Six frameworks, applied in order to every question.
Each framework answers one part of a question and hands its answer to the next. The result is a set of scenarios, a probability on each, and the options that hold up across them.
Each framework answers one part of the question, then hands it on.
Every card has the same parts: what the framework is for, how it answers any question, how it goes deeper when a question needs it, and one honest limit. The named theories and one formula sit under “Read more”.
How the system pushes back.
- What it is for
- Finding the feedback loops that shape how people and institutions behave: what reinforces a trend, what pushes back against it, and where delays hide.
- On any question
- We map the stocks and flows around the decision, such as money, capacity, approvals and people. We mark the loops that amplify a move, the loops that damp it, and the delay between cause and effect.
- Going deeper
- Where a question turns on a pipeline or a bottleneck, we write the structure down and look for the binding constraint, and for the policy resistance that can make a well-meant move backfire.
- Its limit
- A loop diagram shows structure, not size. Without data on each flow, it says which way pressure runs, not how hard.
Read more: the theories and methods behind systems theory
- System dynamics (Forrester, Sterman): stocks, flows, delays, feedback and policy resistance.
- Coordination failure and the big push (Murphy, Shleifer and Vishny).
- Growth diagnostics, the binding constraint (Hausmann, Rodrik and Velasco).
- Weaponised interdependence (Farrell and Newman): hubs and chokepoints in networks.
- Complex interdependence (Keohane and Nye); state capacity; the rentier state.
dS/dt = inflow(t) − outflow(t)The stock-flow equation. A stock changes only through its flows; the delay between them is where a system surprises.
How the players move against each other.
- What it is for
- Working out how the players in a case interact strategically, and using that to identify several scenarios rather than one.
- On any question
- We name the players, what each wants, what each can do and what each knows. We set what they say beside what they have done, and ask which threats and promises are credible and which outcomes are stable.
- Going deeper
- Where a question turns on a bargain or a standoff, we write the game down: who moves first, what a threat costs, and where each side’s acceptable deals overlap. Each stable outcome becomes a candidate scenario.
- Its limit
- This is structured strategic reasoning, not a simulator, and it is only as good as the preferences it assumes. We do not claim computed equilibria with payoff numbers.
Read more: the theories and methods behind game theory
- Best responses and the Nash equilibrium concept; the bargaining model of war (Fearon): information and commitment problems.
- Audience costs (Fearon), with the empirical debate around them.
- Two-level games (Putnam) and veto players (Tsebelis).
- Selectorate theory (Bueno de Mesquita and colleagues).
- Deterrence and compellence (Schelling); time inconsistency and hold-up (Kydland and Prescott); alliance entrapment and abandonment (Snyder); incomplete-information games.
BRi(s−i) = argmaxsi ui(si, s−i)Best responses: for each choice the other side might make, a player’s best reply. In the grid above, B is each side’s best response whatever the other does, so both end at 1,1 though both choosing A pays each more. Where best responses meet is what game theory calls a Nash equilibrium; we use the idea to reason about stable outcomes, and do not claim to compute one.
How real decision-makers depart from the model.
- What it is for
- Correcting game theory’s rational-choice assumptions with what decision science has learned about how leaders and groups actually decide.
- On any question
- We ask how the decision-makers frame the choice: as a gain or a loss, against which reference point, under what time pressure, and through which beliefs about their rivals.
- Going deeper
- Where one leader or a small group decides, we read their record: their beliefs about politics, their traits, and how their circle handles dissent. A leader facing losses may take risks a rational model would not predict.
- Its limit
- Reading a leader from a distance is inference, not diagnosis. Traits and beliefs are estimated from the public record and kept as hypotheses.
Read more: the theories and methods behind political psychology
- Prospect theory (Kahneman and Tversky): losses loom larger than gains, and small probabilities are overweighted, large ones underweighted.
- Bounded rationality and satisficing (Simon).
- Operational code analysis (Leites, George, Walker): a leader’s beliefs about politics.
- Poliheuristic theory (Mintz): options that fail a domestic-politics test are dropped first.
- Groupthink (Janis); perception and misperception (Jervis); leader trait analysis (Hermann).
v(x) = xα for gains; v(x) = −λ(−x)β for losses; λ > 1The prospect-theory value function, after Tversky & Kahneman (1992), who estimated α and β near 0.88 and λ near 2.25: a loss weighs about twice an equal gain.
Which scenarios could unfold, not only the likely one.
- What it is for
- Combining long-run patterns with the system, the game and the psychology to set out several possible and probable scenarios.
- On any question
- We pick the few uncertainties that matter most, combine them, discard the combinations that cannot coexist, and keep a small set of scenarios that do not overlap and together cover every outcome. Each scenario names dated signposts that would show it arriving.
- Going deeper
- We scan the horizon for weak signals, follow decisions forward so each branch changes what the players can do next, and run every option through every scenario. The further out the horizon, the less a number can carry: long-range scenarios are ranked, not priced.
- Its limit
- A scenario set is only as wide as the people who drew it. A scenario nobody named gets no probability, which is why the set always covers “all other outcomes”.
Read more: the theories and methods behind scenario planning
- Scenario planning, the intuitive-logics school (Wack at Shell; Schwartz; van der Heijden).
- General morphological analysis and cross-consistency assessment (Zwicky; Ritchey).
- Cross-impact analysis, horizon scanning and weak signals, and backcasting.
- Wind-tunnelling: testing each option against every scenario.
- Signposts with kill dates: a dated sign, and the day an option is dropped if it has not come.
N = k1 × k2 × … × kn, pruned by consistencyThe morphological count: n uncertainties with k states each give N combinations; those that cannot coexist are removed, and the scenarios that remain form one partition.
What history says, and a number on each scenario.
- What it is for
- Using historical data and patterns to separate the possible from the probable and the likely, and putting a probability on each scenario already identified.
- On any question
- We start from a reference class, how often comparable cases went one way, and only then adjust for what is specific to the case. Every base rate states its class, its count and its dates. The scenarios’ probabilities sum to 100%.
- Going deeper
- Where the counts allow, we fit count models and print an interval, never a bare point. We test whether a gap could be chance, and update as evidence arrives. Every probability is scored when it resolves.
- Its limit
- Small classes give wide ranges. When a class holds only a handful of cases, the brief prints the count and says the gap could be chance.
Read more: the theories and methods behind statistics and probability
- Reference-class forecasting, the outside view (Kahneman and Tversky; Flyvbjerg).
- Bayesian updating: a prior, the evidence, a revised estimate.
- Count models (Poisson, negative binomial) with bootstrap intervals; exact tests for small samples (Fisher).
- Proper scoring rules: the Brier score and its split into calibration and resolution (Brier; Murphy).
- Estimative language, words tied to ranges, and forecasting practice from tournament research (Tetlock).
Σp = 100%The scenarios’ probabilities add to 100%: every outcome falls in exactly one scenario.
P(H | E) = P(E | H) · P(H) / P(E) · Brier = (1/N) Σt (pt − ot)2Bayes’ rule, and the Brier score on its 0-to-1 scale: p is the stated probability, o is 1 if the event happened and 0 if not. 0 is perfect; always saying 50% scores 0.25.
Which strategies hold up across the scenarios.
- What it is for
- Finding the strategies that hold up across all the likely scenarios, and the less likely ones, rather than the one that wins only in the most likely.
- On any question
- Each option is tested in every scenario. An option that pays, or keeps a reader ready, in every scenario is labelled no-regret; one that pays in a single scenario is labelled a bet, with the scenario it bets on and a kill date. These are analytical labels, not recommendations.
- Going deeper
- In Decision Briefs built under this method, the framework takes a lighter, checkable form: a stated robustness comparison that scores each strategy across the priced scenarios, by expected value and by worst case, with the maths in the method annex. In advisory and engagement work it can draw on the full toolkit: linear algebra, calculus and differential equations combined with strategic studies.
- Its limit
- A robust option is only as good as the payoffs entered. No tool picks a reader’s strategy: the comparison shows how options fare, and the decision stays with the reader.
Read more: the theories and methods behind mathematical optimisation
- Decision analysis and minimax regret (Savage).
- Robust decision-making under deep uncertainty (Lempert and colleagues); robustness analysis (Rosenhead).
- Real options and the value of waiting (Dixit and Pindyck).
- Linear and convex programming; dynamic models from strategic studies, such as Richardson’s arms-race equations and Lanchester’s attrition laws.
- Sensitivity checks: drop one scenario, drop one option, find the single number that would flip the answer.
R(a, s) = maxb U(b, s) − U(a, s); a* = argmina maxs R(a, s)Min-max regret: an option’s regret in a scenario is how far it falls short of the best option there; the robust choice keeps its worst regret smallest. It needs the set of scenarios, not their probabilities.
Nine fields turn uncertainty into strategic insight and foresight.
After the six frameworks come the models and theories of the fields that study a changing world order. A few of those used most often:
Geopolitics
Power transition (Organski and Kugler); balance of threat (Walt); the classical geography of Mackinder, Spykman and Mahan.
Geoeconomics
Geoeconomic coercion and sanctions; weaponised interdependence; chokepoints in trade and finance.
Geotechnology
Export-control regimes; chokepoints in technology supply chains; network analysis of who depends on whom.
Political science
Selectorate theory; veto players; audience costs; state capacity.
Economics
Gravity models of trade; real options; coordination failure; tariff pass-through.
Political economy
Two-level games (Putnam); the rentier state; time inconsistency and regulatory hold-up.
Econometrics
Count models with intervals; exact tests for small samples; panel and time-series methods where the data allow.
International relations
The bargaining model of war; hegemonic stability (Kindleberger, Gilpin); complex interdependence (Keohane and Nye).
Global security
Deterrence and compellence (Schelling); the security dilemma (Jervis); escalation and alliance politics.
An illustration: one question through all six frameworks.
Decision Brief No. 021 asked: can Canada turn promised capital into built projects? Edition 1.1, 1 October 2026, evidence to 27 September 2026. The brief was released before the six frameworks were named, so this is an illustration of how they apply to it, not a description of how it was built. Every figure below is printed in the brief or on the forecast register. Where the brief’s own work does not match a framework, we say what the framework would add instead of inventing a finding.
-
1
Systems theory
A stock of promises, and a thin flow out of it.
Seven pledges came to C$180bn. By 27 September 2026 the brief found one placement against them: a loan of undisclosed size. Read as stock and flow, the promises are the stock and placed capital is the flow. The brief also names a loop it did not model: a fund that commits without a government term tells governments they need not write one.
-
2
Game theory
The players followed a political calendar.
The brief named the players: the funds, the federal and provincial governments, regulators and First Nations partners. Of the 5 pledges made since 1 August, 4 fell in the summit week of 11 to 15 September, with no anchor deal found. It printed its own limit: it did not model the bargain between funds and governments over who writes the terms. Writing that game down is this framework’s next step.
-
3
Political psychology Not used in this edition
What it would add: how the choice is framed.
The brief did not apply this framework. It would ask whether pledges made in a summit week behave like commitments or like signals, and how a fund’s board weighs a disallowed overrun, framed as a loss, against an allowed return. These are questions for a later edition, not findings.
-
4
Futures studies and scenario planning
Three public tests give five scenarios that cover every outcome.
A new build stake by a fund the brief tracks, with or without a government term; published airport terms; a fast-track listing. Together they split every outcome by 31 October 2027 into five scenarios: Terms and takers; Takers without the state; Airports open, builds wait; Fast-track lists, capital waits; and Standstill. Each scenario names what would show it coming; for Fast-track lists, it is an order in the Canada Gazette listing a project.
-
5
Statistics and probability
Base rates first, then a number on each scenario.
Of 18 large projects decided since 2020, 6 started inside their announced window and none is on record as late; the other 12 could not be scored. Projects with a government term in place went ahead 6 times in 15 (40%), against 12 in 58 (about 21%) without. An exact test gives p = 0.18, so the brief says the gap could still be chance.
- 16% Terms and takers
- 27% Takers without the state
- 16% Airports open, builds wait
- 14% Fast-track lists, capital waits
- 27% Standstill
Registered 28 Sep 2026 as F-0151 to F-0155; they sum to 100%. 43% = 16% + 27%. Each is our estimate, scored when it resolves on 31 October 2027. Redrawn from the brief’s own figures. -
6
Mathematical optimisation
Three options hold in every scenario; four bets pay in one.
The brief mapped seven options against the five scenarios. Three no-regret options hold in all five: preparing for the federal offers; an airport bid case, binding only once the terms publish; and one overrun rule per asset type, set from a reference class whose median is 1.11 times first cost and whose highest is 1.23. Four bets each pay in one scenario.
A stress test sets the options in the reader’s units. In the brief’s model of one regulated line, the allowed 9.11% return on equity falls to 6.3% if an 11.4% overrun is fully disallowed, and to 4.37% at 23.1%, 0.10 points above the 4.27% long Canada bond. The brief calls this an illustration, not a forecast of the project’s cost or of any regulatory decision.
The brief’s mapping was made by hand, without scores. The scored robustness comparison described above is what the framework adds.
What we have built is kept apart from what we draw on.
- Base rates, each stating its reference class, its count and its dates.
- Count models with bootstrap intervals, which print a range and refuse a bare point estimate.
- Robust choice with minimax regret, with checks that drop one scenario or one option and find the single number that would flip the answer.
- A sealed forecast register: each probability is sealed (hashed with SHA-256) on the day it is registered, listed on the register when its brief is released, and never edited.
The wider toolkit of the six frameworks. Some are applied by hand when a question needs them; none is a tool we have built.
- Monte Carlo simulation, to carry uncertainty through a model.
- Regression, panel and time-series econometrics.
- Tests for regime change and structural breaks.
- Game models of bargaining and deterrence; morphological and cross-impact analysis.
- Operational-code and leader-trait analysis.
- Linear programming and dynamic models built on differential equations.
Registered before the fact, scored when it resolves.
Every probability Strature publishes is our estimate, scored when it resolves.
- 1A question that can be settled, and every outcome.
A test a named third party could resolve from public sources on a stated date, and scenarios that sum to 100%, including “all other outcomes” where the set needs it.
- 2Set and registered.
The Principal sets each probability. It is sealed (the full statement hashed with SHA-256) on the day it is registered, listed on the register when its brief is released, and never edited. Publishing the hashes is our next step.
- 3Scored with the Brier score.
On its resolution date each probability is scored against what happened: 0 is perfect, and always saying 50% scores 0.25.
- 4No skill claim yet.
No aggregate claim about skill is made until enough forecasts have resolved for our 36-month rule to allow it.
Every forecast in our published briefs, resolved and open: see the forecast register.
The same six frameworks, weighted to the brief.
- Decision Brief
- One question, for one class of decision-maker. Decision Briefs built under this method will name all six frameworks in a method annex, with the body kept in plain English and optimisation shown as the robustness comparison. Briefs released so far were not built with the six frameworks named and carry no such annex; they used much of the same thinking without naming it.
- Country Outlook
- One country’s scenarios. Systems, game theory and psychology carry the most weight: the regime, its players and its capacity. Base rates are drawn country by country.
- Scenario Brief
- One event inside a scenario set. Futures studies and scenario planning, with probability, lead: a registered headline probability, the scenarios around it, and dated signposts.
The limits, printed with the work.
Options to consider are illustrations, not instructions. They are general research, not legal, tax, accounting or investment advice, and never a recommendation on any security.
Four more limits, and the rules under every framework
- Every probability may prove wrong. It is our estimate, scored when it resolves.
- Not every framework is used with equal depth on every question. Where one was not applied, the work says so.
- Named theories are what we draw on. Their authors have not reviewed or endorsed this work.
- Client material never leaves its client. It is kept per client, never pooled, and never used in a published brief.
- Four rules hold every piece of work: nothing is invented; a retrieval asserts its own completeness; a scan of nothing is not a clean scan; a conflict is printed, not resolved. They are set out with the Exposure method.
Exposure uses a different method.
Exposure does not forecast. It reads what governments have published, keeps what reaches a company through something the company itself disclosed, and tests each change for materiality. Exposure shows where a company is exposed now; the six frameworks ask what may come next.
Tell us the decision you are facing.
The Principal reads every message and replies from the firm’s own address. Your message is never added to a mailing list unless you tick the box to be kept informed. No tracking of people: we count visits and downloads without cookies.