> ## Documentation Index > Fetch the complete documentation index at: https://docs.gdeltcloud.com/llms.txt > Use this file to discover all available pages before exploring further. # Atlas Overview > Atlas is GDELT Cloud's family of geopolitical indices — a glass-box way to ask how tense a place is, what condition it is in, how capable it is, and where it is heading, computed entirely from our own coded events, narratives, and resolved entities. **Preview.** Atlas is in internal preview (Admin plan) while we validate baselines and coverage. Response shapes, coverage, and band cut-points may change before general availability, and readings are labeled `provisional`. Each layer's page states its honest limits — including the unflattering ones. ## What Atlas is **Atlas** is a family of indices that rate and index the world — continents, regions, countries, and later entities — using **only GDELT Cloud's own derived data**. Most geopolitical risk products hand you a number and ask you to trust it. Atlas is built the other way around: every index is a documented arithmetic path from primitives you can query yourself. If a country reads 180, you can pull the events and narratives that produced it and check the work. ## Why you should care Four things are genuinely hard about geopolitical data. Each Atlas layer exists to solve one of them. Ten protests in Paris and ten in Mogadishu do not mean the same thing. **Atlas GPR** normalizes every place against its own baseline, so a spike means something happened — not that a place is large and well-covered. A long war becomes a country's normal, and a deviation index reports it as unremarkable. **Atlas Posture** scores against fixed reference points, so a war reads as a war however long it has run. Media volume tracks English-language news interest, not strategic weight. **Atlas CNP** measures the capability a state can actually bring to bear. A firm's risk lives in its facilities, routes, and counterparties, not its domicile. **Atlas Vane** indexes entities and questions rather than places. And one commitment that spans all of them: **a number without provenance cannot be acted on.** Every Atlas reading drills to the underlying coded events, narratives, and sources, so an index is an entry point into evidence rather than a replacement for it. ## What it is useful for Atlas is built for workflows where "something changed" has to become "something changed, here, and here is why": * **Tripwires and monitoring.** Alert when a place moves materially above its own norm, then drill straight to the incidents behind it. * **Cross-sectional screening.** Rank the countries you actually operate in by condition rather than headline volume. * **Macro and commodity context.** Geopolitical risk is an established input to oil, defense, and risk-premium models; Atlas provides a multilingual, per-country, drill-through version. * **Narrative-to-asset attribution.** Because Atlas runs on the same primitives as our facilities, filings, and entity data, a regional reading joins to the specific assets exposed to it. **Read each layer's status section before building on it.** Atlas is young — our consistently coded event history begins in **March 2026**, which bounds every layer. World and continent readings are materially better supported than country readings, and today only a handful of countries clear the per-country coverage floor. The per-page status sections give exact numbers. ## The family **Live (preview).** Geopolitical Risk — how tense a place is versus its own norm, through an Attention lens and an event-grounded Events lens. **In development.** Country condition — a dynamic axis that moves with events and a structural axis that moves with institutions. **Planned.** Comprehensive National Power — the capability a state can actually bring to bear. **Planned.** Entity- and question-conditioned readings — exposure and stance for a specific company or actor. **Planned.** Dyadic escalation forecasts, shipped behind a public accuracy scorecard. ## Where the data comes from Every Atlas index is built on the same pipeline. No external index is ingested or resold. We ingest the public, multilingual GDELT article stream — worldwide news in many languages, not just English. We cluster those articles into **Stories** — one cluster per real-world narrative — each with an `article_count`. This is the substrate for the Attention lens. Surfaced by `GET /api/v2/stories`. From those clusters we generate structured **Events** with CAMEO+ / conflict coding, Goldstein tone, and a significance weight. This is the substrate for the Events lens. Surfaced by `GET /api/v2/events`. For the slow axis we load public annual indicators — World Bank WDI/WGI and V-Dem — into a per-country indicator store. Atlas aggregates these into daily readings per geography, normalizes against the stated reference, bands the result, and serves it via `GET /api/v2/intelligence/*`. ## Principles every layer shares These are enforced in the serve layer, not aspirations. Where coverage is too thin to be reliable, a reading is **withheld** (`insufficient_data`) rather than guessed. A `0` from Atlas means a measured zero — a well-covered place with genuinely nothing happening — not an absent measurement. This is why large parts of the country map are blank, and blank is the honest answer. Every index is a documented path: pick a qualifying subset, divide by a denominator, normalize against a stated reference, band the result. No learned weights, no undisclosed blending. Each layer's page states its numerator, denominator, and reference explicitly. Where a layer scores condition or capability, it anchors to fixed reference points rather than ranking countries against each other. A cross-country rank pins any world aggregate near the middle by construction and lets a country improve only because another deteriorated. See [Atlas Posture](/features/atlas-posture) for the full argument. Observations are vintaged. An as-of reading reflects only what was knowable on that date, so a backtest cannot accidentally consume information from the future. Atlas does not re-read the news. It aggregates signals already produced upstream — coded events, clustered narratives, resolved entities — into shares and normalizes them. All classification happens in the coding layer, which is documented separately and independently auditable. Atlas is **inspired by** established indices but does not ingest, license, or redistribute any of them. Every number is computed from GDELT Cloud data. ## Influences Atlas sits in an established tradition of quantitative geopolitical measurement. We build independently, but we name our influences. Each layer's page lists the works specific to it. "Measuring Geopolitical Risk," *American Economic Review* 112(4), 2022. The share of articles in a fixed newspaper panel discussing geopolitical risk, normalized to 100 over a base period, with Threats and Acts subindices and country-specific GPRC. The direct ancestor of [Atlas GPR](/features/atlas-gpr). The newspaper-share methodology GPR itself builds on, and the template for text-frequency indices generally. "A Global Model for Forecasting Political Instability," *AJPS* 54(1), 2010. Found regime type — partial democracy with factionalism — the strongest single predictor of instability. Shapes the structural axis of [Atlas Posture](/features/atlas-posture). Expert-coded democracy indices for most countries back to 1789. Supplies the democracy dimension of Atlas Posture's structural axis. The canonical quantitative national-power measure, and the methodological ancestor of [Atlas CNP](/features/atlas-cnp). The open global news stream our pipeline draws on. GDELT is an **input**; the events, clusters, and entities Atlas is computed from are coded by us on top of it. ## API The Atlas GPR endpoint (preview). Parameters include `level`, `geo`, `variant`, `component`, `window`, and `date_*`. Returns the reading, band, `tension_share`, coverage, and the baseline used — or `insufficient_data` where coverage is thin. Try it in the API Arena.