WATCHTOWER / METHODS ·20OCTOBER2026

Geopolitical Futures

History, resources and alternative strategic paths, with the original evidence and assumptions in view.

Start with a question

Open Geopolitical Futures beside War Room. Choose a country pair, future reference year and map measure. The seven views connect historical parties, geography, demographic/resource scenarios, strategic branches and model validation. The complete country directory is not complete source coverage for every field.

Historical sources and exact roles

UCDP/PRIO Armed Conflict Dataset26.1 and Dyadic Dataset26.1 cover1946–2025. Non-state and within-country organized-violence records cover1989–2025. Onset and termination sources end2024; stated issues and external support end2017; peace agreements end2021. Record units overlap: conflict-years, dyad-years, actors, statements, agreements and country-years cannot be added as unique wars. SideA/B is not an attacker/victim classification. Primary and secondary opposing sides remain explicit; allied co-participation is not coded as opposition.

Source eligibility and country history

The UCDP country-year model target is source-coded state-based event presence within borders, not the government’s decision or a bilateral attack. It derives from eligible UCDP GED events: a boolean1 needs an eligible recorded event, not25 deaths in that particular country-year. Zeros may coexist with violence excluded by source thresholds/coverage; they are source zeros, not verified absence of all violence. Source country/state conventions and reviewed polity breaks constrain consecutive histories. Historical polities without a defensible modern substitution remain unmapped rather than reassigned.

Terrain, land, disputes and beliefs

Existing country encyclopedia public-domain archived fields retain their original text, source revision and reference years. CIA retired Factbook publication on4February2026. Terrain, borders, rivers, basins, aquifers, resource descriptions and dispute wording are reference evidence, not verified present control or sovereignty adjudication. Historical actor-stated objectives may include political or religious goals. Population religious affiliation, ethnicity and age structure are not used to infer aggression or as numerical model predictors.

Earlier historical mechanisms

Twenty-four selected source-linked case studies reach back to431BCE. They provide original analytical summaries and counterpoints, not an exhaustive pre1946 global panel. Twenty-six authored regional study groupings offer starting points; their names and map foci do not establish a currently active conflict. Correlates of War data is not ingested/trained because its publisher expressly restricts commercial use and redistribution without permission.

Population and resource scenarios

World Bank source40 supplies the source-selected population estimates/projections through2050 and under15/65+ shares. No low/high source variants are invented. World Development Indicators supply dated land/water/agricultural and World Bank resource-rent measures. Source years, units, definitions, provider update and retrieval clocks stay distinct. Aggregate rows are excluded, all selected native pages are checked, and missing values remain unknown.

Exact scenario arithmetic

Scenario population = source population at target year ×(1+user population deviation/100). Arable hectares/person = reported arable hectares at its anchor year ×(1+area-change/100) ÷scenario population. Internal renewable water/person = reported billion m³ internal flow ×1e9 ×(1+supply-change/100) ÷scenario population. Hypothetical water stress = reported stress at its anchor year ×target population /source population in that anchor year ×(1+withdrawal-per-person-change/100) ÷(1+renewable-supply-change/100). Each anchor is shown. This is conditional arithmetic, not a climate/hydrology model, carrying-capacity estimate or crop-yield forecast. Internal freshwater omits international inflow; arable area is not food supply; resource rents are neither fiscal revenue nor recoverable reserves.

Strategic tabletop

The user chooses countryA/B, a study or custom pair, a hypothetical trigger and third-party involvement. The evidence brief shows actual historical relationships, dated resource anchors, scenario quantities and archived terrain/dispute wording. Up to12 explicit strategic moves explore talks, access/resource agreements, pressure, assumed limited armed confrontation and standing down. The finite-state transitions are authored teaching/research rules, not calibrated outcome probabilities or recommendations. The first force-choice determines an explicitly hypothetical initiating actor. Negotiated access, contained confrontation and wider spillover remain alternatives. No operational targeting, routes, force optimisation, victory, casualties or duration is calculated.

Historical transition model

The fixed interpretable model uses the previous year’s observed state-based event presence and the number of positive years in the complete previous five-year window, grouped0/1–2/3–5. Groups receive beta(1,1) smoothing; groups below30training cases fall back to frozen training prevalence. A five-bit state distribution propagates exactly for each annual step. No future labels, current headlines, demographics, religion or user resource shocks enter this model. Source zero is used only where the country-year panel supplies it; no missing-country zero imputation.

Temporal validation and future leads

Validation trains through2009 and evaluates later outcomes2010–2017. Held-out testing trains through2017 and evaluates later outcomes2018–2025. For multiyear leads, origin must be at/after the training cutoff; only observed history through origin is used. Leads1,2 and5 are evaluated separately against frozen prevalence and a stated0.99/0.01persistence comparator. Brier score, log-loss, sample counts and reliability bins are exposed. Source vintage is current-retrospective, not historical as-issued data; errors are clustered within countries/conflicts. Historical Wilson bands describe group-frequency sampling uncertainty, not calibrated future prediction intervals.

Publication and prospective evidence

A probability is displayed only for an exact tested lead with at least300held-out cases and lower Brier/log-loss than the prevalence baseline. Persistence comparison remains visible. This gate is not prospective validation. Inputs through2025 make2027 a two-step projection and2030 a five-step projection; longer unqualified leads stay unknown while demographic scenarios remain available through2050. New future records have actual issuance times in an own append-only ledger; no backdating or rewriting old Watchtower forecast/financial/source records. Initial prospective resolutions are zero. Probabilities concern event presence inside borders, including ongoing conflicts, not who will attack whom.

Refresh, export and access

The app reads a bounded retained public snapshot. Its button reloads that snapshot without forcing publishers. The scheduled collector checks daily, with seven-day World Bank and thirty-day pinned UCDP cache lifetimes, source-denial pauses and persistent backoff. Original capture/observation clocks stay visible. History pages use native offset/limit/snapshot; a changed snapshot returns409. MCP field paging is separate. Scenario drafts are session-only and owner/replay-fenced; export preserves assumptions and snapshot. Normal account permissions and native controls apply.

Source downloads and attribution

UCDP datasets: CC BY4.0 with dataset citations. World Bank: Dataset terms, CC BY4.0 and provider attribution. Natural Earth display boundaries: public domain. Archived CIA reference: public domain, pinned source links retained. No source institution endorses these scenario calculations.