all external drivers were held at their 2026 values, the CSSI still rose to a median of 61 by 2035, driven by the continued tightening of couplings already present in the 2026 state and by the slow erosion of buffers. External drivers account for the remainder of the projected increase.

Probability and timing of entry into the breakdown regime

Table 10 gives the cumulative weighted fraction of ensemble runs that have entered the systemic breakdown regime by the end of each year. The headline figure, 71 percent before 1 January 2035, is the final row. The cumulative fraction first exceeds 50 percent during 2033, which is the median entry year²¹. The 90 percent intervals reflect uncertainty arising from the choice of scenario weights and from sampling across configurations; they are estimated by resampling configurations and by varying the weighting scheme within the range considered defensible by the elicitation panel.

The entry curve is steepest in 2032–2034, when 50 percent of all runs enter. This concentration is a property of the model’s dynamics rather than of any scheduled event: in the median run, the slow erosion of buffers and institutional capacity through the late 2020s brings the system to a state in which an ordinary-scale shock, of a kind that occurs several times in a decade, is sufficient to begin a cascade that the model does not resolve within 24 months. In the model, the timing of entry within a run is determined largely by when such a shock occurs; the probability that entry occurs at all within the horizon is determined largely by the state the system has reached by then.

The 29 percent of runs that do not enter the regime before 2035 are not uniformly benign. Approximately 40 percent of them reach a monthly CSSI of 70 at some point without meeting the persistence criterion, and the median CSSI of non-entering runs in 2035 is 63. The breakdown-regime threshold is a defined boundary on a continuous distribution of outcomes, and runs on either side of it near the end of the period differ less than the binary classification suggests.

Attribution to drivers

Table 11 attributes the variance in entry timing across the ensemble to the main drivers, using variance-based sensitivity indices²². The first-order index measures the share of variance attributable to a driver acting alone; the total-order index includes its interactions with all other drivers. The first-order indices sum to 0.66, so that about a third of the variance arises only from interactions. This is the quantitative expression of the report’s central methodological point: the risk is systemic, in the sense that it is a property of the coupled system and not of any of its parts.

The climate hazard pathway illustrates the distinction. Its first-order index is the second smallest in the table, but its total-order index equals that of supply-chain concentration, because its effects operate almost entirely through other systems: correlated harvest failures, derated generation, heat-related health burdens. The institutional-capacity driver shows the opposite pattern to a lesser degree; it has a large first-order effect and the largest total-order effect, because institutional capacity governs recovery in every module.

Pathways into the breakdown regime

Runs that enter the breakdown regime were classified by the sequence of module stresses in the 36 months preceding entry, using scenario-discovery methods²³. Four pathway clusters account for all entering runs. Table 12 describes them. The clusters are not mutually exclusive in mechanism; they are distinguished by which systems carry stress first.

The largest cluster, P4, is the least dramatic in character and in some respects the most significant. In these runs no event in the projection period would, by itself, be recognized as a crisis of unusual scale. The system enters the breakdown regime because recovery from each disturbance is incomplete when the next begins, and because the capacity to coordinate recovery declines over the sequence. Runs in this cluster are also those in which the model’s early-warning indicators, discussed under recommendations, give the longest lead time.

In every run that entered the breakdown regime, at least four of the 24 modeled regions, together containing more than one-fifth of world population, fell below their 2020 levels of capacity in at least two of the three functions named in the definition of severe civilizational breakdown. In the median entering run, eleven regions did so within the first 24 months of the regime.

Regional differentiation

The model reports results for 24 regions, which this report groups into five system types for presentation²⁴. Results are not published at the level of individual regions or jurisdictions, both because regional estimates carry substantially greater uncertainty than global estimates and because the Outlook is not an assessment of any government. Table 13 summarizes the qualitative differences.

Two features of the regional results deserve emphasis. First, the regions that experience the earliest onset of stress are not, in most runs, those that determine whether the global index enters the breakdown regime; that is determined by whether stress reaches the regions with the largest shares of inter-regional flows. Second, advanced economies show a distinctive pattern: high automation and dependence on concentrated imports make them vulnerable to the earliest stages of the P1 and P3 pathways, while greater institutional depth slows the subsequent erosion of governance capacity. Whether that depth is sufficient is the question on which the Model Review Panel’s dissents bear most directly.

11Uncertainty and limitations

Uncertainty in this assessment has four main sources: parametric uncertainty in the model’s calibrated quantities; scenario uncertainty in the external drivers and their weights; structural uncertainty in the model’s representation of systems and their couplings; and deep uncertainty about matters on which no defensible probability can be assigned. The ensemble design addresses the first two directly. The third and fourth can only be examined indirectly.

12Recommendations

The recommendations below are addressed to the Department and, where indicated, to the federal science community. They are chosen to be robust: each is expected to reduce systemic stress or uncertainty across most of the configuration space, not only under the central estimate. They are research and institutional recommendations; they are not public-safety guidance and do not propose action by any other agency.

1. Strengthen the evidence base for the Governance and Social Cohesion module. OSRA and OLSR should conduct a joint calibration study addressing both MRP dissents, extending the historical episode set and developing direct indicators of institutional capacity, with results incorporated before the 2027 edition. This is the largest single source of reducible uncertainty in the assessment.
2. Expand evaluation coverage of automated systems in critical-infrastructure control. OAAI should prioritize evaluation, under the framework in DSC-TR-2026-007, of systems with high common-mode exposure in grid operations, financial clearing and settlement, and communications network management. Evaluation results should be converted into ICSM parameters on a quarterly cycle.
3. Research infrastructure decoupling and graceful degradation. The Department should establish a research line, led by OAAI with CIA-DSC, on the design of critical control systems that retain non-automated fallback operation, limit common-mode dependence on shared underlying systems, and degrade progressively rather than abruptly. The finding that runs retaining such fallbacks enter the breakdown regime at about half the overall rate is the strongest single result of this assessment bearing on research priorities.
4. Characterize buffers and long-recovery dependencies. OSRA should develop, with federal science agencies, improved estimates of inventory depth for critical inputs, spare-equipment holdings for grid restoration, and substitution times for concentrated supply chains, which are among the most influential and least well-measured parameters in the model.
5. Establish quarterly CSSI nowcasting and early-warning indicators. OSRA should publish a quarterly nowcast of the CSSI, together with leading indicators identified in the ensemble, including the rate of recovery from moderate shocks, the concurrence term of the CSSI and the share of dependency paths without fallback. In the P4 pathway, slowing recovery precedes entry by a median of 31 months.
6. Invest in long-duration preservation of knowledge. CIA-DSC and OLSR should continue the Long-Duration Information Preservation Program on its planned schedule, with priority to archival media and facility designs that do not depend on continuous external power, communications or maintenance, consistent with the findings of DSC-TR-2026-005²⁶. The Outlook’s results lengthen the period over which the Department should assume that preserved holdings may need to remain intact without dependable external support.
7. Apply deep-uncertainty methods to multi-decade continuity planning. OLSR should apply the methods in DSC-WP-2026-012 to the Department’s own long-term facility and program planning, using the configuration families of this report as a common scenario set, and testing plans against the non-central configurations as well as the central estimate.
8. Continue the Neural Continuity Research Program on its chartered schedule. The long-term research objective of the Department includes the continuity of human knowledge and, eventually, of human cognition. Whether the latter is achievable is not known; the current state of the art is neuron-level connectome emulation of C. elegans and of partial Drosophila circuits²⁷. The Office recommends that NCRP proceed through its phase gates subject to Research Ethics Review Board and External Scientific Advisory Board review, and does not recommend that its schedule be accelerated on the basis of this report.
9. Sustain independent review and transparency. The Model Review Panel’s mandate should be renewed for the v5 development cycle, with review of the Outlook results as well as the model. Model documentation, scenario definitions and ensemble summary statistics should continue to be published in full.
10. Continue the Outlook series. OSRA should issue the second edition of the Outlook in October 2027, using ICSM v4.3 or later, with a comparison against this edition and a record of any changes to the central estimate and their causes²⁸.

The Office notes that the conditions associated with avoidance of the breakdown regime in the ensemble, namely retained fallback capacity, deeper buffers and institutional capacity that holds under concurrent stress, are not exotic. Each existed to a greater degree in the recent past than it does in the 2026 nowcast. Their erosion was gradual, and the model indicates that their restoration, where it occurs, is gradual also. For that reason the Office regards the early years of the projection period as those in which research and investment have the greatest effect on the distribution of outcomes.