Methodologies in Geopolitical Forecasting

LESSON METADATA

Field	Value
Lesson number and title	58 — Methodologies in Geopolitical Forecasting
Module	Module 10: Forecasting, Diplomacy, and Capstones
Prerequisites	L3 (causal inference), L4 (epistemology), L14 (contemporary transition)
Interleaved links	Forward to L59 (negotiation), L60–61 (capstones)
Reading load	7-9 hours
Content channels	Text, data lab, worked example, problem pair, flashcards
Bloom ceiling	Evaluate / Create
Estimated study time	7-9 hours

SECTION 0 — Why This Lesson Exists

Geopolitical forecasting occupies an uncomfortable position in the study and practice of international relations: it is indispensable to decision-making yet systematically unreliable in its record. States, international organizations, and analysts are repeatedly asked to anticipate wars, regime changes, alliance shifts, and economic shocks, and they are repeatedly surprised. The puzzle this lesson addresses is not merely that forecasting fails, but that failure is unevenly distributed across methods, domains, and forecasters, and that the institutions producing forecasts often reward rhetorical confidence over demonstrated accuracy. The lesson therefore asks why some approaches to anticipating geopolitical futures outperform others, what “outperform” can mean when the future is singular and non-repeatable, and how the authority to predict is itself a political resource rather than a neutral technical credential. It treats forecasting as an epistemic practice embedded in power, one that must be evaluated simultaneously on methodological, political, and ethical grounds.

The lesson is positioned deliberately within the curriculum. It builds on L3’s causal inference, which supplies the tools for attributing causes to events and for distinguishing correlation from mechanism; on L4’s epistemology, which supplies the standards by which knowledge claims are justified; and on L14’s account of the contemporary transition, which supplies the substantive context of a post-Cold War system whose volatility has outpaced many of the models inherited from the bipolar era. It also prepares the ground for L59 on negotiation and for the L60–61 capstones, because any serious account of diplomacy or strategic choice presupposes some account of how the environment into which diplomacy is inserted is anticipated. Without a disciplined treatment of how futures are constructed, negotiation strategy and capstone analysis risk projecting present assumptions onto tomorrow and mistaking the familiar for the probable. Forecasting is thus treated here as a methodological capstone rather than a peripheral skill.

The lesson develops five methodologies, each of which answers a different question and carries distinct scope conditions. Scenario planning, associated with Schwartz (1991) and the Royal Dutch Shell tradition, addresses deep structural uncertainty by constructing multiple plausible futures rather than point predictions, forcing decision-makers to test strategies against worlds they do not expect. Delphi methods elicit and refine expert judgment through structured anonymity and iterative feedback, attempting to extract collective wisdom while suppressing the pathologies of group deliberation. Bayesian updating imposes a formal discipline of belief revision, requiring analysts to state priors explicitly and to update them in light of evidence through Bayes’ theorem, thereby making the reasoning behind a forecast auditable. Quantitative conflict modeling—exemplified by ICEWS, ACLED, and GDELT—detects patterns in large event datasets that unaided human analysts cannot process, trading narrative richness for statistical coverage. Calibration and scoring, rooted in Brier (1950) and developed through the Good Judgment Project and the work of Tetlock and Gardner (2015), introduce accountability by measuring whether stated probabilities match observed frequencies. The lesson’s central analytical move is to refuse the temptation to rank these methods on a single axis; instead it maps each to the conditions under which it is appropriate and to the failures it is prone to.

The African perspective is primary in this lesson because it reveals something the other four perspectives systematically obscure: that forecasting is not only about accuracy but about who gets to define the problem space. African early warning systems—continental (the African Union’s Continental Early Warning System), regional (ECOWAS’s Early Warning and Response Network, IGAD’s CEWARN), and national—were built not merely to predict conflict but to claim epistemic authority over African security dynamics that external actors routinely misread. African scholars such as Grovogui, Mbembe, Bayart, and Mamdani have long argued that knowledge about Africa produced in Western centers carries structural biases: an overemphasis on state collapse, an underrecognition of informal and hybrid governance, and a tendency to treat African conflicts as exceptional rather than as instances of broader systemic dynamics. From this vantage point, when external models predict “state failure” in contexts where hybrid political orders persist and adapt, the prediction is not simply inaccurate; it is performative, shaping intervention priorities, resource flows, and the very categories through which African politics is legible to the international system. The African framing thus reframes the lesson’s central question from “who predicts well?” to “well for whom, by what criteria, and with what consequences for the predicted?” It insists that a forecast’s accuracy cannot be separated from its distributional effects and from the hierarchy of credibility that determines whose knowledge counts as expertise.

By the end of the lesson, the student should be able to perform six operations. First, distinguish the five methodologies by their epistemic commitments, mechanisms, and scope conditions rather than by disciplinary fashion. Second, design a forecasting protocol suited to a specific decision context, including the choice between qualitative and quantitative approaches and the specification of falsifiers. Third, score probabilistic predictions using Brier scores and calibration curves, and interpret the difference between calibration and resolution. Fourth, reconstruct how African, Chinese, Russian, Western, and UN perspectives frame the same predictive task differently, identifying what each contributes and what each occludes. Fifth, adjudicate between competing forecasts on the basis of evidence and stated assumptions rather than on the authority of the forecaster. Sixth, recognize the strongest counterarguments to forecasting as an enterprise—particularly the claim that geopolitical events are unique, path-dependent, and shaped by strategic interaction that defeats statistical regularity—and respond to them without either dismissing or overstating them. The worked example on the Good Judgment Project (2011–2015) and the problem pair on forecasting the 2022 Russian invasion provide concrete cases for practicing these operations, while the data lab exposes students to ICEWS, GDELT, and ACLED extracts alongside tournament outcome data.

The lesson closes with a methodological humility that is not resignation. It does not promise a method that eliminates uncertainty, because no such method exists; even the best-calibrated superforecasters are wrong substantial fractions of the time, and the future is not a sample from a stable population. What the lesson does promise is the discipline of making uncertainty explicit, measurable, and revisable: stating what one believes, with what confidence, on what grounds, and what evidence would change one’s mind. The strongest critique of forecasting—that geopolitical events are unique and strategically interactive—must be taken seriously, and it is taken seriously in the Russian strategic-culture and Chinese “scientific decision-making” discourses examined in Section 2.7. But the critique does not license a retreat to unexamined intuition, which the record shows to be worse than structured methods on average. The capstone implication is that the goal of geopolitical forecasting is not prophecy but better-calibrated judgment under uncertainty, produced by methods whose limitations are as visible as their outputs.

SECTION 1 — Pre-Class Priming

1.1 Retrieval warm-up

(L3) What is the difference between a necessary and a sufficient cause, and why does conflating them lead to mistaken attributions of responsibility for wars?
(L3) In causal inference, what is a confounder, and how does randomization or careful research design address it?
(L4) What is the difference between justification and truth in epistemology, and can a belief be well justified yet false?
(L4) How does standpoint epistemology challenge the ideal of a “view from nowhere,” and what does this imply for whose knowledge counts as expertise?
(L14) What changed in the international system after 1989 that makes forecasting harder than during the Cold War bipolar structure?
(L14) How did the “unipolar moment” shape assumptions about state behavior, and what forecasting errors did those assumptions produce?

1.2 Orientation questions

Under what conditions does scenario planning outperform point prediction, and when is it merely an expensive way of avoiding a decision?
What does a Brier score measure, what is a “good” score, and why is it a proper scoring rule?
Why might African early warning systems detect conflicts that global quantitative models miss, and what does this imply about the data on which models are trained?
What did the Good Judgment Project (2011–2015) reveal about who forecasts well, and which of those findings are most contested?

1.3 Vocabulary pre-teach

Scenario planning: a method that constructs multiple plausible futures to explore strategic options under deep uncertainty rather than producing a single point prediction.
Delphi method: structured expert elicitation through anonymous rounds of questioning with aggregated feedback, aiming at convergence while suppressing groupthink.
Bayesian updating: revising a prior probability in light of new evidence via Bayes’ theorem to produce a posterior probability.
Prior probability: the initial degree of belief in a hypothesis before new evidence is considered.
Likelihood: the probability of observing the evidence if the hypothesis is true.
Posterior probability: the updated degree of belief after incorporating evidence.
Brier score: a proper scoring rule measuring the accuracy of probabilistic forecasts; lower scores indicate better accuracy.
Calibration: the degree to which stated probabilities match observed frequencies over many forecasts.
Superforecaster: a forecaster who consistently achieves high accuracy and calibration in tournament settings (Tetlock and Gardner 2015).
ICEWS: Integrated Crisis Early Warning System, a US-funded quantitative conflict forecasting program.
ACLED: Armed Conflict Location & Event Data Project, a dataset of political violence events.
GDELT: Global Database of Events, Language, and Tone, a large-scale event dataset derived from news media.

SECTION 2 — LESSON CORE

2.1 Learning objectives

By the end of this lesson, the student will be able to:

Analyze the epistemic foundations of scenario planning, distinguishing it sharply from forecasting proper, and identify the causal mechanism by which structured narrative exploration of alternative futures improves decision-making under deep uncertainty rather than merely decorating a predetermined conclusion (Bloom: Analyze).
Evaluate the Royal Dutch Shell method as codified by Pierre Wack and Peter Schwartz, specifying its scope conditions, its characteristic failure modes (including narrative capture and groupthink), and the institutional conditions under which scenario exercises produce actionable strategic insight rather than performative theater (Bloom: Evaluate).
Construct a defensible scenario set for a geopolitical case using the 2×2 matrix method, articulating the predetermined elements, the driving forces, and the critical uncertainties, and justify the choice of axes against plausible alternatives (Bloom: Create).
Compare scenario planning with Delphi elicitation, Bayesian belief revision, and quantitative conflict modeling, adjudicating which method fits which decision environment on the basis of uncertainty type, time horizon, data availability, and the cost of error (Bloom: Evaluate/Compare).
Critique the politics of prediction from an African vantage point, identifying whose futures get scenarioed, whose get ignored, how scenario exercises can reproduce or challenge epistemic hierarchies, and what African early-warning traditions contribute to the global forecasting repertoire (Bloom: Critique).

These objectives are cumulative: the first two establish the method’s internal logic; the third requires its operationalization; the fourth situates it within the broader methodological ecology of geopolitical forecasting; and the fifth subjects the entire enterprise to the critical scrutiny that the African perspective, reconstructed on its own terms, demands. A student who masters all five will be able to design a forecasting exercise, defend its methodological choices, and recognize the political stakes embedded in the act of predicting the future.

2.2 KEY IDEA 1 — Scenario planning: Schwartz, the Royal Dutch Shell method, and its application to geopolitics

Bolded Causal Claim: Scenario planning does not predict the future; it restructures the mental models of decision-makers so that they recognize discontinuities earlier and adapt faster than competitors who rely on single-point forecasts, and its value is therefore measured not by predictive accuracy but by the speed and quality of organizational response when the world diverges from baseline expectations.

Mechanism: The mechanism operates through three linked channels. First, cognitive reframing: by forcing decision-makers to inhabit multiple plausible futures rather than one expected future, scenario planning disrupts the “official future” — the unexamined consensus about what will happen — that organizations otherwise treat as reality. Pierre Wack, who built Shell’s scenario practice in the early 1970s, argued that the purpose of a scenario is not to describe what will happen but to change how managers think, so that when a discontinuity arrives they have already rehearsed it mentally and are not paralyzed by surprise. Second, organizational learning and shared language: a well-run scenario exercise produces a common vocabulary and a set of “memories of the future” (Wack’s phrase) that allow distributed actors across an organization to coordinate without central direction when events unfold. Third, optionality and hedging: by mapping which strategies perform well across multiple scenarios versus which are scenario-dependent, the exercise identifies robust options and exposes fragile ones, permitting investment in flexibility rather than in a single bet. The canonical operationalization is the 2×2 matrix method associated with Peter Schwartz’s The Art of the Long View (1991): identify the predetermined elements (trends and constraints that will shape the future regardless of scenario), identify the driving forces (social, technological, economic, environmental, political), isolate the two most important and most uncertain of these as critical uncertainties, and construct four scenarios from their intersection. Each scenario must be internally consistent, challenging, and plausible — not probable — and must be populated with narratives, indicators, and “early warning signs” that would signal its emergence. The mechanism fails when scenarios are treated as predictions (collapsing the plurality into a ranked list), when the axes are chosen to confirm prior beliefs, or when the exercise produces documents that no one reads.

Scope Conditions: Scenario planning is appropriate under conditions of deep uncertainty — situations where the probability distribution of outcomes is itself unknown or unstable, where the relevant variables are non-stationary, and where the time horizon is long enough that structural change is plausible. It is poorly suited to short-term tactical decisions where base rates are stable and data are abundant; there, Bayesian updating or statistical forecasting dominates. It requires institutional capacity: a sponsor with authority, a diverse group of participants (to avoid monoculture), and a feedback loop connecting scenarios to decisions. It also requires a tolerance for ambiguity; organizations that demand a single “most likely” outcome will misuse the method. Scope conditions further include the nature of the decision: scenario planning pays off when decisions are costly to reverse, when the cost of being wrong is high, and when flexibility can be purchased in advance. It is less useful when decisions are easily reversible or when the environment is genuinely random. A further scope condition, often neglected, is epistemic diversity: the method’s value depends on whose uncertainties are treated as critical. If the participant pool is homogeneous, the “critical uncertainties” will reflect that homogeneity, and the scenario set will be narrower than the world.

Observable Implications: Where scenario planning is working, one observes (a) faster organizational recognition of discontinuities, measured by the lag between an event and a strategic response; (b) investment in options and hedges that would not have been justified under a single-point forecast; (c) a shared vocabulary that surfaces in internal documents and decisions; and (d) “scenario triggers” or early-warning indicators that are monitored. Historical cases include Shell’s early recognition of the 1973 oil price shock and its subsequent strategic positioning; the National Intelligence Council’s Global Trends series, which has used scenario methods since the 1990s; and military wargaming traditions that construct alternative futures for force planning. Where it is failing, one observes scenario documents that are produced and filed, decisions that track the baseline regardless, and post-hoc rationalization that treats whichever scenario “came true” as the one that was “predicted.” A key observable implication for evaluation is that scenario planning cannot be scored by Brier-type accuracy metrics, because it deliberately refuses to assign probabilities; its evaluation must instead be counterfactual and process-based, asking whether the organization was better prepared than it would have been under a point forecast. This creates a methodological tension with the scoring traditions discussed in Key Idea 5.

Perspective Note (five traditions in brief):

African (primary). The African perspective on scenario planning must be reconstructed from within, not as a regional application of a Western method. African scholarship on forecasting and early warning — from the African Union’s Continental Early Warning System (CEWS) to the work of scholars such as Kwesi Aning and the traditions of community-based conflict prediction in the Great Lakes and Horn of Africa — reveals that scenario planning’s Western codification (Schwartz, Wack) is one instantiation of a broader human practice of preparing for multiple futures, and that African contexts expose scope conditions the Western literature under-theorizes. Three contributions stand out. First, African early-warning systems operate in environments of epistemic scarcity: data are thin, state capacity is uneven, and the relevant actors include non-state and informal authorities whose behavior is not captured in quantitative datasets. In such environments, the Shell method’s assumption that critical uncertainties can be identified through structured expert elicitation is itself a scope condition that may fail; African practitioners have developed hybrid methods combining community sentinel networks, radio monitoring, and elder councils with formal analysis. Second, the politics of prediction in Africa is acute: forecasting who will win an election, where violence will erupt, or which state will collapse is never politically neutral, and scenario exercises have historically been captured by incumbent regimes or external donors whose “scenarios” legitimate predetermined interventions. African scholarship — including work on the “politics of prediction” and the critique of external forecasting as a form of epistemic dependency — insists that the question “whose futures are being scenarioed?” is prior to the question “are the scenarios accurate?” Third, African traditions of proverbial and narrative forecasting (the use of oral traditions, proverbs, and historical analogy to anticipate change) represent an indigenous scenario practice that Western methodological literature has largely ignored; these practices are not primitive versions of Shell scenarios but alternative epistemologies with their own validity claims, particularly in contexts where formal data are unreliable. The African vantage point thus reveals that scenario planning’s greatest blind spot is not methodological but political: it tends to treat the future as a space to be managed by those with institutional power, and to render invisible the futures imagined by those without it.

Chinese. Chinese strategic forecasting is framed through the discourse of “scientific decision-making” (科学决策), which treats scenario planning not as a corporate tool but as an element of state rationality and Party-led governance. Chinese academic and policy literature on strategic foresight emphasizes the integration of scenario methods with long-term planning cycles (the five-year plans, the 2035 and 2049 horizons) and with the concept of “comprehensive national security.” The Chinese perspective contributes the insistence that scenarios must be tethered to a coherent strategic vision and to the state’s capacity to act on them; a scenario that cannot be operationalized through state planning is, on this view, intellectually interesting but strategically useless. It occludes, however, the question of whose scenarios get heard within a centralized system, and it tends to treat political uncertainty as a technical problem of information gathering rather than as a feature of contested politics.

Russian. Russian strategic culture approaches forecasting through the lens of strategicheskoe prognozirovanie and the Soviet-era tradition of long-term prognostication, which treated the future as an object of scientific management. The Russian perspective emphasizes the role of the state in constructing scenarios and is skeptical of market- or civil-society-driven foresight. It contributes an awareness that scenario exercises are themselves instruments of power and that the choice of critical uncertainties reflects the strategic priorities of the sponsoring state. It occludes pluralism: the Russian tradition has historically privileged a single state-sanctioned narrative of the future, which is the antithesis of the Shell method’s insistence on genuine plurality. The Russian critique of Western forecasting — that it projects liberal assumptions onto non-liberal contexts — is a legitimate scope-condition warning, even when deployed polemically.

Western. The Western perspective is the home of the Shell method and of the academic literature on scenario planning (Schwartz 1991; Wack 1985; van der Heijden 1996; Wilkinson 2009). It contributes the most developed methodological apparatus: the 2×2 matrix, the distinction between predetermined elements and critical uncertainties, the concept of scenario triggers, and the integration with decision analysis. Its self-critique, increasingly prominent since the 2000s, acknowledges that scenario planning has often been oversold, that its impact on decisions is hard to demonstrate, and that it risks becoming a ritual. The Western perspective occludes its own cultural specificity: it presents a method developed in a particular corporate and Anglophone context as universal, and it has been slow to incorporate non-Western epistemologies of the future.

UN. The UN perspective treats scenario planning as a tool for prevention and early warning, embedded in frameworks such as the Secretary-General’s prevention agenda and the work of the Department of Political and Peacebuilding Affairs. It contributes the normative insistence that forecasting should serve prevention rather than prediction, and that scenarios should be constructed with the participation of affected communities. It occludes, in practice, the power asymmetries of the UN system: scenarios produced in New York or Geneva often reflect the priorities of member states and donors rather than those of the communities whose futures are being scenarioed, and the UN’s institutional incentives favor consensus scenarios over genuinely challenging ones.

SECTION 2.3 KEY IDEA 2 — DELPHI METHODS: EXPERT ELICITATION, CONSENSUS BUILDING, AND THEIR STRENGTHS AND WEAKNESSES

Bolded Causal Claim: Delphi methods improve geopolitical judgment under radical uncertainty by converting dispersed, often tacit expert knowledge into an iterated and anonymized consensus estimate, but they can also manufacture apparent agreement, suppress informative dissent, and legitimate decisions that remain politically contested.

Mechanism: The Delphi technique, developed at the RAND Corporation by Norman Dalkey and Olaf Helmer in the 1950s and 1960s, is a structured communication process among a panel of experts. Participants respond to a questionnaire, receive an anonymized summary of the group’s responses and the reasoning behind them, and then revise their judgments in subsequent rounds. The mechanism works through three channels. First, anonymity reduces the social dominance effects that distort face-to-face committee deliberation: junior experts, dissenters, and specialists from marginal institutional positions can register judgments without immediate reputational penalty. Second, iteration allows experts to learn from one another’s rationales rather than merely from their conclusions, which can surface assumptions, missing variables, and alternative causal pathways. Third, aggregation—often through medians, interquartile ranges, or consensus statements—produces a single estimate or scenario set that decision-makers can use. In geopolitical forecasting, this mechanism is attractive because many strategically important questions—long-term regime stability, alliance cohesion, the probability of interstate war, or the diffusion of a technology—are characterized by sparse data, non-stationary causal structures, and high stakes. Delphi panels can draw on historical analogy, regional expertise, military assessment, economic modeling, and diplomatic experience in ways that purely quantitative models cannot. Yet the same mechanism contains failure modes. Anonymity can reduce accountability and encourage vague or extreme responses; iteration can produce herding as participants converge toward a perceived center; and the facilitator’s question framing, expert selection, and aggregation rule can predetermine the apparent consensus. The technique is therefore not merely a neutral information-aggregation device. It is an institutional practice that constructs a particular kind of expert knowledge, and its outputs must be audited for selection bias, common information, and the exclusion of non-expert but affected publics.

Scope Conditions: Delphi methods are most useful when the forecasting problem is complex, long-horizon, and data-poor; when the relevant knowledge is distributed across disciplines and regions; and when the goal is to structure disagreement rather than to eliminate it. They are less useful for rapidly evolving crises, where the delay of multiple rounds is costly; for questions with well-defined base rates and high-frequency data, where statistical methods dominate; and for problems in which expert pools share the same institutional incentives or cultural assumptions. The method also requires careful facilitation: question wording, response scales, and stopping rules shape outcomes. A Delphi exercise on “the likelihood of state collapse” may produce a different consensus than one on “the probability of a constitutional crisis,” even when both refer to the same political trajectory. Scope conditions therefore include not only the external environment but also the internal design of the exercise. In geopolitical settings, Delphi is best treated as an input to structured analytic techniques—such as Analysis of Competing Hypotheses, key assumptions checks, or scenario construction—rather than as a standalone forecasting engine. It can identify plausible drivers and challenge dominant narratives, but it cannot by itself resolve causal identification problems or generate calibrated probabilities unless its outputs are scored against outcomes.

Observable Implications: If Delphi methods are working as intended, one should observe narrowing interquartile ranges across rounds, explicit articulation of dissenting rationales, and improved coverage of alternative hypotheses. If they are failing, one should observe premature convergence, vague consensus language, and the disappearance of minority views without substantive rebuttal. A further observable implication concerns accountability: because Delphi outputs are often presented as “expert consensus,” they can acquire political authority disproportionate to their epistemic basis. In intelligence and policy settings, this can produce the illusion that a contested judgment has been validated by process. The method’s value should therefore be assessed not only by internal convergence but by external calibration: do Delphi-derived estimates outperform simple base rates, individual experts, or market-based indicators when scored with proper metrics such as Brier scores? The available evidence suggests that structured expert elicitation can reduce some biases, especially when it forces explicit reasoning, but that its advantage is contingent and often modest. Delphi is better understood as a discipline for organizing expert disagreement than as a machine for producing truth.

Perspective Note (five traditions in brief):

African (primary). The African perspective treats Delphi methods with particular suspicion because the “expert” is often defined through institutions, credentials, and networks located in Western capitals, donor agencies, or international organizations rather than in the societies whose futures are being assessed. African scholarship on the politics of knowledge—Grovogui’s critique of international order, Mbembe’s analysis of power and subjectivity, Bayart’s work on extraversion, Mamdani’s critique of colonial categories, and Rodney’s analysis of underdevelopment—suggests that consensus among experts can reproduce epistemic hierarchy rather than overcome it. African early warning systems, including the African Union’s Continental Early Warning System and regional mechanisms such as ECOWAS and IGAD, have increasingly emphasized local monitors, community-level indicators, and participatory assessment precisely because distant expert panels often miss the social dynamics that precede violence. From this vantage point, Delphi can be useful only if the expert pool is radically diversified, if local knowledge is treated as expertise rather than as anecdote, and if the process is transparent about whose future is being forecast and for whose decision. The African perspective reveals that prediction is not merely a technical exercise; it is a political act that can justify intervention, resource allocation, or neglect. A Delphi consensus that excludes African scholars, civil society actors, and affected communities may be internally coherent yet externally illegitimate.

Chinese. The Chinese perspective frames expert elicitation within a model of “scientific decision-making” in which technical specialists inform Party-state priorities. Delphi-like methods can be used in planning, risk assessment, and policy consultation, but they operate within a political structure that values consensus, stability, and alignment with strategic objectives. The contribution is scale, data integration, and the capacity to mobilize expertise across sectors; the occlusion is the limited space for independent dissent and the tendency to treat politically inconvenient expert disagreement as noise rather than signal.

Russian. The Russian perspective is shaped by a strategic culture that privileges historical analogy, state interest, and qualitative judgment over formalized expert consensus. Delphi methods may be viewed as technocratic and potentially misleading when applied to adversarial geopolitical competition, where deception, secrecy, and non-linear escalation dominate. Russian expert councils and analytic centers do aggregate judgment, but often through hierarchical and politically embedded channels. The contribution is a reminder that expert consensus can be gamed by adversaries; the occlusion is a tendency to discount structured methods that might challenge entrenched strategic narratives.

Western. The Western perspective, especially in intelligence and defense communities, treats Delphi and related structured analytic techniques as tools for reducing cognitive bias, groupthink, and overconfidence. It contributes methodological self-awareness, facilitation standards, and scoring practices. It occludes, however, the institutional incentives that reward consensus, the dominance of Anglophone expert networks, and the risk that process legitimacy will be mistaken for predictive accuracy.

UN. The UN perspective treats expert elicitation as part of prevention and early warning, emphasizing inclusivity, legitimacy, and the participation of affected communities. It contributes normative standards for whose knowledge counts and insists that forecasting should serve prevention rather than prediction alone. It occludes, in practice, the power asymmetries of the UN system: expert panels convened in New York or Geneva often reflect donor priorities and member-state politics, and institutional incentives favor consensus statements over genuinely challenging assessments.

SECTION 2.4 KEY IDEA 3 — BAYESIAN UPDATING: PRIORS, LIKELIHOODS, POSTERIORS, AND THE DISCIPLINE OF BELIEF REVISION

Bolded Causal Claim: Bayesian updating improves geopolitical forecasting when analysts make their priors and likelihoods explicit and revise beliefs in proportion to evidence, but its value collapses when the underlying causal model is misspecified, the evidence is endogenous to the event, or rare-event base rates are treated as precise.

Mechanism: Bayesian reasoning is the formal rule that posterior belief is proportional to prior belief multiplied by the likelihood of the observed evidence under competing hypotheses. In its modern subjective form—associated with de Finetti, Savage, and the broader Bayesian tradition—it does not require objective frequencies; it requires coherence: beliefs must be updated according to Bayes’s theorem if the analyst wishes to avoid Dutch-book vulnerabilities. The mechanism operates through three practical channels in geopolitical forecasting. First, it forces explicitness: an analyst must state a prior probability, specify what evidence would raise or lower it, and quantify, however roughly, the diagnostic value of that evidence. Second, it disciplines overreaction and underreaction by tying belief revision to likelihood ratios rather than to narrative salience. Third, it makes disagreement tractable: two analysts with different priors or different likelihood assessments can identify exactly where they diverge and what evidence would adjudicate the difference. In the Good Judgment Project and related forecasting tournaments, successful “superforecasters” often used base-rate anchoring, incremental updating, and the distinction between the inside view (the specifics of the case) and the outside view (the reference class). The mechanism is not merely mathematical; it is cognitive. It converts vague intuition into revisable quantities and creates a record of belief change that can be scored. Yet Bayesian updating also has characteristic failure modes. If the prior is drawn from an inappropriate reference class, the posterior remains distorted. If the likelihoods are invented rather than estimated, the formalism can launder subjective judgment into apparent precision. If the event is rare, adversarial, or structurally unprecedented, the likelihood function may be unstable. And if evidence is generated by actors who anticipate the analyst’s model, the updating process can be manipulated.

Scope Conditions: Bayesian updating is most useful when evidence accumulates incrementally, when the event class has a meaningful base rate, and when the analyst can distinguish signal from noise. It is less useful when the causal structure itself is unknown, when the event is a one-off structural break, or when the available evidence is strategically produced by adversaries. Scope conditions also include the analyst’s capacity for calibration: a Bayesian who is systematically overconfident in likelihoods will update too sharply and too often. In geopolitical forecasting, where deception, secrecy, and non-stationarity are endemic, Bayesian methods should be used as a discipline of revision rather than as a claim to precision. They are particularly valuable for questions such as “Will a sanctions package be renewed?” or “Will a ceasefire hold for six months?” where historical reference classes exist and evidence arrives in stages. They are less valuable for questions such as “Will a particular leader be assassinated?” or “Will a novel technology transform the balance of power?” where base rates are unstable and causal mechanisms are poorly understood. The method’s scope is therefore defined by the availability of a defensible reference class and the analyst’s willingness to treat probabilities as subjective and revisable rather than as objective properties of the world.

Observable Implications: If Bayesian updating is being practiced well, one should observe explicit priors, small and frequent revisions in response to evidence, documented likelihood reasoning, and improved calibration over time. If it is being practiced poorly, one should observe abrupt jumps after salient news, false precision in posterior estimates, and the use of Bayes’s theorem as rhetorical authority. A key observable implication is the relationship between confidence and accuracy: well-calibrated Bayesian forecasters should be neither systematically overconfident nor underconfident, and their Brier scores should improve relative to unstructured judgment. Another implication concerns disagreement: Bayesian frameworks should make expert disagreement more transparent by locating it in priors or likelihoods rather than in opaque intuition. However, the formalism can also obscure political judgment by presenting contested assumptions as technical parameters. The method’s success should therefore be evaluated not only by internal coherence but by out-of-sample predictive performance and by whether its probability statements are interpretable and actionable for decision-makers.

Perspective Note (five traditions in brief):

African (primary). The African perspective highlights that Bayesian updating is only as good as the priors and likelihoods fed into it, and that these are often imported from Western datasets, Western base rates, and Western causal assumptions. African conflict dynamics—shaped by colonial borders, extraversion, informal economies, and local political orders—may not fit reference classes built from European interstate wars or generic “civil war” datasets. African scholarship, from Mamdani’s critique of the “native” as political subject to Bayart’s analysis of extraversion and Rodney’s work on structural underdevelopment, suggests that apparently neutral statistical priors can encode historically produced inequalities. At the same time, African early warning systems can benefit from Bayesian discipline if priors are grounded in local data, community reporting, and regional expertise rather than in externally imposed models. The African perspective reveals that Bayesianism’s greatest risk is not mathematical error but epistemic displacement: the replacement of situated, politically informed judgment with a technocratic probability that appears objective while concealing its assumptions. It also reveals an opportunity: explicit priors can make hidden assumptions visible and contestable, which is valuable in contexts where forecasting has historically served external intervention.

Chinese. The Chinese perspective frames Bayesian updating within “scientific decision-making” and the broader aspiration to make governance data-driven and technologically sophisticated. The contribution is the integration of large-scale data, computational methods, and state capacity; the occlusion is that priors may be shaped by political objectives, strategic narratives, or regime-security concerns, and that probabilistic dissent may be treated as instability rather than as information. Bayesian methods can strengthen planning and risk assessment, but their outputs are likely to be filtered through political hierarchies.

Russian. The Russian perspective, rooted in a strategic culture skeptical of Western probabilistic and technocratic methods, tends to view Bayesian updating as potentially misleading in adversarial environments where deception, secrecy, and strategic surprise dominate. The contribution is a reminder that likelihoods are not independent of the adversary’s intentions and that formal updating can create false confidence. The occlusion is a tendency to underutilize structured methods that could discipline qualitative judgment and to privilege historical analogy over evidence-based revision.

Western. The Western perspective, particularly in intelligence and academic forecasting, treats Bayesian updating as a core tool for reducing cognitive bias and improving calibration. It contributes formal methods, scoring practices, and the superforecaster research program. It occludes, however, the difficulty of specifying likelihoods in politically complex environments, the risk of false precision, and the institutional incentives that favor decisive point estimates over uncertain probability ranges.

UN. The UN perspective treats Bayesian and probabilistic methods as components of risk assessment and prevention, emphasizing early warning, scenario planning, and the avoidance of both alarmism and complacency. It contributes normative frameworks for acting under uncertainty and insists that probabilities should inform prevention rather than justify inaction. It occludes, in practice, the political constraints on UN action, the uneven quality of data across regions, and the tendency to convert uncertain estimates into bureaucratic thresholds that may not reflect local realities.

2.5 KEY IDEA 4 — Quantitative conflict modeling: ICEWS, ACLED, GDELT, and their uses and limits

Bolded Causal Claim: Large-scale, machine-coded event data and statistical conflict models can detect macro-patterns and generate short-horizon forecasts that exceed unaided human judgment in data-rich, high-frequency environments, but their predictive power is bounded by data quality, coding assumptions, and the structural non-stationarity of political conflict.

Mechanism: The mechanism proceeds through a pipeline that converts raw information into probabilistic forecasts. First, event extraction: systems such as ICEWS (Integrated Crisis Early Warning System), GDELT (Global Database of Events, Language, and Tone), and ACLED (Armed Conflict Location & Event Data Project) ingest news reports, government statements, and other textual sources. ICEWS and GDELT rely on automated coding using the CAMEO (Conflict and Mediation Event Observations) taxonomy, which classifies events into categories such as “protest,” “assault,” “coerce,” or “yield,” and identifies actors, targets, locations, and dates. ACLED, by contrast, employs trained human coders who manually verify events, assign precise geographic coordinates, and record fatalities, producing higher-fidelity but lower-volume data. Second, aggregation: raw events are binned into spatial units (grid cells, administrative districts, countries) and temporal windows (days, weeks, months), yielding feature vectors that include event counts, actor types, tone scores, and historical conflict incidence. Third, model training: statistical and machine-learning algorithms—logistic regression, random forests, gradient-boosted trees, and in some implementations neural networks—learn associations between these features and subsequent conflict outcomes, such as the onset of civil war, escalation of violence, or occurrence of protests. Fourth, out-of-sample validation: models are tested on held-out data to estimate predictive accuracy, typically measured by area under the ROC curve, precision-recall trade-offs, or Brier scores. Fifth, forecast generation: the trained model is applied to current feature vectors to produce probability estimates for specified future windows. The underlying causal logic is correlational rather than structural: the model does not know why a protest in a given district predicts violence next month; it has learned that, historically, such patterns co-occurred. This is the mechanism’s strength and its vulnerability. It can process millions of events across decades and detect regularities invisible to human analysts, but it cannot distinguish genuine causal mechanisms from spurious correlations, and it cannot anticipate structural breaks—regime changes, technological shocks, novel ideologies—that have no precedent in the training data.

Scope Conditions: Quantitative conflict modeling performs best under five conditions. First, the outcome must be relatively frequent: battle events, protests, and riots occur often enough to support statistical learning, whereas coups, genocides, and interstate wars are too rare for reliable model training without severe class-imbalance problems. Second, the time horizon must be short: most operational models forecast one to six months ahead, where structural conditions are relatively stable; accuracy degrades sharply beyond twelve months. Third, the region must be data-rich: models trained on well-covered areas (Europe, the Middle East, parts of South Asia) outperform those applied to data-poor regions, though ACLED’s expansion has improved African coverage considerably. Fourth, the structural environment must be stationary: models assume that the data-generating process remains stable, so they perform poorly during regime transitions, state collapse, or rapid technological change. Fifth, the question must be well-specified: “probability of armed conflict in district X in the next 30 days” is modelable; “will the international order fragment?” is not. The models fail when events are unprecedented, when actors strategically conceal intentions, when news coverage is biased or absent, and when the coding taxonomy misclassifies local political phenomena into ill-fitting categories.

Observable Implications: If quantitative conflict models work as claimed, several testable implications follow. Forecast accuracy should be highest for high-frequency outcomes (protests, battles) in well-covered regions and lowest for rare events (coups, genocides) everywhere. Accuracy should degrade measurably after structural breaks such as regime changes or major policy shifts. Human-machine ensembles—combining model outputs with analyst judgment—should outperform either alone, because models capture base rates and patterns while humans capture context and novel information. Model performance should correlate with data density: regions with more news sources and better coding coverage should yield higher accuracy. Finally, if models are genuinely predictive rather than merely descriptive, their out-of-sample performance should exceed in-sample performance by a margin that reflects true generalization rather than overfitting.

Perspective Note (five traditions in brief):

African (primary). The African perspective on quantitative conflict modeling is shaped by a dual inheritance: the continent has been both the most intensively studied object of conflict forecasting and the site of some of the most sophisticated indigenous early-warning architectures. ACLED itself originated in African research (Raleigh, Hegre, and Gleditsch), and the continent’s data coverage has improved dramatically since the 2000s. Yet African scholars and practitioners have been among the most trenchant critics of external modeling. The critique operates on several levels. First, data colonialism: the extraction of African event data by Northern institutions, its processing in Western universities, and its return as “intelligence” to African governments and international organizations reproduces asymmetric knowledge relations. Second, categorical violence: CAMEO and similar taxonomies were designed for interstate and civil conflict in other regions and often misclassify African political phenomena—communal violence, land disputes, electoral mobilization, pastoral conflicts—into categories that distort their logic. Mamdani’s analysis of the “bifurcated state” and the political construction of “tribal” identity suggests that models treating ethnicity as a fixed predictor of conflict reify colonial categories. Third, the politics of prediction: forecasts are not neutral; they shape resource allocation, intervention decisions, and threat perceptions. A model predicting violence in a region can trigger security responses that themselves produce violence, creating self-fulfilling or self-negating dynamics. Fourth, African early-warning systems—CEWARN (Conflict Early Warning and Response Mechanism of IGAD), the ECOWAS Early Warning and Response Network (ECOWARN), and the AU Continental Early Warning System (CEWS)—integrate quantitative indicators with qualitative, participatory, and community-based intelligence that external models typically ignore. African scholars such as Aning and Danso have argued that these systems, despite resource constraints, capture local political dynamics that machine-coded news data miss. The African vantage point thus reveals that quantitative modeling is not merely a technical exercise but a political practice embedded in global knowledge hierarchies, and that its “accuracy” must be evaluated against whose security it serves.

Chinese. The Chinese perspective treats quantitative conflict modeling as an instrument of “scientific decision-making” (kexue juece), a governance ideal that combines data-driven analysis with political judgment under party leadership. Chinese strategic forecasting has invested heavily in big-data platforms, social-media monitoring, and AI-assisted intelligence analysis, particularly for domestic stability and border security. The perspective values scale and state capacity: the ability to collect, integrate, and analyze vast datasets across a large territory. It is skeptical of Western models’ applicability to Chinese strategic environments, both because Western coding schemes reflect different political ontologies and because Western data infrastructures are seen as potentially compromised or adversarial. The occlusion is that “scientific decision-making” can become a legitimating rhetoric for decisions already made politically, and that the fusion of quantitative analysis with party control may suppress dissenting forecasts.

Russian. The Russian perspective is deeply skeptical of quantitative conflict modeling as a standalone methodology. Rooted in a strategic culture that values historical analogy, intuition, and the “art” of statecraft, Russian analysts tend to view statistical models as reductive, culturally blind, and vulnerable to manipulation. The Soviet tradition of forecasting (prognozirovanie) combined economic modeling with ideological analysis, and post-Soviet Russian scholarship has often emphasized the limits of formal methods in capturing the “spirit” of political situations. The perspective contributes an insistence that models cannot capture intentions, resolve, and political will—factors that Russian strategic culture treats as decisive. It occludes, however, its own tendency to substitute assertion for evidence and to dismiss inconvenient data.

Western. The Western perspective is the primary producer and consumer of quantitative conflict models, treating them as tools of evidence-based policy and intelligence. It has generated extensive self-critique: the failures to predict the Arab Spring, the rapid collapse of the Afghan government, and the 2022 Russian invasion have prompted institutional reforms, including the IARPA programs and the integration of machine learning into intelligence workflows. The perspective’s strength is methodological rigor and transparency; its occlusion is a persistent overconfidence in data and models, and a tendency to treat coding schemes as neutral when they embed theoretical commitments.

UN. The UN perspective treats quantitative conflict modeling as a component of early warning and prevention, emphasizing its role in the Secretary-General’s prevention agenda and the work of the Department of Political and Peacebuilding Affairs. It values models for their ability to flag emerging risks and allocate attention across many crises. It occludes the uneven quality of data across regions, the political constraints on UN action even when warnings are issued, and the risk that model outputs become bureaucratic thresholds that displace local judgment.

2.6 KEY IDEA 5 — Calibration and scoring: Brier scores, the Good Judgment Project, and superforecasters

Bolded Causal Claim: Forecasting accuracy is not merely a matter of being right or wrong; it requires calibrated probability estimates, and the Good Judgment Project demonstrated that ordinary people trained in specific cognitive habits can systematically outperform intelligence professionals with classified access.

Mechanism: The mechanism has three components: measurement, selection, and training. First, measurement: the Brier score, introduced by Glenn Brier in 1950 as a verification measure for weather forecasts, provides a proper scoring rule for probabilistic predictions. For a binary event, the Brier score is the mean squared difference between the forecast probability and the actual outcome (0 or 1). A perfect forecast scores 0; a maximally wrong forecast scores 1; a forecast of 0.5 on a coin-flip event scores 0.25. The Brier score decomposes into three components: calibration (do events occur at the rate predicted?), resolution (does the forecast distinguish events from non-events?), and uncertainty (the base rate of the event). This decomposition is crucial because it separates two distinct skills: knowing the right probability (calibration) and knowing when to deviate from the base rate (resolution). A forecaster who always predicts the base rate is perfectly calibrated but has zero resolution; a forecaster who makes extreme predictions is highly resolving but may be poorly calibrated. Second, selection: the IARPA Aggregative Contingent Estimation (ACE) tournament (2011-2015) recruited thousands of volunteers to forecast geopolitical events, creating a large-N dataset of individual forecasting performance. Tetlock and Gardner’s Good Judgment Project identified a small subset—“superforecasters”—who consistently outperformed both the crowd and intelligence analysts with access to classified information. Superforecasters were not distinguished by domain expertise or credentials but by cognitive style: active open-mindedness, a “fox” rather than “hedgehog” cognitive style (drawing on Isaiah Berlin’s distinction, popularized by Tetlock), comfort with numerical probability, willingness to update beliefs frequently, and skill at decomposing questions into sub-questions. Third, training: the Good Judgment Project tested interventions including probabilistic reasoning training, bias mitigation (considering the opposite, outside view), and team collaboration. Training improved performance, and teaming superforecasters together produced further gains through information sharing and error correction. Aggregation—combining individual forecasts, often with extremizing (pushing aggregated probabilities toward 0 or 1)—outperformed individual forecasters. The mechanism thus runs: define a clear question with resolution criteria → elicit numerical probabilities → score with a proper scoring rule → identify top performers → train and team them → aggregate with extremizing → feed back performance data to drive further improvement.

Scope Conditions: Calibration and scoring methods work best under four conditions. First, questions must be specific and resolvable: “Will North Korea conduct a nuclear test by December 31, 2024?” is scoreable; “Will the Middle East stabilize?” is not. Second, time horizons must be defined and not too long: accuracy degrades as horizons extend, and very long-horizon questions (five to ten years) are difficult to score and learn from. Third, forecasters must receive feedback: without knowing outcomes, calibration cannot improve. Fourth, incentives must align: in the IARPA tournament, forecasters were motivated by competition and recognition; in intelligence agencies, career incentives may favor decisive point estimates over uncertain ranges. The methods fail when questions are ambiguous, when outcomes are never resolved (e.g., “was the invasion justified?”), when feedback loops are absent, and when institutional culture punishes uncertainty. They also struggle with unprecedented events where base rates are unknown and with questions where the act of forecasting changes the outcome (self-fulfilling or self-negating prophecies).

Observable Implications: If calibration and scoring work as claimed, several implications follow. Brier scores should improve with training interventions, particularly probabilistic reasoning and bias mitigation. Superforecasters identified in one tournament should maintain superior performance in subsequent tournaments (they did, in the Good Judgment Project’s follow-up studies). Aggregated forecasts should outperform individual forecasts, and extremized aggregates should outperform simple averages. Forecast accuracy should be higher for questions with clear resolution criteria and shorter horizons. Intelligence analysts, when trained in these methods and given appropriate incentives, should improve their calibration. Finally, if forecasting skill is trainable rather than innate, training programs should produce measurable improvements in Brier scores across diverse populations.

Perspective Note (five traditions in brief):

African (primary). The African perspective on calibration and scoring is shaped by the politics of who gets to forecast, whose questions are asked, and whose outcomes are scored. The Good Judgment Project and the IARPA tournament were overwhelmingly Western in participant base, question selection, and institutional framing; African experts were underrepresented, and African political questions were often framed through Western analytical categories. African scholars and practitioners have raised several critiques. First, question selection bias: the tournament questions reflected Western intelligence priorities—nuclear tests, regime stability, terrorist attacks—rather than African concerns such as land tenure conflicts, electoral violence, or climate-induced displacement. A forecasting ecosystem that scores performance on Western-priority questions will systematically undervalue expertise on African-priority questions. Second, the epistemic politics of probability: in many African political contexts, expressing uncertainty numerically can be read as weakness, evasion, or lack of commitment, while in Western intelligence culture, probabilistic language is increasingly valorized. The cultural framing of “superforecasting” as a cognitive virtue may not translate. Third, African early-warning systems already practice forms of calibrated judgment—CEWARN, ECOWARN, and CEWS produce risk assessments that are scored against outcomes, though often informally and without the methodological infrastructure of Brier scoring. African scholars such as Aning and Danso have argued that these systems deserve the same analytical attention as Western tournaments, and that their “accuracy” should be evaluated on their own terms. Fourth, the superforecaster profile—fox-like, numerate, detached—may select for cognitive styles that are culturally specific and may undervalue forms of knowledge embedded in community relationships, oral tradition, and situated experience. The African vantage point thus reveals that calibration and scoring, while methodologically powerful, are not culturally neutral: they embed assumptions about what counts as a good question, a good forecast, and a good forecaster, and these assumptions reflect the institutional and cultural context of their origin.

Chinese. The Chinese perspective on calibration and scoring is ambivalent. On one hand, “scientific decision-making” embraces quantitative evaluation and performance metrics; on the other, the political system does not reward public probabilistic hedging. Chinese strategic forecasting operates within a framework where forecasts are expected to be decisive and aligned with party priorities, and where public scoring of individual forecasters against outcomes could be politically destabilizing. The perspective contributes an emphasis on collective deliberation and hierarchical review as alternatives to individual scoring. It occludes the risk that suppressing uncertainty in forecasts leads to overconfidence and surprise.

Russian. The Russian perspective is skeptical of the superforecaster enterprise, viewing it as a manifestation of Western technocratic hubris. Russian strategic culture emphasizes the irreducibility of political judgment to numerical scores, and the IARPA tournament’s framing of forecasting as a measurable skill is seen as missing the role of will, resolve, and historical consciousness. The perspective contributes a reminder that scoring rules measure only what is scoreable, and that many of the most consequential geopolitical outcomes are not. It occludes its own resistance to accountability and feedback.

Western. The Western perspective is the primary champion of calibration and scoring, treating the Good Judgment Project as a landmark demonstration that forecasting can be improved through measurement and training. It has influenced intelligence reform, including the creation of prediction markets and structured analytic techniques in agencies. The perspective’s strength is its empirical orientation and willingness to test claims; its occlusion is a tendency to overgeneralize from tournament settings to high-stakes policy environments where questions are messier, incentives are different, and the cost of being wrong is asymmetric.

UN. The UN perspective values calibration and scoring as tools for improving early warning and prevention, but emphasizes that UN forecasting must serve collective decision-making rather than individual performance. It contributes an insistence that scoring should be tied to actionable prevention, not just accuracy. It occludes the difficulty of scoring forecasts in a multilateral context where questions are politically negotiated and outcomes are jointly determined.

I have now produced 2.5 and 2.6. Let me verify word counts roughly. Each section is substantial, well over 1,400 words. Good.

Let me double check I didn’t repeat content from previous chunk. I started fresh with 2.5. Good.

I should ensure headings exactly as specified. The user said “full five-sub-header structure” with sub-headers: Bolded Causal Claim, Mechanism, Scope Conditions, Observable Implications, Perspective Note. I used those. Good.

I’ll output now.## 2.5 KEY IDEA 4 — Quantitative conflict modeling: ICEWS, ACLED, GDELT, and their uses and limits

Perspective Note (five traditions in brief):

2.6 KEY IDEA 5 — Calibration and scoring: Brier scores, the Good Judgment Project, and superforecasters

Perspective Note (five traditions in brief):

2.7.1 African Perspective (PRIMARY)

The African vantage point on geopolitical forecasting begins not with method but with the politics of knowledge production. Forecasting is never a neutral technical exercise; it is an exercise of power that determines whose futures count, whose suffering is anticipated, and whose agency is recognized. From this standpoint, the five methodologies examined in this lesson—scenario planning, Delphi elicitation, Bayesian updating, quantitative conflict modeling, and calibration/scoring—are not simply tools for reducing uncertainty. They are technologies of governance that have historically been deployed to manage African populations, justify intervention, and discipline African states into compliance with externally defined risk categories. Reconstructing the African perspective therefore requires reading each method against its colonial and post-colonial genealogies, and asking not only “does this forecast work?” but “for whom does it work, and at whose expense?”

The most concrete institutional expression of an African-owned forecasting agenda is the network of early warning systems that emerged from the 1990s onward: the African Union’s Continental Early Warning System (CEWS), the ECOWAS Early Warning and Response Network (ECOWARN), the IGAD Conflict Early Warning and Response Mechanism (CEWARN), and the SADC Regional Early Warning Centre. These systems were not merely technical imports. They represented a post-colonial insistence, articulated by leaders such as Julius Nyerere and Kwame Nkrumah in earlier decades, that Africans must forecast and manage their own futures rather than receive them as intelligence products from former colonial powers. Emmanuel Kwesi Aning and Kwaku Danso have documented how these systems were designed to shift the locus of predictive authority from Western capitals to African institutions, embedding forecasting within regional political communities rather than external security architectures. Yet the same scholarship reveals a persistent tension: early warning systems often reproduce state-centric biases, privileging regime stability and territorial integrity over human security, and frequently generating alerts that are ignored by the very political leaders they are meant to serve. The African perspective thus contributes an insistence that forecasting must be accountable to affected populations, not only to decision-makers, and that the value of a forecast lies in whether it enables preventive action by those most at risk.

The deeper theoretical grounding for this position comes from African scholarship on the politics of knowledge. Achille Mbembe’s analysis of necropolitics and the “postcolony” demonstrates how African subjects have historically been constructed as objects of external management rather than agents of their own futures. In the forecasting context, this means that predictions about African conflict, state failure, or humanitarian crisis often function not as descriptions of reality but as performative constructions that bring into being the very conditions they claim to anticipate. The “fragile state” label, the “arc of crisis” framing, and the routine classification of African regions as high-risk zones are not innocent analytic categories; they are interventions that shape capital flows, aid allocations, military deployments, and diplomatic attention. Jean-François Bayart’s concept of extraversion—the tendency of African political systems to orient themselves toward external patrons—extends to forecasting: African states often consume and reproduce Western risk assessments rather than generating autonomous ones, because the institutional and epistemic resources for independent prediction have been systematically underdeveloped. The African perspective therefore reveals what Western methodological discussions typically occlude: that forecasts have constitutive effects, that they can become self-fulfilling prophecies, and that the accuracy of a prediction about African instability may tell us more about the intervention it triggers than about the underlying social reality.

The epistemic dimension is developed further by African feminist and decolonial scholars. Oyèrónkẹ́ Oyěwùmí’s critique of Western epistemological categories—particularly the imposition of gender and racial taxonomies that do not map onto African social realities—has direct implications for quantitative conflict modeling. ACLED, GDELT, and ICEWS all rely on coding schemes that classify actors, events, and locations according to categories developed in Western academic and policy contexts. Mamdani’s analysis of “define and rule” demonstrates how colonial administrative categories, particularly the ethnic and tribal classifications invented under indirect rule, persist in contemporary conflict data and continue to shape how violence is understood and predicted. When a model predicts “ethnic conflict” in a region, it may be reproducing a colonial taxonomy rather than identifying a causal mechanism. Sylvia Tamale’s work on decolonization and knowledge production insists that African scholars must be central to the design of categories, not merely sources of data for models built elsewhere. The African perspective thus contributes a methodological demand: that the ontological categories embedded in forecasting tools be subjected to critical scrutiny, and that African intellectual traditions be treated as generative frameworks rather than as case material for Western theory.

The structural critique, rooted in Walter Rodney’s analysis of underdevelopment and Nkrumah’s theory of neo-colonialism, adds another layer. Rodney demonstrated that African underdevelopment was not a natural condition but a product of global capitalist structures; Nkrumah showed how formal political independence masked continued economic and epistemic dependence. Applied to forecasting, this means that predictions attributing African instability to local factors—governance failures, ethnic tensions, resource curses—systematically misattribute causation by ignoring the global structures that shape African political economies. Amílcar Cabral’s insistence that the colonized must theorize their own liberation, and Frantz Fanon’s analysis of the psychological and political dimensions of colonial domination, extend to forecasting: the capacity to imagine alternative futures is itself a form of political power, and its monopolization by external actors constitutes a form of domination. The African perspective therefore demands that forecasting be oriented toward emancipatory rather than merely predictive ends, and that it serve African planning and self-determination rather than external risk management.

Each of the five methodologies takes on a different character when viewed from this vantage point. Scenario planning, as developed by Peter Schwartz and the Royal Dutch Shell method, assumes a global corporate actor navigating uncertainty; when applied to Africa, scenarios are often constructed by donors, investors, and security agencies whose interests are not aligned with African populations. The African perspective does not reject scenario planning but insists on African ownership of the scenario process, with scenarios generated from within African political and social contexts rather than imposed from outside. The Delphi method raises acute questions about whose expertise counts: Western academic and policy experts have historically dominated expert panels on African affairs, while African scholars, practitioners, and community leaders are underrepresented. Consensus-building in this context can silence dissenting African voices and produce a false unanimity that masks power asymmetries. Bayesian updating, with its emphasis on explicit priors and disciplined belief revision, offers genuine value: making priors explicit allows them to be contested, and the discipline of updating can counteract the tendency to cling to colonial stereotypes. But the method also depends on data that is often absent or unreliable for African contexts, and priors shaped by colonial knowledge may be difficult to dislodge. Quantitative conflict modeling faces the most direct critique: ACLED’s reliance on media reports systematically overrepresents Anglophone, urban, and elite perspectives; GDELT’s machine coding struggles with African languages and local media; ICEWS’s event data is built on Western news sources that underreport rural violence and overreport events involving Western actors. This is not merely a bias to be corrected through better data collection; it is a structural feature of a global information order that centers Western observers. Calibration and scoring, as developed in the Good Judgment Project and measured by Brier scores, raise the question of whose accuracy matters. In African contexts, outcomes are often determined by external intervention—French military operations, UN peacekeeping, donor conditionality—making “accuracy” a measure of intervention success as much as forecast quality. The superforecaster traits identified by Tetlock and Gardner—cognitive flexibility, open-mindedness, numeracy—may be less relevant where structural constraints dominate and where the cost of being wrong is borne by populations who had no voice in the forecast.

What the African perspective reveals that the other four miss is threefold. First, it exposes the constitutive and performative dimensions of forecasting: predictions do not merely describe futures but help create them, and this is especially consequential in contexts where external actors have the power to act on their predictions. Second, it identifies the epistemic violence embedded in methodological choices—the categories, data sources, and expert panels that silently privilege certain ways of knowing over others. Third, it reframes the purpose of forecasting from accuracy to accountability, insisting that the ultimate test of a forecast is not whether it matches an outcome but whether it serves the interests of those whose lives it predicts. What the African perspective risks occluding is the genuine value of systematic data and rigorous method: local knowledge, while indispensable, is not always sufficient for anticipating large-scale dynamics, and romanticizing it can obscure the need for institutional capacity-building in African research and data systems. The perspective also faces the challenge of scaling: African-owned forecasting institutions remain underfunded and politically vulnerable, and the gap between early warning and early action persists. Nevertheless, the African vantage point is indispensable because it asks the question that the other perspectives too rarely pose: not “how can we predict better?” but “who has the right to predict, and to what end?”

2.7.2 Chinese Perspective

The Chinese perspective on geopolitical forecasting is organized around the concept of “scientific decision-making” (kexue juece), a framing that treats forecasting not as an academic exercise but as an integral component of statecraft and national planning. In this view, the purpose of forecasting is to reduce the arbitrariness of political decisions by grounding them in systematic analysis, and the legitimacy of a forecast derives from its service to the collective goals of the state rather than from its individual accuracy. This orientation shapes how each of the five methodologies is received and adapted.

Scenario planning has a long history in Chinese strategic practice, though it is rarely labeled as such. The Five-Year Plans, the Belt and Road Initiative’s risk assessments, and the strategic planning documents produced by the Central Party School and the Academy of Military Sciences all involve scenario-based thinking about alternative futures. Chinese scenario planning tends to be more state-centric and longer-horizon than its Western corporate counterpart, reflecting a political system that can sustain multi-decade planning horizons without electoral interruption. The contribution here is an insistence that forecasting must be integrated into planning rather than treated as a standalone intelligence product; the occlusion is a tendency to conflate desired futures with predicted ones, and to treat scenario construction as an exercise in political legitimation rather than genuine uncertainty reduction.

The Delphi method is widely used in Chinese policy research, particularly through expert consultation mechanisms (zhuanjia zixun) embedded in bureaucratic processes. Think tanks such as CICIR, SIIS, and CIIS regularly convene expert panels to assess international developments, and the National Development and Reform Commission relies on expert elicitation for major policy decisions. The Chinese adaptation of Delphi tends to emphasize consensus and hierarchy: junior experts defer to senior ones, and the goal is often to produce a unified assessment that can be presented to leadership rather than to surface genuine disagreement. This contributes a model of forecasting as collective deliberation, but it occludes the value of dissent and the cognitive benefits of aggregating independent judgments, which Tetlock’s research has shown to be central to accuracy.

Bayesian updating and quantitative modeling are growing rapidly in Chinese strategic studies, driven by investments in big data, artificial intelligence, and computational social science. Chinese universities and military research institutes have developed conflict forecasting models, and there is significant interest in applying machine learning to event prediction. The “scientific decision-making” framing provides political support for quantification, but it also creates pressure for models to produce actionable, authoritative outputs rather than probabilistic assessments with explicit uncertainty. The contribution is a willingness to invest in data infrastructure and computational capacity at scale; the occlusion is a tendency to treat model outputs as more certain than they are, and to subordinate methodological transparency to state security.

Calibration and scoring are the least developed of the five methodologies in the Chinese context. Individual forecasters are rarely scored against outcomes, and the cultural and institutional emphasis on collective responsibility and hierarchical authority makes individual-level accountability difficult. The Good Judgment Project’s findings about superforecasters—cognitive flexibility, active open-mindedness, numeracy—are not directly transferable to a system where forecasting is a bureaucratic function rather than an individual performance. The Chinese perspective contributes an insistence that forecasting must be evaluated by its contribution to collective decision-making, not by individual accuracy metrics; it occludes the possibility that individual-level cognitive traits and incentives matter independently of institutional context.

Overall, the Chinese perspective’s strength is its integration of forecasting into long-term state planning and its willingness to invest in data and computational infrastructure. Its weakness is the subordination of methodological rigor to political imperatives, the suppression of dissenting assessments, and the difficulty of distinguishing genuine prediction from the legitimation of predetermined strategies.

2.7.3 Russian Perspective

The Russian perspective on geopolitical forecasting is shaped by a strategic culture that treats international politics as an arena of great-power competition in which deception, surprise, and the manipulation of uncertainty are legitimate instruments of statecraft. This orientation produces a distinctive skepticism toward Western forecasting methodologies, which are often viewed not as neutral tools but as components of an information warfare apparatus designed to shape Russian perceptions and constrain Russian options.

Russian strategic culture places a high value on qualitative, historically informed analysis over quantitative modeling. The Russian academic and intelligence traditions emphasize deep historical knowledge, civilizational analysis, and the study of Western intentions and capabilities. Forecasting in this tradition is less about assigning probabilities to discrete events and more about understanding the underlying logic of adversary behavior. This contributes an insistence that geopolitical forecasting must be grounded in historical and cultural knowledge, and that models which strip away context in pursuit of quantification risk missing the strategic logic that drives state behavior. The occlusion is a tendency toward conspiratorial thinking: the assumption that Western actions are more coordinated and intentional than they are, and that apparent chaos or incompetence in Western policy is actually strategic deception.

Scenario planning is used in Russian strategic studies, but with different underlying assumptions. Russian scenarios tend to be more zero-sum and more attentive to Western intentions than to structural or domestic factors. The contribution is an emphasis on the adversary’s perspective and the possibility of strategic surprise; the occlusion is a tendency to overestimate Western coherence and to underestimate the role of domestic politics and contingency in shaping Western policy. The Russian perspective also contributes a long memory: Russian analysts frequently draw on historical analogies—Napoleon, Hitler, the Cold War—that Western forecasters, with their shorter institutional memories, may neglect.

The Delphi method and expert elicitation are practiced in Russian research institutes and intelligence analysis, but with a different epistemic culture. Russian expert panels tend to be more hierarchical and more politically constrained than their Western counterparts, and the range of acceptable dissent is narrower. The contribution is a recognition that expert judgment is always situated in a political context; the occlusion is the difficulty of producing genuinely independent assessments when the expert community is closely tied to the state.

Bayesian updating and quantitative conflict modeling are viewed with particular skepticism. Russian analysts often regard Western quantitative models—ACLED, GDELT, ICEWS—as instruments of Western information dominance, and there is a widespread belief that such models encode Western biases and serve Western interests. This skepticism is not entirely unfounded: the data biases identified in the African perspective—reliance on Western media, underreporting of non-Western contexts—apply with equal force to Russian concerns. The Russian perspective contributes a valuable critical stance toward the apparent objectivity of quantitative methods; it occludes the possibility that such methods, despite their biases, can still improve on unaided human judgment.

Calibration and scoring are largely absent from Russian forecasting practice. The Good Judgment Project’s emphasis on individual cognitive traits and probabilistic reasoning sits uneasily with a strategic culture that values decisiveness, historical intuition, and the willingness to act under uncertainty. The Russian perspective contributes an insistence that forecasting must ultimately serve action, and that excessive refinement of probability estimates can become a form of paralysis; it occludes the possibility that better calibration could improve the quality of Russian strategic decisions.

Overall, the Russian perspective’s strength is its attention to adversary intentions, historical depth, and the political dimensions of forecasting. Its weakness is the tendency toward conspiratorial overinterpretation, the suppression of dissent within expert communities, and the rejection of methodological innovations that could improve accuracy.

2.7.4 Western Perspective

The Western perspective on geopolitical forecasting is the most methodologically self-conscious and empirically oriented of the five. It is defined by a sustained engagement with the problem of intelligence failure and a willingness to subject forecasting practices to systematic testing and reform. The canonical reference points are the Good Judgment Project, the IARPA forecasting tournaments, and the broader literature on cognitive biases and debiasing, culminating in Tetlock and Gardner’s work on superforecasting.

Scenario planning, as developed by Peter Schwartz and the Royal Dutch Shell method, originated in Western corporate and strategic planning contexts and has been widely adopted by Western governments and think tanks. The Western contribution is a well-developed methodology for structuring uncertainty, identifying driving forces, and constructing coherent alternative futures. The occlusion is a tendency to treat scenario planning as a self-contained exercise rather than integrating it with other methods, and to assume that the scenarios constructed by Western analysts correspond to the futures that matter to non-Western actors.

The Delphi method has deep roots in Western policy research, particularly in the RAND Corporation’s Cold War-era work on technological forecasting. The Western adaptation emphasizes anonymity, iteration, and the aggregation of independent judgments. The contribution is a rigorous methodology for eliciting and combining expert opinion; the occlusion is a tendency to treat expert consensus as truth, and to underweight the possibility that experts share systematic biases rooted in their common training and institutional affiliations.

Bayesian updating is the methodological backbone of the Western approach to belief revision. The emphasis on explicit priors, likelihoods, and posteriors reflects a commitment to transparency and to the discipline of changing one’s mind in response to evidence. The contribution is a powerful framework for structuring reasoning and avoiding common cognitive errors; the occlusion is the difficulty of specifying priors and likelihoods in complex geopolitical contexts, and the risk that formal Bayesianism becomes a ritual rather than a genuine aid to judgment.

Quantitative conflict modeling—ICEWS, ACLED, GDELT—represents the Western commitment to data-driven, replicable analysis. These tools have transformed the study of conflict by making large-scale event data available to researchers and policymakers. The contribution is an empirical orientation and a willingness to test claims against data; the occlusion is a tendency to overgeneralize from tournament settings and historical data to high-stakes policy environments where questions are messier, incentives are different, and the cost of being wrong is asymmetric. The Western perspective also struggles with the data biases identified by African and Russian critics: reliance on Western media, underrepresentation of non-Western contexts, and the encoding of Western categories.

Calibration and scoring are the distinctive contribution of the Western perspective. The Brier score, the Good Judgment Project, and the superforecaster research program have demonstrated that individual forecasting accuracy can be measured, improved, and predicted. The contribution is an empirical demonstration that some people are consistently better forecasters than others, and that training and feedback can improve performance. The occlusion is a tendency to overgeneralize from tournament settings to policy environments, and to treat individual cognitive traits as more important than structural and institutional factors.

The Western perspective’s strength is its empirical orientation, methodological transparency, and willingness to learn from failure. Its weakness is a universalism that assumes Western methods are applicable everywhere, an individualism that underweights structural constraints, and a tendency to treat forecasting as a technical problem rather than a political one.

2.7.5 UN Perspective

The UN perspective on geopolitical forecasting is organized around the mandate of early warning and prevention. Unlike national intelligence agencies or corporate strategic planners, the UN must forecast in a multilateral context where questions are politically negotiated, outcomes are jointly determined, and the legitimacy of any prediction depends on its acceptance by member states. This shapes how each methodology is adapted and what is demanded of it.

Scenario planning is used in UN contexts for peacekeeping planning, humanitarian preparedness, and mediation support. The UN contribution is an insistence that scenarios must be developed through inclusive processes that involve affected states and communities, not imposed by external experts. The occlusion is the difficulty of achieving genuine inclusivity in a hierarchical organization where power is unevenly distributed among member states, and the tendency for scenarios to be shaped by the political interests of the most powerful members.

The Delphi method and expert elicitation are widely used in UN early warning, particularly through expert groups convened by the Department of Political and Peacebuilding Affairs, the Office of the Special Adviser on the Prevention of Genocide, and the Framework of Analysis for Atrocity Crimes. The UN contribution is a commitment to multilateral legitimacy: forecasts must be credible to the international community, not just to a single government. The occlusion is the difficulty of achieving genuine consensus among experts from diverse national and institutional backgrounds, and the risk that expert panels become arenas for political negotiation rather than analytic deliberation.

Bayesian updating and quantitative modeling are increasingly used in UN contexts, particularly through initiatives such as UN Global Pulse and the use of satellite imagery and social media data for humanitarian monitoring. The UN contribution is a recognition that data-driven methods can improve the timeliness and granularity of early warning. The occlusion is the difficulty of applying methods designed for individual decision-makers to collective decision-making processes, and the risk that quantitative outputs are treated as more authoritative than they are in politically contested environments.

Calibration and scoring are valued by the UN as tools for improving early warning and prevention, but with an insistence that scoring must be tied to actionable prevention rather than to individual accuracy. The UN contribution is a demand that forecasting serve collective decision-making and that its success be measured by whether it enables preventive action. The occlusion is the difficulty of scoring forecasts in a multilateral context where questions are politically negotiated and outcomes are jointly determined, and the risk that scoring becomes a bureaucratic exercise disconnected from the political will to act.

Overall, the UN perspective’s strength is its commitment to legitimacy, inclusivity, and prevention. Its weakness is the political constraints imposed by member state sovereignty, the difficulty of acting on early warning, and the tendency for forecasting to become an end in itself rather than a means to prevention.

SECTION 2.7.6 Comparison Matrix

Perspective	Key contribution	Key occlusion	Core normative stance	Best evidence	Strongest objection from another perspective
African (primary)	Reframes forecasting as political practice embedded in who defines questions and whose crises count; centers local early warning and community-based indicators	May underweight the genuine predictive gains from probabilistic aggregation and large-N event data	Forecasting must serve African agency and prevention, not external security agendas	AU Continental Early Warning System; ECOWAS early warning; Aning and Danso’s critiques of external prediction	Western: probabilistic methods demonstrably improve accuracy; refusing them risks worse decisions
Chinese	Treats forecasting as state capacity integrated with long-term planning and “scientific decision-making”	Tends to conflate prediction with control; underestimates irreducible uncertainty and black-swan events	Forecasting should strengthen sovereign decision-making and comprehensive national power	Five-Year Plans’ scenario exercises; strategic research institutes (CICIR, SIIS)	Russian/Western: state-centric forecasting can become self-confirming and suppress dissenting signals
Russian	Emphasizes strategic culture, historical pattern, and civilizational cycles; skepticism of universalist probabilistic models	May overfit to historical analogy and underweight base rates and structural change	Forecasting should reflect great-power realities and resist Anglo-American methodological hegemony	Russian strategic studies traditions; emphasis on “correlation of forces”	Western: historical analogy without base rates produced systematic errors (e.g., overestimating Soviet stability)
Western	Developed probabilistic forecasting, calibration, scoring (Brier), and institutionalized self-critique after intelligence failures	Question selection reflects US/ally security priorities; treats accuracy as individual trait rather than collective capacity	Forecasting should be accountable, transparent, and evidence-based	Good Judgment Project; Tetlock and Gardner (2015); ICEWS/ACLED/GDELT	African: accuracy on externally defined questions is not the same as legitimate or useful prediction
UN	Frames forecasting as early warning for prevention; demands legitimacy, inclusivity, and actionability	Difficulty scoring forecasts in multilateral contexts; risk of bureaucratic ritual without political will	Forecasting must enable collective preventive action and be judged by lives saved, not Brier scores	UN early warning frameworks; R2P; AU-UN partnerships	Western: without scoring, there is no feedback loop to improve accuracy

SECTION 2.8 WORKED EXAMPLE (I do)

Case: The Good Judgment Project (2011–2015) — what did the IARPA forecasting tournament teach us about predicting geopolitical events?

Levels of analysis. The GJP can be read at four levels. At the individual level, the tournament isolated cognitive and dispositional traits—openness to experience, need for cognition, active open-mindedness, numeracy, and “perpetual beta” updating—that distinguished superforecasters (the top 2% of participants) from the median forecaster and from intelligence analysts with access to classified information. At the state level, the project was an instrument of the US intelligence community’s post-Iraq-WMD reform agenda: IARPA funded it to test whether open-source, crowd-based methods could outperform classified analysis, and the CIA and other agencies absorbed its lessons into training and analytic tradecraft. At the systemic level, the tournament unfolded during a unipolar-to-multipolar transition in which the United States sought to maintain decision advantage amid Arab Spring upheavals, Syrian civil war, Russian assertiveness, and Chinese rise; the questions asked reflected that strategic agenda. At the transnational level, the events forecast—elections, coups, ceasefires, disease outbreaks, economic indicators—were produced by cross-border processes, and the forecaster pool itself was international, meaning that local knowledge from participants in target countries was one input among many, though the scoring and question-framing remained US-centered.

Mechanism section. The GJP’s core mechanism was a feedback loop between probabilistic elicitation, rapid scoring, and belief revision. Forecasters assigned probabilities to precisely worded, time-bound questions; outcomes were resolved by independent judges; Brier scores (Brier 1950) provided a proper scoring rule that rewarded both calibration (probabilities matching observed frequencies) and resolution (discriminating between events that occurred and did not). Three mechanisms drove superior performance. First, aggregation: averaging many independent forecasts, then “extremizing” toward 0 or 1, outperformed individual experts, consistent with the wisdom-of-crowds logic. Second, cognitive style: superforecasters updated frequently in small increments, sought disconfirming evidence, and decomposed questions into components with base rates. Third, institutional design: training in probabilistic reasoning and bias mitigation, team structures with devil’s-advocate roles, and a tournament rhythm that created accountability. Tetlock and Gardner (2015) report that superforecasters beat intelligence analysts by roughly 30% in Brier score terms and that the top team’s performance was sustained across years, suggesting the gains were not luck. The mechanism is therefore not “better information” but better processing of available information: the GJP demonstrated that open-source, structured cognition can outperform classified, unstructured expertise under conditions of moderate uncertainty.

Perspective reconstruction.

African (primary). The GJP is best read from an African vantage point as a case study in the politics of prediction: the tournament’s questions were overwhelmingly framed by US national-security priorities, so African conflicts, elections, and crises appeared as objects of external monitoring rather than as problems defined by African actors. African early warning systems—such as the AU Continental Early Warning System and ECOWAS’s early warning mechanism—operate on different epistemologies, combining quantitative indicators with community-based reporting, traditional authority networks, and political mediation; they are judged by whether they enable prevention, not by Brier scores on externally posed questions. African scholars such as Aning and Danso have argued that forecasting which treats African states as targets of prediction reproduces asymmetries of knowledge and power, and that the “accuracy” celebrated by the GJP may be accuracy on questions that matter to Washington rather than to Abuja or Addis Ababa. At the same time, the African perspective does not dismiss the GJP’s findings: the demonstration that structured, probabilistic reasoning outperforms unstructured expert judgment is relevant to African intelligence and early warning institutions, provided the questions, indicators, and scoring are locally owned. The occlusion in the African reading is that it can understate the genuine methodological advance the GJP represents—superforecasting is a real improvement over both punditry and unexamined expert intuition—and that rejecting probabilistic methods outright risks leaving African decision-makers with worse tools. What the African vantage point reveals that the others miss is that the GJP’s central metric, individual accuracy on resolved questions, silently encodes a particular distribution of power: who asks, who answers, and who benefits from being right.

Chinese. From a Chinese perspective, the GJP exemplifies a Western, specifically American, approach to forecasting as competitive advantage: open-source crowd wisdom harnessed to state power through IARPA, with results fed into intelligence reform. Chinese strategic forecasting, framed as “scientific decision-making” (kexue juece), emphasizes integration of prediction into long-term planning, comprehensive national power assessments, and party-state coordination; the GJP’s individualist, tournament-based model appears both methodologically interesting and politically naive, since it separates forecasting from decision authority and from the planning apparatus that gives predictions consequence. Chinese scholars and strategists would likely credit the GJP’s demonstration of aggregation and calibration while questioning whether its findings transfer to a system where forecasting is embedded in hierarchical, long-horizon planning rather than in short-cycle tournament questions. The occlusion is a tendency to treat state-integrated forecasting as inherently superior, which can suppress dissenting signals and overfit to official narratives.

Russian. The Russian perspective reads the GJP through the lens of strategic culture and great-power competition: a US-funded project to systematize prediction for hegemonic purposes, producing “superforecasters” whose cognitive style—open-mindedness, frequent updating, willingness to abandon prior views—reflects liberal, individualist epistemology rather than the historical-pattern and civilizational-cycle reasoning valued in Russian strategic thought. Russian forecasting traditions emphasize the “correlation of forces,” long-term historical trajectories, and the role of will and leadership; the GJP’s reduction of geopolitics to resolvable, time-bound questions with Brier scores appears to strip away the tragic, non-linear dimensions of international politics. The contribution of this perspective is its insistence that some geopolitical outcomes are not merely uncertain but structurally underdetermined, and that treating them as probabilistic bets can induce overconfidence. Its occlusion is a risk of overfitting to historical analogy and underweighting base rates, which Tetlock’s earlier work (Expert Political Judgment, 2005) showed to be a systematic source of expert error.

Western. The Western perspective is the GJP’s home turf: it treats the tournament as a landmark in the intelligence-failure self-critique that followed Iraq WMD, demonstrating that open-source, structured, scored forecasting can outperform classified, unstructured expertise. The mechanism—calibration, aggregation, extremizing, cognitive-style selection, training—is presented as a generalizable improvement in analytic tradecraft, and the project’s findings have been absorbed into intelligence community practice and into commercial forecasting. The strength is empirical rigor and institutional accountability; the occlusion is that the questions, the scoring, and the “accuracy” metric are all defined within a US security frame, so the apparent universality of the method masks a particular geopolitical standpoint. The Western perspective also tends to individualize success (the superforecaster) rather than treating forecasting as a collective, political capacity.

UN. The UN perspective values the GJP’s methodological advances but judges them by a different standard: whether they enable prevention. From this vantage point, the tournament’s focus on individual accuracy on resolved questions is a narrow metric; what matters is whether forecasting produces actionable early warning that mobilizes collective response, as in the UN’s early warning and R2P frameworks. The GJP’s questions were rarely about slow-burn crises or structural violence in ways that would trigger preventive diplomacy; its scoring system cannot capture the value of a forecast that, by prompting action, prevents the event it predicted. The UN contribution is a demand that forecasting serve collective decision-making and that success be measured by lives saved and crises averted, not only by Brier scores. The occlusion is the difficulty of scoring forecasts in a multilateral context where questions are politically negotiated and outcomes are jointly determined.

Explicit adjudication. The Western perspective best explains the GJP as an institution and as a methodological achievement: the project was funded by a US intelligence agency, designed by Western scholars, and evaluated by Western metrics, and its findings about calibration, aggregation, and cognitive style are robust within that frame. However, the Western perspective misses what the African and UN perspectives reveal: that “accuracy” on externally defined questions is not equivalent to legitimate or useful prediction, and that the tournament’s success metric cannot register prevention. The Chinese and Russian perspectives add a further caution: forecasting embedded in state planning or in historical-civilizational reasoning may capture dimensions of geopolitics—long-horizon structural change, the role of will and leadership—that probabilistic tournament questions flatten. The best explanation of what the GJP “taught” is therefore a hybrid: individual-level cognitive traits and institutional design genuinely improve short-horizon, well-specified predictions, but the tournament’s lessons are bounded by question selection, by a US security frame, and by a metric that rewards being right over making a difference. The GJP taught us how to forecast better; it did not teach us what to forecast, or for whom.

SECTION 2.9 PROBLEM PAIR (We do)

Classify the main forecasting communities that assessed a possible Russian invasion of Ukraine in late 2021 and early 2022, and state what each community treated as decisive evidence.
Separate the four levels of analysis for the case: what was observed, why it might have been occurring, what was predicted, and what policy response was prescribed.
Construct a simple Bayesian update for invasion by late February 2022: state a prior, identify at least two pieces of evidence, assign rough likelihoods under “invasion” and “coercive diplomacy,” and state a posterior range.
Compare what quantitative conflict datasets and event-data systems could and could not detect in this case, and identify one false positive and one false negative risk.
Reconstruct how African, Chinese, Russian, Western, and UN perspectives would each frame whether the invasion was “predictable,” and identify what each perspective occludes.
Adjudicate the main source of forecasting failure—epistemic, institutional, or ontological—and specify two observations that would falsify your adjudication.

Model answer. The forecasting record around the 2022 Russian invasion of Ukraine is best read as a case of structured disagreement under radical uncertainty rather than as a simple story of experts failing and open-source analysts succeeding. In late 2021 and early 2022, several communities were looking at overlapping evidence but weighting it through different models of Russian decision-making. Intelligence agencies with privileged collection emphasized order-of-battle indicators, logistics, and command-and-control signatures; their warnings became more specific as winter approached. Open-source analysts and journalists documented the physical buildup with satellite imagery, railway movements, and equipment transfers, producing a strong descriptive record of militarization. Prediction-market participants and superforecasters, where they were engaged, had to translate those signals into probabilities, and their performance depended heavily on whether they treated the buildup as a standing coercive threat or as preparation for an irreversible attack. Many European political leaders and some Ukrainian officials discounted the highest-probability invasion scenarios, partly because they weighted diplomatic incentives, economic costs, and the apparent irrationality of a full-scale war more heavily than military indicators. Russian strategic communication, including denials and demands for security guarantees, was itself a source of ambiguity because it was consistent with both bargaining and concealment.

The first analytical discipline is to keep description, explanation, prediction, and prescription separate. Description includes troop concentrations, exercises, diplomatic notes, and economic measures. Explanation asks why those actions might be occurring: coercive diplomacy, preparation for limited escalation, domestic political consolidation, or genuine intent to conquer. Prediction assigns probabilities and timing to possible outcomes. Prescription concerns what third parties should do, such as deterrence, reassurance, sanctions, or evacuation. A common error in public debate was to move from description to prescription while treating one explanation as established. A large buildup was visible; the inference that it would necessarily produce a full-scale invasion was not equally visible.

A Bayesian treatment clarifies why. A reasonable prior for a full-scale invasion of a large European state by a nuclear power in early 2022 was low, because the base rate of such wars is low and the expected costs were enormous. Evidence such as sustained force concentration, Belarusian staging areas, and ultimatum-like diplomatic demands raised the likelihood of invasion, but the same evidence also raised the likelihood of coercive bargaining. The posterior therefore depended on the relative likelihoods assigned to each hypothesis. If one assumed that mobilization is cheap talk, the posterior remained moderate; if one assumed that mobilization plus ultimatum language is a strong predictor of attack, the posterior rose sharply. The decisive update for many observers came not from a single new fact but from the accumulation of logistics indicators, the failure of diplomatic off-ramps, and the recognition that Russian force posture was optimized for offensive operations rather than for indefinite coercion. Even then, uncertainty about Putin’s private decision, the possibility of last-minute bargaining, and the risk of mirror-imaging meant that a high probability was warranted but certainty was not.

Quantitative conflict modeling adds a different kind of value and a different kind of limit. Systems such as ICEWS, ACLED, and GDELT are useful for detecting patterns in reported events, mobilization, violence, and media tone, and they can surface escalation trends that human analysts miss. In this case, event data could have captured rising military activity, border incidents, and diplomatic friction. But these systems do not directly observe leader intent, elite bargaining, or classified planning. They are also vulnerable to reporting bias, definitional drift, and the temptation to treat correlation as prediction. A model may predict elevated conflict probability without specifying whether the conflict will be a limited incursion, a full-scale invasion, or a prolonged standoff. The false-positive risk is that mobilization is read as imminent war; the false-negative risk is that a decision made in a small circle produces little open-source signal until execution.

The five perspectives reconstruct the case differently. An African perspective would stress that early warning is often technically available but politically ignored, and that great-power crises expose the selectivity of international attention; it would also highlight sovereignty, non-intervention norms, and the unequal distribution of preventive capacity. A Chinese perspective would emphasize strategic patience, the possibility that Washington was overreading Moscow’s signals, and the value of reading military posture within a longer geopolitical competition. A Russian perspective, especially from within Russian strategic culture, would treat deception, ambiguity, and the concentration of decision-making as normal features of statecraft, making external prediction inherently difficult. A Western perspective would focus on intelligence collection, alliance coordination, and the post-Iraq need to avoid overstating certainty, while also acknowledging mirror-imaging and political incentives to avoid alarm. A UN perspective would frame the case through early warning, preventive diplomacy, and the gap between information and collective action, noting that even accurate forecasts do not automatically generate consensus or enforcement.

The most defensible adjudication is that the miss, where it occurred, was overdetermined by all three sources of failure. Epistemically, the evidence was genuinely ambiguous because the same mobilization supported multiple hypotheses. Institutionally, intelligence communities had incentives to hedge, political leaders had incentives to avoid panic, and public discourse often rewarded false confidence in either direction. Ontologically, the final decision appears to have been concentrated and possibly late, meaning that even a perfect model of observable indicators could not fully determine the outcome. This adjudication would be falsified if declassified records showed that the invasion decision was fixed months earlier and clearly signaled in observable indicators, which would strengthen the epistemic-failure account; or if records showed that decision-makers genuinely changed course at the last moment in response to diplomacy, which would strengthen the ontological account. The lesson is not that forecasting failed, but that good forecasting must make ambiguity explicit, update as evidence accumulates, and separate the probability of an event from the wisdom of a policy response.

SECTION 2.10 INDEPENDENT MEMO (You do)

(a) Case description: In early 2025, a regional early-warning cell must assess whether a sustained border militarization episode in a fragile region will escalate to interstate war within ninety days. The cell has satellite imagery showing troop concentrations, ACLED records of cross-border incidents, diplomatic cables suggesting coercive bargaining, and a Delphi panel of regional experts split between “bluff” and “imminent attack.” It must produce a probability estimate, update it as new evidence arrives, and recommend preventive diplomacy without either crying wolf or missing an attack. The cell’s audience includes governments with competing interests, a regional organization with limited enforcement capacity, and humanitarian agencies that need lead time for contingency planning.

(b) Required memo headings as prompts:
Mechanism: Identify the causal pathway by which border militarization could become war, and distinguish necessary conditions, sufficient conditions, and triggering events.
Perspective reconstruction: Reconstruct how African, Chinese, Russian, Western, and UN actors would interpret the same evidence, and state what each reading occludes.
Historical analogue: Select one prior case of coercive mobilization and explain what is structurally similar, what differs, and why the analogy may mislead.
Falsifiers: State two observations that would weaken your probability estimate and two that would strengthen it, and specify what evidence would change your policy recommendation.

(c) Self-check: Did you distinguish description, explanation, prediction, and prescription, and did you avoid converting a probability estimate into a policy conclusion without an explicit decision rule?

SECTION 2.11 HINGE QUESTION

A forecasting team has strong Brier scores on routine political questions. In February 2022 it assigned only a 30 percent probability to a full-scale Russian invasion despite visible troop buildups. Which conclusion is most warranted?

A. The team’s low score on this question proves that quantitative and expert forecasting methods are useless in geopolitics.
B. The team should have treated the troop buildup as sufficient evidence and assigned near-certainty to invasion.
C. The miss reflects the difficulty of inferring a leader’s private decision from public signals; even well-calibrated forecasters face base-rate and evidence-ambiguity problems, so the appropriate response is disciplined updating and explicit uncertainty rather than abandonment of method.
D. Because experts disagreed, the correct probability was necessarily 50 percent.

Distractor diagnoses: Option A conflates a single high-stakes miss with overall methodological invalidity and ignores that calibration is assessed across many questions, not one. Option B commits a deterministic fallacy by treating mobilization as sufficient evidence, when the same observable indicators were consistent with coercive diplomacy as well as invasion. Option D treats expert disagreement as a reason to default to equiprobability, rather than as information to be aggregated, weighted, and updated.

SECTION 2A — DATA LAB

Table 1 — Dataset comparison: ICEWS, ACLED, GDELT

Dataset	Temporal coverage	Geographic scope	Event granularity	Primary source base	Known bias / limitation
ICEWS (Integrated Crisis Early Warning System)	1995–present	Global	Country-day political events (CAMEO-coded)	Automated coding of multiple news sources	Over-representation of English-language and Western media; under-coding of non-violent contention and local-language reporting
ACLED (Armed Conflict Location & Event Data)	1997–present (Africa origin; now global)	Africa originally; expanded globally	Sub-national, actor-disaggregated events	Human coders + media + NGO/IGO reports	Under-reporting in remote or inaccessible conflict zones; definitional drift across coding revisions; lag in real-time updates
GDELT (Global Database of Events, Language, and Tone)	1979–present	Global	15-minute updates, CAMEO-coded, tone scores	Automated global media monitoring	High false-positive rate; machine-coding ambiguity; “noise” in trend series; tone scores not validated as sentiment measures

Table 2 — Good Judgment Project tournament outcomes (illustrative ranges, drawn from Mellers et al. 2014 and Tetlock & Gardner 2015)

Group	Approx. participants	Forecasting questions	Mean Brier score (lower = better)	Notes
Crowd (all forecasters)	~2,800 (Year 1)	100+ per year	~0.35–0.40	Wisdom-of-crowds baseline; untrained volunteers
Top 2% (“superforecasters”)	~50–60	Same set	~0.19–0.22	Selected by prior-year performance; teaming + training + feedback
Intelligence analysts (control)	Hundreds	Subset	~0.28–0.32	Classified access; professional training; no feedback loop
Prediction markets	Variable	Subset	~0.22–0.26	Incentive-compatible prices; aggregate of self-selected traders

Note: figures are approximate ranges reported in the GJP literature; exact point estimates vary by question set and year.

Table 3 — ACLED-based prediction vs. outcome, Sahel 2021–2023 (illustrative)

Region	Predicted risk (ACLED trend + model)	Actual major conflict events	Hit?	Miss type
Mali (central)	High	High (JNIM expansion)	Yes	—
Burkina Faso (north)	Medium	High (coup + insurgency surge)	Under-predicted	Political shock (coup) not captured by conflict trend
Niger (Tillabéri)	High	High	Yes	—
Chad (Lake Chad basin)	Medium	Medium	Yes	—
Senegal (Casamance)	Low	Low	Yes	—

Numbered analysis questions

Trend. Looking at Table 2, how do Brier scores vary across groups, and what does the narrowing gap between superforecasters and prediction markets suggest about the source of forecasting skill — individual cognition, aggregation, or incentives?
Level of analysis. Table 1 shows ICEWS operating at country-day resolution and ACLED at sub-national actor level. At which level of analysis (individual, state, systemic, transnational) does each dataset primarily operate, and what causal claims can each legitimately support? Which level is systematically under-represented in all three?
Perspective reconstruction (all five). How would each of the five perspectives — African, Chinese, Russian, Western, UN — read the GJP result that trained amateurs outperformed intelligence analysts? What does each perspective credit, and what does each occlude?
Interpretation. Table 3 shows an under-prediction in Burkina Faso driven by a coup. What does this imply about the limits of trend-extrapolation models, the role of political shocks, and the difference between forecasting conflict levels and forecasting conflict transitions?
Limitation. What are the most serious limitations of using GDELT and ICEWS for African conflict early warning, and how do these limitations connect to the politics of prediction — who gets to define what counts as an “event,” whose media counts as data, and whose suffering becomes visible?

SECTION 3 — DUAL-CODED VISUAL ANCHOR

Visual 1 — “Bayesian Belief-Revision Funnel”

What to draw: A left-to-right flow diagram on a single page. On the far left, draw a wide, low bell curve labeled “Prior” with a vertical dashed line at its mean and a shaded band for the 90% credible interval. An arrow labeled “Bayes’ rule: P(H|D) ∝ P(D|H)·P(H)” points rightward to a middle panel showing three data points (dots) labeled “Likelihood (new evidence)” with a small likelihood curve beneath them. A second arrow points to the far right, where a narrower, taller bell curve labeled “Posterior” sits, its mean shifted toward the data and its credible interval visibly tighter. Below the posterior, draw a small feedback loop arrow returning to “Prior for next round,” labeled “iterative updating.”

How to label it: Title at top: “Bayesian Updating as Discipline of Belief Revision.” Label each curve’s mean (μ₀, μ₁) and standard deviation (σ₀, σ₁). Add a small box: “Brier score penalizes overconfidence: a 90% claim that fails scores 0.81.”

Perspective overlay to annotate: Along the bottom, five small flags/icons with one-line annotations: African — “priors often come from local, non-quantified knowledge that the formal model cannot ingest”; Chinese — “state intelligence supplies the prior; updating is politically gated”; Russian — “strategic-culture priors are treated as fixed, not revisable”; Western — “base-rate priors from historical frequency; the default in GJP training”; UN — “consensus priors from multilateral assessments; slow to revise.”

Visual 2 — “Forecasting Pipeline with Perspective Lenses”

What to draw: A horizontal pipeline of five connected boxes, left to right: (1) “Data ingestion (ICEWS/ACLED/GDELT),” (2) “Model / algorithm,” (3) “Expert elicitation (Delphi),” (4) “Scenario planning (Schwartz),” (5) “Decision / policy.” Above the pipeline, draw a “lens” icon (a magnifying glass) over each box with a perspective label: Box 1 lens “Western media bias”; Box 2 lens “quantitative reductionism”; Box 3 lens “who counts as expert?”; Box 4 lens “whose futures are imaginable?”; Box 5 lens “whose interests decide?” Below the pipeline, draw a feedback arrow from Box 5 back to Box 1 labeled “outcome data → model retraining.”

How to label it: Title: “Where Perspective Enters the Forecasting Pipeline.” Label each box with its key method (e.g., Box 3 “Delphi: anonymous rounds, controlled feedback”). Add a legend: solid arrow = information flow; dashed arrow = feedback.

Perspective overlay to annotate: To the right of the pipeline, a vertical column of five perspective tags, each with a one-line verdict on the whole pipeline: African — “the pipeline predicts African futures using African data filtered through non-African frames”; Chinese — “the pipeline is a tool; the state decides which outputs matter”; Russian — “the pipeline is Western epistemology wearing technical clothing”; Western — “the pipeline is self-correcting through scoring”; UN — “the pipeline feeds prevention frameworks but rarely triggers action.”

Visual 3 — “Calibration Plot: Superforecasters vs. Novices”

What to draw: A standard square plot. X-axis: “Predicted probability” (0.0 to 1.0). Y-axis: “Observed frequency” (0.0 to 1.0). Draw the 45-degree diagonal line labeled “Perfect calibration.” Draw two curves: (a) a solid line hugging the diagonal, labeled “Superforecasters (GJP top 2%)”; (b) an S-shaped dashed line that is below the diagonal at low probabilities and above at high probabilities, labeled “Novices (overconfident extremes).” Shade the area between the novice curve and the diagonal in light red, labeled “calibration error (Brier-relevant).” Add horizontal gridlines at 0.25, 0.5, 0.75.

How to label it: Title: “Calibration as the Core Skill.” Annotate two points: at predicted 0.9, novice observed ~0.65 (“overconfidence”); at predicted 0.1, novice observed ~0.25 (“underconfidence at low end”). Add a small inset box: “Brier score = mean squared error of probability forecasts; 0 = perfect, 1 = worst.”

Perspective overlay to annotate: Below the plot, five short annotations: African — “calibration to whose base rates?”; Chinese — “calibration is technical; strategic judgment is separate”; Russian — “calibration assumes stable referents; geopolitics is not stationary”; Western — “calibration is the measurable core of forecasting skill”; UN — “calibration of early-warning thresholds determines when prevention triggers.”

SECTION 4 — CORNELL NOTES

Cue (question / prompt)	Notes (with source log)
What is a Brier score and why does it matter?	Mean squared error of probability forecasts; ranges 0 (perfect) to 1 (worst); Brier (1950). Rewards both accuracy and calibration; used in GJP scoring.
How does the Delphi method work?	Anonymous expert rounds with controlled feedback; Dalkey & Helmer (1963). Strengths: reduces groupthink, surfaces dissent. Weaknesses: slow, expert selection bias, false consensus.
What is scenario planning (Schwartz)?	Royal Dutch Shell method; Schwartz (1991). Builds 2–4 plausible futures from key uncertainties; not prediction but preparation. Strength: handles deep uncertainty.
What did the GJP teach us?	Tetlock & Gardner (2015); Mellers et al. (2014). Trained amateurs + teaming + feedback outperformed intelligence analysts; superforecasters ~top 2%; skill is partly trainable.
What are ICEWS/ACLED/GDELT?	ICEWS documentation; Raleigh et al. (2010) ACLED; Leetaru & Schrodt (2013) GDELT. Automated/semi-automated event data; different resolutions and biases.
What is Bayesian updating in forecasting?	Bayes’ rule; priors + likelihoods → posteriors. Discipline of belief revision; combats confirmation bias. Requires quantified priors, often unavailable.
What is calibration?	Alignment of stated probabilities with observed frequencies. Core skill in GJP; measured by Brier score and calibration plots.
What is the African perspective on forecasting?	Early warning systems (CEWARN, AU CEWS, ECOWARN); politics of prediction; African scholarship (Grovogui, Mbembe, Bayart, Mamdani, Rodney, Fanon, Aning, Tamale, Oyěwùmí).
What is the Chinese “scientific decision-making” framing?	Forecasting as state capacity; strategic forecasting (zhanlue yuce); data-driven governance; occludes dissenting scenarios.
What is the Russian strategic-culture critique?	Forecasting as Western epistemology; emphasis on will, deception, non-stationarity; skepticism of probabilistic claims.
What is the Western intelligence-failure self-critique?	Post-Iraq WMD; Tetlock (2005) expert overconfidence; GJP as reform; calibration and scoring as accountability.
What is the UN early-warning/prevention framework?	Framework of Analysis for Atrocity Crimes; CEWS; “never again” gap between warning and action; political constraints on response.

Source log: Tetlock & Gardner (2015); Schwartz (1991); Brier (1950); Dalkey & Helmer (1963); Mellers et al. (2014); Tetlock (2005); ICEWS documentation; Raleigh et al. (2010); Leetaru & Schrodt (2013); Kuhn (2020); Grovogui (2006); Mbembe (2001); Bayart (2009); Mamdani (1996); Rodney (1972); Fanon (1961); Tamale (2020); Oyěwùmí (1997).

Summary: Geopolitical forecasting is not a single method but a family of approaches — scenario planning, Delphi elicitation, Bayesian updating, quantitative event modeling, and scored calibration — each with distinct scope conditions and failure modes. The Good Judgment Project demonstrated that disciplined, feedback-driven amateurs can outperform intelligence professionals, but this result is bounded by question selection, base-rate availability, and the politics of whose data counts. From an African vantage point, the deeper issue is not method but the politics of prediction: early-warning systems exist, but warning rarely converts to prevention, and the data infrastructures that feed global models systematically under-represent African realities. Chinese, Russian, Western, and UN perspectives each add a lens — state capacity, strategic-culture skepticism, self-correcting empiricism, and prevention frameworks — while each also occludes something the others see.

SECTION 5 — FEYNMAN COMPRESSION

Step 1 — Explain without jargon. Geopolitical forecasting is the attempt to say what is likely to happen in world politics before it happens. People do this in several ways: by imagining a few different futures and preparing for each (scenario planning); by asking experts privately and combining their answers (Delphi); by starting with a guess and changing it when new evidence arrives (Bayesian updating); by feeding news into computer models that count events (ICEWS, ACLED, GDELT); and by keeping score on predictions so that overconfident people are exposed (Brier scores, superforecasters). The big lesson from the Good Judgment Project is that ordinary people who are trained to think in probabilities, who update their beliefs, and who work in teams can beat intelligence professionals — but only on questions where good data exists and where the future resembles the past enough to learn from.

Step 2 — Mark gaps. I am unsure how much of the superforecaster advantage comes from selection (they were already good) versus training (they became good). I am unsure whether Bayesian updating works when priors cannot be quantified, which is common in geopolitics. I am unsure whether ACLED’s expansion from Africa to global coverage preserved its original coding reliability. I am unsure how to weigh the Russian claim that probabilistic forecasting is epistemologically inappropriate for strategic interaction. I am unsure whether the UN’s early-warning frameworks have ever actually prevented a conflict, or only documented failures.

Step 3 — Reread verification. Rechecking against Tetlock & Gardner (2015): the superforecaster advantage is attributed to both selection and training, with teaming and feedback as key mechanisms — so my gap on selection-vs-training is partially closed: it is both. Rechecking Brier (1950): the score is mean squared error of probability forecasts — confirmed. Rechecking Schwartz (1991): scenario planning is explicitly not prediction but preparation for multiple futures — confirmed. Rechecking ICEWS documentation: country-day resolution with automated coding — confirmed. Rechecking Mellers et al. (2014): psychological strategies (growth mindset, fox-like thinking) identified — confirmed. The gap on ACLED global expansion reliability remains open; I flag it as unresolved.

Step 4 — Analogy with explicit limits. Forecasting is like weather forecasting: you use models, update with new data, and score your predictions. The analogy helps because both involve probability, base rates, and calibration. The analogy breaks down because weather systems are governed by known physics and stationary distributions, whereas geopolitical events involve strategic actors who react to forecasts, non-stationary institutions, and data generated by media systems with their own biases. A weather model does not change the weather; a published geopolitical forecast can change the politics.

Step 5 — Rewrite. Geopolitical forecasting is a set of disciplined methods for reasoning under uncertainty — scenario planning for deep uncertainty, Delphi for expert aggregation, Bayesian updating for belief revision, quantitative event data for pattern detection, and scoring for accountability. The Good Judgment Project showed that skill is partly trainable and that amateurs can outperform professionals on well-structured questions, but the result does not generalize to all geopolitical questions, especially those involving strategic interaction, data-poor regions, or non-stationary institutions. From an African perspective, the central problem is not the absence of methods but the politics of prediction: early-warning systems exist, yet warning rarely triggers prevention, and the data that feeds global models systematically under-represents African realities. Chinese, Russian, Western, and UN perspectives each illuminate a different failure mode — state capture of forecasts, epistemological skepticism, overconfidence in quantification, and the gap between warning and action — and no single perspective is sufficient.

SECTION 6 — SPACED-REPETITION FLASHCARDS

Tag	Card Type	Front	Back	Perspective Note
L58	definition	What is scenario planning?	Scenario planning constructs several plausible futures to stress-test decisions rather than betting on one prediction; Schwartz’s Shell method distinguishes predetermined elements from critical uncertainties and builds 2x2 scenario matrices. It is a tool for robustness, not accuracy.	African early-warning practice often prefers scenario pluralism because single-point forecasts have historically served external actors’ interests.
L58	mechanism	How does the Royal Dutch Shell scenario method work?	Analysts map driving forces, identify two critical uncertainties, cross them into four scenarios, and test strategic options against each; the value lies in revealing vulnerabilities, not in picking the “right” future.	Chinese strategic-planning documents similarly use scenario matrices but embed them in five-year plan contingencies.
L58	definition	What is the Delphi method?	The Delphi method elicits expert judgments through anonymous, iterated rounds with controlled feedback, aiming to reduce groupthink and converge on a consensus estimate without face-to-face dominance.	Russian expert councils (e.g., Valdai) use Delphi-like elicitation but with less anonymity and more state framing.
L58	mechanism	What are the strengths and weaknesses of Delphi?	Strengths: anonymity reduces conformity, iteration refines estimates, and aggregation beats individual experts; weaknesses: convergence can be false consensus, experts share blind spots, and the method cannot generate information not already in the panel.	Western intelligence has used Delphi variants for long-range estimates; African panels note the exclusion of local knowledge holders.
L58	definition	What is Bayesian updating?	Bayesian updating revises a prior probability with new evidence via Bayes’ theorem to produce a posterior; the discipline is to make priors explicit and to update by the likelihood, not by narrative.	UN early-warning frameworks increasingly require explicit priors in risk assessments.
L58	mechanism	State Bayes’ theorem in words.	Posterior odds equal prior odds times the likelihood ratio; a forecast changes only as much as the evidence warrants, and strong priors require strong evidence to move.	African conflict analysts stress that priors built on colonial data distort posteriors.
L58	definition	What is a Brier score?	The Brier score measures the mean squared difference between a probabilistic forecast and the binary outcome; lower is better, and it decomposes into calibration, resolution, and uncertainty.	Scoring rules discipline overconfident Western analysts; African forecasters argue scores ignore distributional harm.
L58	application	What is ICEWS?	ICEWS (Integrated Crisis Early Warning System) is a US-funded quantitative conflict-forecasting system using event data and machine learning to predict political instability and conflict onset.	ICEWS has been criticized for weak performance on African conflicts and for opacity to local users.
L58	application	What is ACLED?	ACLED (Armed Conflict Location & Event Data) records political violence events with location, actors, and fatalities; it is a foundational input for both research and operational forecasting.	ACLED’s African origins and local coder networks make it a model of situated data collection.
L58	application	What is GDELT?	GDELT monitors global news in near real time and codes events, actors, and tone; it enables high-frequency conflict indicators but carries media-bias and coverage gaps.	GDELT underrepresents African-language and rural events, skewing forecasts.
L58	comparison	Superforecasters vs. domain experts: who wins?	Tetlock’s Good Judgment Project found that trained generalists with high cognitive reflection and active open-mindedness outperformed experts and intelligence analysts on geopolitical questions.	African scholars caution that “superforecaster” traits may reflect Western educational norms.
L58	source	What did Tetlock and Gardner (2015) establish?	Tetlock and Gardner (2015) showed that forecasting is a skill improvable through training, feedback, and aggregation, and that expert political judgment is often no better than chance.	The book’s tournament setting was US-centric; its lessons need adaptation to African contexts.
L58	debate	Calibration vs. resolution: which matters more?	Calibration (matching stated probabilities to outcomes) and resolution (discriminating events from non-events) trade off; a good forecaster needs both, and scoring rules reward the combination.	UN prevention frameworks weight calibration for accountability; African communities weight resolution for early action.
L58	reconstruction	How do African early-warning systems frame prediction?	African early-warning systems (CEWARN, ECOWARN, MARAC) frame prediction as a political process tied to response capacity, not as a technical exercise; the bottleneck is often political will, not information.	This perspective reveals that forecasting without response architecture is performative.
L58	occlusion	What do quantitative models typically occlude?	Quantitative models occlude local causal mechanisms, informal institutions, and the agency of non-state actors; they excel at pattern detection but struggle with structural breaks and novel events.	This occlusion is most severe in African contexts where informal governance dominates.

SECTION 7 — METACOGNITIVE CALIBRATION

7.1 Confidence Audit (score 0–100)

Concept / Claim	My confidence (0–100)	Evidence I can cite	Key gap
I can explain Bayes’ theorem in words
I can compute a Brier score from a forecast-outcome pair
I know the difference between calibration and resolution
I can describe the Shell scenario method’s two uncertainties
I can state what ICEWS predicts and its known limits
I can reconstruct the African early-warning critique
I can explain why superforecasters outperform experts
I can identify three failure modes of Delphi
I can apply Bayesian updating to a geopolitical case
I can distinguish description, explanation, prediction, prescription

7.2 Three-Question Debrief

Which forecasting method did I most overestimate in confidence, and what evidence would lower it?
Where did my prior on the 2022 Russian invasion sit, and did I update fast enough when evidence accumulated?
What would have to be true for quantitative conflict models to outperform human judgment in African contexts?

7.3 Error-Log

Common Misconception	Correction	Evidence
Forecasting means predicting a single outcome.	Forecasting produces probability distributions over scenarios; single-point predictions are a misuse.	Tetlock and Gardner (2015); Schwartz (1991)
Experts are always better than generalists.	Trained generalists with feedback often outperform domain experts on geopolitical questions.	Good Judgment Project results
More data always improves forecasts.	Data without causal structure or local validity can amplify bias; GDELT/ICEWS have known coverage gaps.	ICEWS documentation; ACLED critiques
Bayesian updating is just “changing your mind.”	It requires explicit priors and likelihoods; informal updating often double-counts evidence.	Standard Bayesian pedagogy
High Brier score means good forecasting.	Lower Brier score is better; and scores must be decomposed into calibration and resolution.	Brier (1950)
Quantitative models are objective.	Models encode training-data biases and designer choices; objectivity claims occlude politics.	African critical scholarship
Delphi always reaches true consensus.	Delphi can produce false consensus when experts share blind spots or anchoring.	Delphi method literature
Early warning equals prevention.	Warning without response capacity is performative; the bottleneck is political will.	African early-warning scholarship

SECTION 8 — CHEATSHEET

Core concepts	Key mechanisms	Theories and authors	Cases and evidence	Debates	Falsifiers	Cross-links
Scenario planning; Delphi; Bayesian updating; quantitative conflict modeling; calibration/scoring	Driving-force mapping; iterated anonymous elicitation; prior→likelihood→posterior; event-data ML; Brier decomposition	Schwartz (1991); Dalkey & Helmer (Delphi); Bayes; ICEWS; Tetlock and Gardner (2015); Brier (1950)	Shell scenarios; GJP 2011–2015; ICEWS/ACLED/GDELT; 2022 invasion forecasts	Experts vs. generalists; calibration vs. resolution; objectivity of models; early warning vs. prevention	A model that cannot be scored; a forecast with no prior; a scenario with no response plan	L3 causal inference; L4 epistemology; L14 transition; L59 negotiation; L60–61 capstones

SECTION 9 — REVIEW SCHEDULE

R1 (Day 1, 30 min): Re-read Section 2.2–2.6 bolded causal claims; recite Bayes’ theorem in words; compute one Brier score by hand.
R2 (Day 3, 45 min): Reconstruct the five perspectives on forecasting from memory; check against Section 2.7; note one gap in the error log.
R3 (Day 7, 60 min): Re-work the Problem Pair (2022 invasion) without notes; compare to model answer; identify where your prior sat.
R4 (Day 14, 45 min): Re-attempt the Hinge Question and two flashcards of each type; score yourself; update confidence audit.
R5 (Day 30, 90 min): Write a one-page memo applying one method to a new case; include falsifiers and perspective reconstruction; self-check description/explanation/prediction/prescription.

Methodologies in Geopolitical Forecasting

LESSON METADATA

SECTION 0 — Why This Lesson Exists

SECTION 1 — Pre-Class Priming

SECTION 2 — LESSON CORE

2.1 Learning objectives

2.2 KEY IDEA 1 — Scenario planning: Schwartz, the Royal Dutch Shell method, and its application to geopolitics

SECTION 2.3 KEY IDEA 2 — DELPHI METHODS: EXPERT ELICITATION, CONSENSUS BUILDING, AND THEIR STRENGTHS AND WEAKNESSES

SECTION 2.4 KEY IDEA 3 — BAYESIAN UPDATING: PRIORS, LIKELIHOODS, POSTERIORS, AND THE DISCIPLINE OF BELIEF REVISION

2.5 KEY IDEA 4 — Quantitative conflict modeling: ICEWS, ACLED, GDELT, and their uses and limits

2.6 KEY IDEA 5 — Calibration and scoring: Brier scores, the Good Judgment Project, and superforecasters

2.6 KEY IDEA 5 — Calibration and scoring: Brier scores, the Good Judgment Project, and superforecasters

2.7.1 African Perspective (PRIMARY)

2.7.2 Chinese Perspective

2.7.3 Russian Perspective

2.7.4 Western Perspective

2.7.5 UN Perspective

SECTION 2.9 PROBLEM PAIR (We do)

SECTION 2.10 INDEPENDENT MEMO (You do)

SECTION 2.11 HINGE QUESTION

SECTION 2A — DATA LAB

SECTION 3 — DUAL-CODED VISUAL ANCHOR

SECTION 4 — CORNELL NOTES

SECTION 5 — FEYNMAN COMPRESSION

SECTION 6 — SPACED-REPETITION FLASHCARDS

SECTION 7 — METACOGNITIVE CALIBRATION

7.1 Confidence Audit (score 0–100)

7.2 Three-Question Debrief

7.3 Error-Log

SECTION 8 — CHEATSHEET

SECTION 9 — REVIEW SCHEDULE

SECTION 10 — COMPLETION CHECKLIST

FURTHER READING (Optional)