Deep Conflict Research

Open Advanced / Investigate → Geopolitical Futures → Deep research. You can also open a profile in Conflicts & wars and select its deeper-research action.

What is included

The library now contains 178 authored conflict and war dossiers, including 32 additional historical cases. Each narrative field exposes its bibliography and editorial status. Separate descriptive evidence statements bind exact source records, measures, units and reporting periods. An attached source is not automatically proof of every causal interpretation in a historical narrative.

The new research API and workbench join existing UCDP annual records with reviewed dossier links and additional imports:

• UCDP Candidate Events: a monthly provisional edition. Initial qualification imported 16,633 records for January–September 2026. The dataset can contain aggregated reports, uncertain dates and representative locations. It is not a live military-position feed or a census of every incident.
• UCDP External Support 18.1: 10,852 supporter–recipient–opponent/year rows, 1975–2017, including source-coded no-support and alleged-support rows. Direction, support type, coalition, substate and both-sides flags remain visible. These do not establish current alliances.
• Existing annual UCDP evidence: 1,928 actor identities, 374 peace agreements, 1,786 stated-objective records, plus conflict/dyad/non-state/termination rows and the earlier dyad-year support projection. Their periods and overlapping units differ.
• UNHCR: annual origin-country refugee, asylum-seeker, IDP, return and other-protection categories. Initial import contains 7,404 field observations across 1,234 origin-years, 2020–2025, including missing values and explicitly reported zeros. Stocks and annual flows are separate.
• World Bank: GDP in current US dollars, real GDP growth, consumer inflation and ILO-modelled unemployment. Initial import contains 13,824 country/indicator/year slots, 2010–2025, including unavailable values. Negative growth and inflation are retained. These national statistics are not measured costs caused by a particular war.
• 51 historical/cultural map snapshots, 5000 BCE–2010 CE, from André Ourednik and historical-basemaps contributors, pinned to a source revision. The complete, unmodified GeoJSON files and GPL-3.0 licence are available from the map source notice. They are work-in-progress historical references, not authoritative borders or current territorial control.

Counts are dated qualification observations. These collections overlap and must not be added into a total of wars, independent events, or independently corroborating sources.

Select your research scope

Choose a profile, source association, country context and historical year range. Reviewed conflict links use explicit UCDP identity mappings and the profile's period. Broader country context deliberately selects geographically related records and is not an assertion that every record belongs to the same war. Country references also select humanitarian/economic context and recent reporting.

Negative calendar years mean BCE; there is no year zero. Closed historical profiles retain their endpoints. To explore later country developments independently, choose Global source exploration and the relevant country. Unknown links and source gaps stay visible in Coverage & freshness.

Dossier and claim evidence

The dossier retains actors, causes, triggers, resources, religion and ideology, outside powers, civilian impacts, settlement, timeline and limits. Each field has an expandable bibliography labelled editorial synthesis.

Source-backed descriptive claims separately identify exact event or statistical record IDs, reporting periods and values. For example, a UNHCR origin-country total is not a count of people displaced by one selected conflict. Later source evidence does not silently rewrite a background narrative or promote a ceasefire claim to verified compliance.

The original recent narrative backgrounds remain through their displayed cutoff, often 2024. The workbench adds subsequent 2025 annual observations, 2026 provisional evidence and separately dated reporting. Source retrieval in 2026 is not a fresh review of every historical explanation.

Timeline and maps

The timeline distinguishes authored milestones, qualifying annual source records and historical agreements. Source years do not imply uninterrupted fighting; absent years are not verified peace. Excel agreement dates are converted only when they agree with the supplied source year. Unparsed/undated milestones keep their wording without an invented day.

The map has two modes:

1. Provisional events: filter by source type, date interval and geographic precision. Event intervals overlap the selected date window; uncertain totals are not divided into invented daily counts. Markers show the source's exact, approximately 25-km, administrative, broad-region, country or international-location class. Larger/fainter markers indicate less precise source references. Click a marker to inspect the complete coded record and bibliography. Load additional pages when the map bound is reached.
2. Historical reference boundaries: browse the 51 discrete supplied snapshots with the selector or slider. The actual displayed map year is explicit. A timeline shortcut chooses a nearby supplied frame and labels that choice. No interpolation, precise modern-state equivalence or military-control claim is made. Cultural areas may overlap, and physical shorelines and rivers may have changed.

Actors, relationships, objectives and peace

Actor profiles retain source identifiers, original names and source-location context. Candidate-only identities are labelled where the full annual actor profile is unavailable. Current leaders or commanders are not inferred from an organization's historical name.

The graph distinguishes historical opposition, provisional event relationships, confirmed/source-recorded support and source-alleged support. Only the source-qualified support and one-sided-violence relations use directional arrows; opposing sides do not establish the initiator of a larger war. Undated predecessor/alliance/member links are displayed separately inside actor profiles. Drawing bounds are disclosed; the paged tables and exports retain the complete selection.

Stated objectives are actor claims and coding from the original 1989–2017 issue dataset, including undated founding records. They are not proven motives or population-wide religious beliefs. Peace agreements retain dates, named parties, original text links and coded provisions. A provision's existence does not establish implementation or current compliance.

Humanitarian and economic context

Select a country and chart measure. Missing observations break chart lines and remain unknown, while explicit source zeros remain zero. Source definitions, years and units are displayed beside each value.

UNHCR's IDP category includes conflict-generated people it protects or assists and does not cover all global internal displacement. Refugee stocks, asylum seekers, other-protection categories and return flows are not summed. These API totals do not supply complete UNRWA-mandate Palestinian refugee coverage. UNHCR annual-edition discovery advances only when an available publication announcement identifies the annual year; if the methodology page denies access, the reviewed annual edition remains pinned while its separately accessible API is refreshed.

GDP levels use current US dollars; GDP growth is real annual change. Inflation and unemployment have separate denominators and methods. No synthetic reconstruction-cost figure or causal war-loss estimate is generated.

Compare and explore conditional futures

Choose two to four profiles for a side-by-side comparison of all narrative dimensions and separately labelled source evidence. Shared countries and overlapping conflicts cannot be added as independent observations; no similarity or aggression score is implied.

Futures & indicators offers negotiation, continued confrontation and wider-involvement branches. Set explicit assumptions about diplomacy, external support, resource access and an illustrative horizon, then record what evidence still needs review. Each branch lists observable questions and links the currently selected evidence. Drafts are session-only and exportable; account/replay transitions clear them. No branch probabilities, inferred attacker, tactical routes or predicted casualty/victory totals are supplied. Closed historical cases are mechanism/counterfactual exercises. A shortcut opens the original strategic tabletop for an explicitly chosen country comparison.

Watchlists, alerts and Pro Brief

Follow / alerts saves a normal account-private My Watchtower watch using country context and topic-phrase matching. You choose whether to enable inbox alerts. The existing server evaluator runs while the browser is closed, establishes an initial baseline, and applies source qualification, duplicate suppression, cooldown and delivery budgets. Country/text matches are contextual evidence, not automatic verification of escalation or ceasefire status. There is no new email or external push subscription.

Saved watches can be paused, deleted, or managed in My Watchtower. Renaming a watch does not delete it; it remains in the native watchlist even if it no longer has the Conflict research prefix.

For exact statistical monitoring, choose Set source alert on a humanitarian/economic observation. The native editor binds the exact record, unit, reporting year and definition for threshold, material-change or source-staleness rules. Put in Pro Brief uses the existing owner-bound source reader and append workflow and allows you to keep researching in Geopolitical Futures.

Exports and freshness

JSON and CSV exports traverse all pages of the selected source view with a coherent native snapshot. A source change interrupts export rather than combining different editions. CSV text cells are escaped against spreadsheet-formula execution. Research print packs include the narrative, current panel and source metadata; use the browser's print dialog to save PDF. Complete large event/statistical ledgers remain separately exportable as JSON/CSV.

Coverage & freshness shows reviewed mappings, narrative cutoffs, annual years, candidate coverage, citations and gaps for every dossier. It is a coverage inventory, not an accuracy percentage. Source checks use their own due times and persistent denial/backoff controls; opening a view does not force upstream collection. The research collector is scheduled every six hours, with daily candidate checks, weekly statistical reads and monthly pinned historical-support refreshes. Ordinary reads retain original dates and disclose aged or unavailable sources.

The public read-only /agent/api/conflict-research dataset is available through the reviewed Watchtower data catalogue/MCP and native research tools. Public source data and account-private watches retain their separate access boundaries.

Attribution

UCDP: CC BY 4.0; Hegre, Croicu, Eck & Högbladh (2020) for Candidate Events; Meier, Karlén, Pettersson & Croicu (2022) for External Support; original annual-table citations and codebooks remain available in Sources & definitions. UNHCR population data: CC BY 4.0 except where indicated, with its category/coverage definitions. World Bank: CC BY 4.0 dataset terms and original-provider attribution. Historical-basemaps: André Ourednik and contributors, GPL-3.0, separate unmodified map data with source downloads, copyright/licence notices and no warranty. Original narrative references remain attached to their fields.