Scenario methodology

Three layers

The scenarios sit between two other bodies of work, and each does a different job.

Outcomes

Outcomes are the second thing this site keeps, beside the scenarios. A trend, which GINC keeps on a site of its own, is a slow structural force. A scenario is an event: a cause, with effects across many capabilities at once. An outcome is neither. It is an isolation point: one named place in the capability framework crossing a threshold, whatever caused it. ‘A regional war closes a sea route’ is a scenario; ‘a collapse in energy security’ is an outcome it may produce, and so may others. Outcomes are what a government fears or wants, and they are where the reverse stress test starts.

Each outcome is defined on a basket of capability groups. A failure outcome is met when the basket’s mean score falls a full band, three points. A prosperity outcome is met when it rises half a band.

There are fifty, twenty-five of each kind.

• Eighteen are set by rule. For each of the nine capability domains, the whole domain falling a band and the whole domain rising half a band. These guarantee that every one of the 73 capability groups sits in an outcome, and no hand chose them.
• Thirty-two are named. Narrower isolation points of two to four capability groups, inside one domain or across several: a failure of digital lifelines, a loss of strategic deterrence, research translated into value. They were built the way the trends and scenarios were in the second pass: three AI models each proposed thirty-two separately, each then saw all three sets and settled on its own, and a fourth made the final cut. Ten of the thirty-two are held by all three models, twenty-one by two and one by one. Between them they reach 72 of the 73 capability groups.

The first reverse stress test ran on twelve outcomes chosen by one hand, which reached 28 of the 73 groups and nothing in national security or in information and influence. The fifty replace them.

Three ways to run a scenario against a nation

A scenario can be run against a nation’s capability scores in three ways. Each answers a different question. All three have been run for illustration against the 2026 capability scores, with exposure set nation by nation. The results are published with their settings, the load and capture tests are summed into a vulnerability index and an opportunity index, and the reverse stress test is published outcome by outcome and nation by nation. The methods and their settings are proposals for the review committee, not adopted practice.

All three use the same inputs:

• Baseline. The nation’s capability scores and domain tiers from the Atlas.
• Load. The scenario’s loading table, mapped from working domains to the capabilities beneath them through the domain map. High, Medium and Low become a size of shock.
• Exposure. How directly the scenario reaches this nation, from 0 to 1. It starts from the scenario’s regional exposure and is refined by Atlas metrics that the scenario’s own parameters name: the share of crude arriving through one strait, months of import cover, the share of the budget funded by aid.
• Absorption. How much of the shock the nation’s buffers soak up, from 0 to 1. Buffers are the capabilities the scenario’s stakeholder actions point to: stockpiles, fiscal space, backup power, surge capacity.
• Policy response. The grammar’s ninth slider: none, current plans executed, or best practice.

1. Load test

What happens to this nation under this scenario?

The load test is the forward stress test, and the closest to what bank supervisors do. The scenario is fixed at Standard Severe and applied to every nation alike, so results compare.

For each capability the scenario loads:

stressed score = baseline score − load × exposure × (1 − absorption)

1. Map each loaded working domain to its capabilities.
2. Turn the load into a shock: High is sized to move a fully exposed, unprotected capability down one tier; Medium half of that; Low a quarter.
3. Scale by the nation’s exposure and reduce by its absorption.
4. Recompute each domain’s score and tier.
5. Run three times, at each policy response. The gap between ‘none’ and ‘current plans’ is the capability finding. The gap between ‘current plans’ and ‘best practice’ is the policy recommendation.

It reports: stressed scores by domain; the number of domains that fall a tier; the capability that falls furthest; the recovery horizon from the scenario’s scorecard.

Strengths. Simple, transparent and the same for every nation, so it can be published as a comparable result. It uses only what the Library and the Atlas already hold.

Limits. It is linear: it does not capture one failed capability pulling down another. The load sizes are judgements. It says what happens at one intensity and nothing about how close the nation is to something worse.

2. Reverse stress test

What would it take to produce this outcome, and how likely is that?

The reverse stress test works backwards. The other two tests fix a scenario and read the result. This one fixes the result, such as a collapse in energy security or a step up in a critical technology, and searches for the conditions most likely to produce it. Run towards failure it exposes what threatens resilience. Run towards a gain it exposes what drives prosperity.

1. Take an outcome. The test runs on each of the fifty outcomes. A failure outcome is met when its basket of capability groups falls a full band, three points. A prosperity outcome is met when the basket rises half a band. The bar for gains is lower because, in this model, a scenario delivers gains more slowly than it delivers losses.
2. Search the conditions. A condition is one scenario or two together, each at Major, Severe or Extreme, with the nation’s current plans holding or failing. Every downside scenario and every pair is tried for a failure outcome; every upside scenario and every pair for a prosperity outcome. The load test supplies the fall and the capture test supplies the gain.
3. Keep the routes. A condition that meets the outcome is a route. Each nation has its own routes, because exposure and buffers differ.
4. Rank the routes by likelihood. A route’s likelihood starts from the scenario’s published likelihood and is adjusted for severity, for whether plans must fail, and for two scenarios having to fall close together.

It reports: the likelihood of the outcome over ten years and three; the most likely route; headroom, which is how many times Standard Severe the nearest single scenario would have to be to do it alone, so that below 1 means the standard case is already enough; the capability in the basket that moves first; and the lever, which is the weakest buffer against the lead scenario for a failure, or the weakest required capability for a gain.

What it assumes. These are GINC assumptions and are published with the run.

• Major is 0.6 of the Standard Severe load and Extreme is 1.6. An event at Major or worse is 2.5 times as likely as one at Severe or worse; an event at Extreme is one fifth as likely.
• Plans are more likely to fail where buffers are weak: the chance is one half times the share of the scale by which the nation’s buffers fall short.
• Two unrelated scenarios in the same ten years compound only if they fall close together, which is taken as three times in ten. Where one scenario is recorded as triggering the other, the second follows the first one time in four.
• An upside scenario lifts the capabilities it loads by the same points as a downside scenario lowers them, scaled by the nation’s capture rate: the mirror of the load test. In the first run it lifted only the capabilities the scenario requires, which left most prosperity outcomes out of reach.

Strengths. It answers the question a government asks, which starts from the thing it fears or wants and not from a list of scenarios. It finds compound failures that a single scenario misses: in this run most routes to failure need two scenarios together. It does not depend on any one scenario’s likelihood being right, because an outcome usually has several routes.

Limits. The search can only find routes made of scenarios in the Library, so an outcome with no route is not an outcome that cannot happen. A full band is the same fall for every nation: for a strong nation it is serious but leaves it far above where a fragile state starts, so each result shows the band before and after. The likelihood combines routes that share scenarios and is an estimate, not an exact probability. The upside half rests on the capture test and is the least tested part of the method.

3. Capture test

How much of the upside does this nation capture?

The capture test is the positive test. It is built for the upside scenarios, where the question is not what breaks but who gains, and it also applies to the nations that come out ahead in a downside scenario. Here the loading table is read the other way: the capabilities a scenario loads are the ones a nation needs in order to take the gain.

net position = gain × readiness − loss exposure × (1 − adaptability)

1. Readiness. Take the capabilities the scenario loads High. Readiness is the nation’s score on them relative to the frontier, and it is held down by the weakest: a nation with cheap power and no grid to carry it captures little of an energy breakthrough.
2. Loss exposure. Measure the nation’s dependence on what the scenario makes obsolete: hydrocarbon revenue as a share of the budget, knowledge-work exports, industries built behind tariffs.
3. Adaptability. Score the capabilities that let a nation move out of a losing position: fiscal buffers, retraining, government effectiveness.
4. Combine the three into a net position.

It reports: a capture rate from 0 to 100; a net position of gainer, mixed or loser; the binding capability, which is the one whose improvement would raise the capture rate most; and the nation’s position among its peers.

Strengths. It turns a scenario into an investment priority and shows that the same event sorts nations into winners and losers. It gives the upside scenarios a result of their own, since ranking them by risk would misread them.

Limits. The size of the gain in an upside scenario is far less certain than the size of a loss in a downside one. Readiness measured today may not hold over a scenario that runs for five years. It is the least tested of the three.

How the three fit together

The three are complementary, not alternatives. The load test runs every nation against every downside scenario and is the comparable public result. The capture test does the same for the upside scenarios. The reverse stress test is built on both: it takes the falls from the load test and the gains from the capture test and asks which combinations of them produce an outcome a government cares about.

The run so far

The results pages show the load test on the ten downside scenarios and the capture test on the ten upside ones, for 197 nations. This is the third run, on the scenario set of the second pass. The reverse stress test is run on fifty outcomes, twenty-five of failure and twenty-five of prosperity. All settings are published in the test settings file on the data page.

• Scores. Capability scores for 73 capability groups in nine domains, on a scale of 0 to 20 in seven bands of three points, from the GINC National Capability Index 2026. The index describes them as candidate ratings, not adopted ratings.
• Load. High is three points, which is one band; Medium is 1.5; Low is 0.75. Each scenario singles out about ten capability groups that take its domain’s load in full. Other groups in a loaded domain take the load one level lower.
• Exposure, nation by nation. Each scenario has its own exposure formula: a weighted blend of two or three indicators and the scenario’s regional exposure. The indicators come from the 2025 World Factbook and are scaled from 0 to 1: the share of oil and gas use not met at home, months of imports covered by reserves, public debt, trade relative to GDP, natural hazards, land borders with states in conflict, the share of the population over 65, and others. A regional war that closes a sea route, for example, weighs energy import dependence at 45 per cent, the regional exposure at 35 and trade openness at 20. Where a nation does not report an indicator, its weight passes to the rest. Each scenario’s results page lists its formula.
• Absorption. A nation’s mean score on the scenario’s buffer capabilities, as a share of the top of the scale, times 0.75.
• Policy response. No response sets absorption to zero. Current plans uses the nation’s own buffers. Best practice lifts each buffer to 15, the foot of the Advanced band, where it is lower.
• Capture test. Gain and loss exposure are set from indicators where one fits: energy import dependence for the gain from cheap energy, fuel exports for the loss; ageing for pension strain. Where none fits they are set by region, with named exceptions.

What the results report. For the load test: the change in the nation’s capability index in points and per cent; the rank shift, which is the change in its place among all nations once every index is stressed; the number of capability groups and domains that fall a band; and the weakest point, the singled-out capability left lowest. For the capture test: capture, loss and net position; gainer, mixed or loser; rank; and the binding capability.

Two indices

The scenario results are summed into two indices for each nation, weighting each scenario by its likelihood.

• Likelihood. Each likelihood band is turned into a probability at the midpoint of its range: 0.1, 0.6, 3 and 15 per cent for bands 1 to 4, and 40 per cent for band 5, which is open-ended. The indices are computed at two horizons: ten years, from the ten-year band, and three years, scaled up from the two-year band on a constant yearly chance.
• Idiosyncratic scenarios. For scenarios that strike one nation at a time, a nation’s probability is the scenario’s probability scaled by its exposure relative to the average nation. A state with thin reserves and heavy debt is then more likely to meet a funding crisis, as well as harder hit by one.
• Vulnerability index. Expected loss: the sum over the downside scenarios of probability times the share of the baseline capability index lost. It is scaled so that the most vulnerable nation is 100. The loss is taken as a share of baseline, not in points, because a weak state that loses one point has lost far more of what it has than a strong one.
• Worst case, and tail. The largest single fall across the scenarios, with the scenario that causes it, and the mean of the three largest. These sit beside the index because an average hides the one scenario that breaks a nation.
• Opportunity index. The likelihood-weighted mean net position across the upside scenarios, from −100 to +100, with the best single case beside it.
• Other measures. Breadth is the number of downside scenarios in which a capability domain falls a band. Shock absorbed is the share of the load a nation’s own buffers take, across all downside scenarios. The policy dividend is the expected index points recovered by moving from current plans to best practice. The weakest link is the capability that is the nation’s weakest point in the most scenarios.
• Four groups. Each nation is placed against the median of each index: well placed, high stakes, insulated or exposed.
• Seven archetypes. Nations are also grouped by the whole shape of their results. Each nation’s result in all 20 scenarios is standardised and the nations are clustered into seven groups by k-means, with a fixed seed so the grouping does not change between runs. The groups are found by the method; the names and descriptions are GINC’s reading of them. An archetype describes a pattern of results in this run, not a nation.

How to read them. The run shows the method working on real scores. It is not a finding about any nation, and the indices are not adopted rankings. Three cautions apply. Scenarios are treated as independent when one can trigger another, so the sum misses compounding. The two indices move against each other, because the capabilities that absorb a shock are largely those that capture a gain. And very small states produce odd results, because several capabilities do not apply to them.

The second pass

The first pass of this site was a GINC desk build: the trends, the scenarios and every score were drafted in one hand. In October 2026 the whole build was put through a second pass, to test it against other judgement before the expert panel sits. Three AI models from three developers were used as a stand-in panel, and a fourth made the final cut to GINC’s brief. Everything on this site now comes from that pass: the twenty scenarios, their ratings and the results. The trend list and its scores against the scenarios came from the same pass and are kept at GINC Trends.

The record of the pass, with each decision and the reason for it, is on the data page.

How it was run

The same three steps were used for the trends and then for the scenarios.

1. Separately. Each model was given the first-pass material and asked to improve it, without sight of the others.
2. Together. Each model was shown all three results and asked to settle on a final version, taking the best of each and not simply returning its own.
3. Final cut. The three final versions were put side by side and a fourth model made one list from them, weighing how many models held each item but deciding on merit, and writing each entry afresh.

Agreement is reported throughout as support: the number of the three models whose final version held the item.

The trends

The trends were revised first, because the scenarios are rebuilt from them. Each model revised the 250 and balanced them across the nine capability domains, and the final list was cut one domain at a time. Of the final 250, 219 are held by all three models. How the list changed, and why, is set out in the Trends methodology.

The scenarios

Each model was given the revised trends and the 24 first-pass scenarios and asked to rebuild the set from first principles as twenty: ten downside and ten upside. Ratings, loads and regional exposure in the final set are the median of the models that proposed each scenario.

Three things changed in kind.

• Fewer, sharper downside scenarios. Fourteen became ten. Four pairs were folded together where the panel judged one to be the sharpest form of the other, and mass displacement was retired because two of the three models held that its loads are tested elsewhere. With ten downside scenarios, the ranked ten are now the whole downside set.
• One new scenario. A cryptographic trust break, in which a practical attack on public-key cryptography leaves networks running and trust in them broken, was proposed by one model and kept in the final cut because nothing else tests it.
• Upside scenarios became events. The first pass described several upside scenarios as open-ended shifts. The panel restated each as something that can be seen to have happened, with a threshold: firm clean power 40 per cent below new fossil plant, a discovery step verified in three scientific fields, a compact that enters into force. The superintelligence takeoff is no longer the standard case; it is carried as the Extreme setting of the automated discovery step.

The panel also rated near-term likelihood lower than the desk had. No scenario now reaches Tier I, and six of the ten downside scenarios tie on priority.

Each of the twenty records was then written in full from its specification, starting from the first-pass record where the scenario continues one, and checked against the schema. The test settings for each scenario, which name the capabilities it loads and the indicators that set each nation’s exposure, were drafted with the record.

The scores

Every trend was then scored against every scenario by each of the three models on a seven-point rubric, 5,000 pairs each, and the median kept. Each scenario page shows the trends rated Very high or Critical for it. The scores, and how far the models agreed on them, are published with the trends at GINC Trends.

Recalculation

With the scenarios in place, everything downstream was recomputed: the load, capture and reverse stress tests on the twenty scenarios; the two indices; and the seven archetypes, which were re-read and two of them renamed.

What to make of it

A panel of models is not a panel of experts, and this pass does not replace the elicitation described below. It does three useful things. It shows where the first pass rested on one view: the items that changed. It gives each item a measure of agreement. And it exercises the whole pipeline, from list to results, on material GINC did not write.

Its weaknesses are stated plainly.

• The models share sources. Three models trained on overlapping text agree more readily than three independent experts would. Support of three is weaker evidence than it looks.
• One model could not hold a count. Asked for exactly 250 trends or exactly ten scenarios, it returned more or fewer through several corrections. Its lists were settled by asking it to choose into numbered places, one domain at a time. Its judgement of which items to keep is its own; the count was enforced.
• One model’s first scenario set stayed close to the first pass. Most of the new thinking on scenarios came from a single model in the first round and was taken up by a second in the second round.
• Figures in the records were written from memory. The historical anchors are real events, but the figures quoted for them were not checked against sources in this pass, and several records say so in their open questions. They need a sourcing pass before release.
• Likelihoods fell without new evidence. The lower two-year likelihoods are the panel’s judgement, not a finding. They change the ranking and the indices, and should be read as one more reason the ranking is provisional.

Two objects: the Library and the Edition

The Library is the stable set of named scenarios. Each scenario has a permanent identifier and slug, a record on one schema and a rating history. The Library grows over time; scenarios are retired, not deleted.

The Edition is the annual cut, released each January. It ranks ten downside scenarios from the Library on likelihood and impact, records movement against the previous edition and publishes the paper and the data. Rankings move; scenarios persist.

Version strings follow the two objects. The Library carries a semantic version (this is v0.4). An Edition carries its year (2027). Every scenario page shows both.

Three rules shape every page:

1. Likelihood is global; impact is per nation. A scenario’s likelihood is a property of the world. Its impact is a property of each nation’s balance sheet. The Edition rates systemic impact and a typical national impact; nation-specific impact arrives when the Atlas connects.
2. The data is the product; the site is a view of it. Every scenario is a data object rendered into a page.
3. Scenarios are shocks, not trends. Slow structural forces such as polarisation, inequality, climate change, demographic ageing, antimicrobial resistance and economic stagnation belong to GINC Trends. Each scenario names the trends that raise its likelihood.

Upside scenarios

Not every shock is a loss. The Library carries ten upside scenarios beside the ten downside ones, two in each family: a debt service reset, a market access opening, great-power détente, a Middle East settlement, a clean power threshold, a water-efficient food leap, an automated discovery step, a public service leap, a chronic disease advance and an outbreak defence that succeeds. In the second pass each was restated as an event that can be seen to have happened, with a stated threshold, where the first pass had described several as open-ended shifts.

An upside scenario is a shock that most would count as progress. It is not upside for everyone, so each record names who gains and who loses. It is specified on the same grammar and rated on the same scales, read as follows:

• Likelihood is unchanged: the probability that the Standard Severe event occurs within the horizon.
• Impact measures the scale of change, in either direction: output gained or redistributed, and capability band shifts up or down. Levels 4 and 5 are labelled Major and Transformative for these scenarios, not Severe and Catastrophic.
• Load marks the capabilities a nation needs in order to capture the gain or absorb the loss.

Upside scenarios are not ranked. The ranked ten order downside risk, and a priority score would read a large gain as a large threat. Some of these scenarios are classed elsewhere as risks, automated discovery above all: its Extreme setting is a superintelligence takeoff, and the record says so and lists the open question.

The backbone: every variable a scenario is built from

Every scenario and every shock is specified on the same set of variables. This is what makes two scenarios comparable, and it is what will let a custom scenario be built and run on the same footing as the published ones. A scenario is a value chosen for each variable below, plus a short list of parameters of its own.

Each variable answers one plain question.

What kind of event it is

Event reach changes what the likelihood means, so it is worth reading twice.

• Global event: one event, many nations. Some scenarios happen to the world at once: a pandemic, a global funding seizure, a regional war that closes a sea route. There is one event, and the question for each nation is how hard it is hit. Likelihood is the chance that the event happens at all.
• Single-nation event: one nation at a time. Others happen to a single nation: hybrid coercion, a legitimacy rupture, a disaster in the core. Somewhere in the world this happens most years. The scenario asks what happens when it is you, and likelihood is the chance that it is you.

Sixteen of the 20 scenarios are global events and four are single-nation events. The Library and the data use the older terms systemic and idiosyncratic for the same two values.

How it unfolds in time

Onset pace separates shocks from accelerated trends. A cut-off that lands in days leaves only what is already in stock; a diffusion that takes years leaves time to adapt and time to fall behind. Warning time is a different thing: a storm can be forecast days ahead and still arrive in hours. Recovery horizon was split from duration after the 2026 closure of the Strait of Hormuz showed about five weeks of closure followed by months of partial recovery.

How large it is and where it lands

Scope and event reach are close but not the same. Scope is the footprint of the damage. Event reach is how the event is counted. A global scope removes the option of leaning on neighbours, because they are hit too.

What is assumed around it

‘None’ is the honest setting for external support and the one that reveals sovereignty gaps. Compound settings are where most real failures live. On policy response, the gap between ‘none’ and ‘current plans’ is the capability finding, and the gap between ‘current plans’ and ‘best practice’ is the policy recommendation.

What it does to capability

What is judged, not set

The variables above are chosen. The ratings below are judgements made about the scenario once it is specified, and they are described in the sections that follow.

How the records use the values

A record takes one standard value for each variable and may add a qualifier in brackets where the event needs it: ‘national (regional cluster)’, ‘1 year (two seasons)’, ‘adversarial (great power)’. The standard value is what scenarios are compared on; the qualifier is for the reader. Onset pace takes the four standard values only, with no qualifier.

Presets and custom scenarios

There are three presets: Standard Major, Standard Severe and Standard Extreme. Standard Severe is the comparability anchor. Every scenario is rated at it, following the Swiss practice of assessing all hazards at one common intensity. Standard Extreme is the reasonable worst case in the UK National Risk Register’s wording: ‘the worst plausible manifestation of that particular risk (once highly unlikely variations have been discounted)’.

A custom scenario is built the same way as a published one: choose a value for each variable, name the parameters that define the event, and mark the load on each of the nine capability domains. Because it sits on the same backbone it can be put through the same load, break-point and capture tests. A custom run is labelled ‘non-standard run’ and is left out of comparisons with the published set.

In this release the variables are displayed at their preset values and are not adjustable, and the site does not yet offer a way to build a custom scenario. The tests read the capability loading and the likelihood; they do not yet read onset pace, warning time or the other settings.

Likelihood

Likelihood is published in five bands, following the UK National Risk Register convention, at two horizons: two years (to end-2028 for Edition 2027) and ten years.

Each scenario carries an event reach, which sets what its likelihood means.

• Global event (systemic). The event is global or near-global when it occurs. Likelihood is the probability that an event meeting the Standard Severe definition occurs within the horizon.
• Single-nation event (idiosyncratic). The event strikes one nation at a time. Likelihood is the probability that a given nation experiences the Standard Severe event within the horizon. It is estimated from the base rate across the 197 GINC nations, with clustering noted where exposure is concentrated: hybrid coercion in Europe and the neighbours of hostile states.

Impact

Systemic impact is global and uses five-year accounting. It follows the GDP@Risk convention developed by Lloyd’s and the Cambridge Centre for Risk Studies, extended from 107 to 197 nations.

National impact is rated for a typical highly exposed nation on the same five labels. It is judged on seven dimensions adapted from the National Risk Register’s impact framework and mapped to the National Capability Framework: human welfare, essential services, economic damage, security, international position, social cohesion and environment. The score is the highest dimension reached, not an average.

Calibration points: the five-year, probability-weighted losses published by Lloyd’s (pandemic US$13.6 trillion, geopolitical conflict US$14.5 trillion, food and water US$5 trillion, cyber US$3.5 trillion, space weather US$2.4 trillion) sit between levels 2 and 3 of the systemic scale. Their extreme variants (pandemic up to 6.4 per cent of global GDP, conflict up to US$50 trillion) reach level 5. Supervisory ‘severely adverse’ scenarios (Federal Reserve 2026, Bank of England 2025) are level 5 events for the economies they model.

Capability loading

Each scenario maps to nine working domains, three in each dimension of the National Capability Framework: Hard, Soft and Economic. A domain’s load is High where capability band shifts are expected under current plans, Medium where band shifts occur only with no policy response, and Low where there is strain without a band shift.

The working domain labels are placeholders. Their reconciliation to the canonical framework domains is recorded in domain-map.json on the data page; two of the nine do not yet map one-to-one.

Confidence and priority

Confidence has five levels, from very low to very high, each with a one-line reason. Confidence refers to the rating, not the scenario.

The development ranking uses a priority score: two-year likelihood multiplied by the higher of systemic and national impact. Tier I is 15 and above, Tier II is 9 to 14 and Tier III is under 9. Ties are broken by systemic impact, then confidence, then ten-year likelihood.

The panel, not the arithmetic, sets the Edition. The priority score exists for development and transparency. Where the published ranking departs from it, the departure is recorded as a judgement in the edition’s data file and shown on the edition page.

Stakeholder relevance

Stakeholder views order the ranked scenarios by a relevance score from 1 to 5. In this release the score is a development default derived from the loading table: each group reads two working domains; High counts 2, Medium 1 and Low 0; the score is the sum plus one. From the mid-2027 release it is replaced by the split of the panel’s own ratings by stakeholder group.

Borrowed calibration

Where a supervisor or an established scenario programme has published a calibration, the Standard Severe preset borrows it and cites it. GINC’s own judgement is labelled as such. Each programme is described in full, with its strengths and limitations, on the benchmarks page.

Figures marked ‘about’ or ‘around’ on scenario pages are approximate. Anchors from 2026 rely on press reporting and are to be re-checked against primary data before the paper is typeset.

Elicitation method for the Edition

The scores on this site are provisional GINC desk scores. The 2027 Edition replaces them with the results of a structured expert elicitation:

• 40 to 60 named panellists, drawn across the nine regions and the four stakeholder groups.
• Two rounds: structured individual ratings with a written rationale, then a calibrated discussion round.
• Publication of the distribution of ratings, not only the median.
• Panel tagging by region and stakeholder group, which produces the stakeholder and regional splits.
• Adoption of the final ratings by the review committee.
• Conflicts of interest declared.

Limitations

• Loading tables are judgement-based until the Atlas connects.
• There is no agreed loss model for nations; national impact is a structured judgement.
• Likelihoods for adversarial scenarios are politically sensitive and rest on thin evidence.
• Bands and levels carry a risk of false precision. Each is published with its range and a confidence level for that reason.

Trends

The slow structural forces behind the scenarios are kept on a site of their own, GINC Trends: 250 trends, each assigned to the one domain of the National Capability Framework it bears on most, each scored against every scenario here and against every other trend, and each followed for the public attention it draws.

This site uses the trends in one place. Each scenario page lists its key trends: those scored Very high or Critical for it on a seven-point rubric, where a trend relates to a scenario when it changes how likely the scenario is, how hard it lands, or is itself sharply changed by it. Trends are evidence and context. They are not an input to any test or index here.

This site keeps a copy of the trend list and its scores, taken from GINC Trends, so that the two can be read together. How the list was built, how the scores were given and how the trend network is found are set out in the Trends methodology.

Relationship to the Atlas, the Shock Register and Trends

The Atlas is the balance sheet: every nation’s capabilities, metrics and sources. Scenarios is the load. This site links to the Atlas for every capability domain, and will link to it for every nation once nation-level results are published.

The Shock Register holds the historical events that calibrate the scenarios; each scenario’s anchors reference it. GINC Trends holds the slow structural forces; each scenario names the trends that amplify it and links to them there.