A short, honest account of where the numbers come from, how often they refresh, which canonical scores PRISM relies on, and the caveats that come with each.
PRISM is not a snapshot. It's a live instrument that pulls from authoritative APIs on schedules ranging from quarter-hourly to daily, persists every pull to disk, and computes the country-level headlines on demand from the most recent reconciled records.
Architecturally the platform rests on four pillars. None of them are clever; together they make the difference between a dashboard you trust and a dashboard you have to second-guess.
Every long-running data source — ReliefWeb, ACLED, IOM DTM, IDMC, OECD DAC, ACAPS, the FTS funding cache — has its own SQLite database on a persistent disk. Concurrent reads stay fast; writes don't block dashboards.
A single Excel data_store snapshot acts as the canonical baseline for cross-source joins. Every API pull layers on top, never replacing ground truth without an explicit refresh cycle.
Daily ReliefWeb pulls do not overwrite — they accumulate. The archive grows by ~30 reports a day across the 74 HPC HNO countries, with full provenance, retraction handling and admin moderation.
Country headlines (PiN, funding coverage, severity classifications) are recomputed at request time from the latest reconciled records. There are no stale precomputed figures sitting in the page — only canonical sources resolved live.
The canonical scope is the OCHA Humanitarian Programme Cycle's HNO country list — the 74 operations with a Humanitarian Needs Overview in the current cycle. Several datasets reach further (climate, ODA, EU migration), and PRISM exposes that wider coverage where it adds analytical value.
Country identity is reconciled to ISO-3166-1 alpha-3 codes on ingestion. The 74-country scope is hard-coded in the data processors so the daily refresh runs the same surface every cycle, regardless of which agencies happened to publish that day. Datasets with broader native coverage (OECD DAC's 33-donor universe, ACLED's 198-country event archive, Copernicus' global ERA5 reanalysis, IDMC's 200+ country displacement database) are kept at their native resolution and joined on request.
This methodology covers the public analytics layer. Project-level field data is handled separately by PRISM Project, which anonymises records on ingest, encrypts them end to end, and keeps each project private to its assigned team. Only aggregate, anonymised findings are surfaced.
Cadence varies by source. The table below is the operational schedule — what PRISM actually pulls, and how often.
Of the 40+ datasets PRISM ingests, twelve carry the analytic weight on the country pages. Each card lists the issuing institution, refresh class, contribution and an honest caveat. The remainder are summarised below.
The canonical source for People in Need, plan requirements and cluster/AoR funding. Drives the headline figures across every country profile.
PiN figures shift between draft, final and revised HNO snapshots; PRISM uses the most-recent published value.
Reported funding flows for every appeal. Per-donor explore views match FTS published totals exactly — same donor names, same currencies, same scope rules.
FTS captures reported, not actual, disbursements; late or unreported flows can shift coverage by several points.
Situation reports, press releases, assessments and analyses from humanitarian organisations. The archive (~3,100 reports, growing daily) is the substrate for full-text search and the weekly LLM situation summaries.
Coverage depth varies by country; well-covered crises generate more reports than equally severe but under-reported ones.
Monthly political-violence and civilian-targeting event counts, with fatalities. Admin-1 / Admin-2 resolution for HRP countries; trend lines for all.
Open-tier counts are aggregated, not event-level; event detail requires ACLED's research-tier subscription.
Validated annual stocks (people displaced at year-end) plus preliminary event-level Internal Displacement Updates from the last 180 days. PRISM keeps stock and new-displacement metrics rigorously separate — they cannot be summed.
Recent-event figures are preliminary; reconciled annual totals supersede them.
Live protection-event monitoring: abduction, attacks on civilians, GBV, forced displacement and 11 more categories. Each event carries a justification and attribution. Country pages show the three most recent; full lists are CSV-exportable.
ACAPS coverage depends on field-network presence; absence of events in a country is not absence of risk.
Composite humanitarian-severity classification per country (Very low → Very high) combining impact of the crisis, conditions of affected people and complexity. Used as a canonical severity tag, not blended with other scores.
Severity moves slowly; sudden-onset events may not appear until the next quarterly update.
Composite of hazard exposure, vulnerability and lack of coping capacity. PRISM surfaces the index and the three constituent dimensions side by side so users see what's driving the score, not just the headline.
A forward-looking risk score, not a real-time severity indicator; cross-reference INFORM Severity for current conditions.
Internal-displacement totals by cause and by sub-national admin-1, where DTM is operational. Complements IDMC's stock view with operational granularity at the sub-country level.
Coverage is operation-dependent; some countries with active displacement have no DTM presence.
Official Development Assistance — $174B+ across the DAC universe (2025 preliminary). The "Rise and Fall" view plots per-donor ODA volume, year-over-year change and the ODA/GNI ratio against the 0.7% UN target.
DAC reporting lags by one year; "preliminary" figures are revised in the spring of the following year.
Authoritative historical disaster archive. PRISM uses EM-DAT for the long-run climate and disaster lens — affected populations by year, butterfly breakdowns by disaster type, and CSV export for downstream analysis.
EM-DAT requires events to meet inclusion thresholds (≥10 deaths, ≥100 affected, declaration of emergency, or call for international assistance).
Systematic protection-severity scoring across protection risks and countries. Powers PRISM's dedicated Protection Severity page; complementary to ACAPS's event-level view.
Severity scoring reflects validated cluster assessments; not all 74 HPC countries are scored in every release.
Beyond the twelve sources above, PRISM ingests the following without dedicated editorial framing — they appear inside thematic pages, deep dives or country profiles where they're relevant.
Every country page surfaces canonical scores from their issuing institutions — INFORM Risk and Severity from the JRC and ACAPS, IPC phase classification from the IPC Global Support Unit, protection severity from the GPC. PRISM does not aggregate them into a single index.
This is deliberate. Composite scores hide their assumptions; users see a number, not the trade-off behind it. PRISM's editorial line is to keep the canonical scores side by side — INFORM Severity next to INFORM Risk next to IPC Phase 3+ population next to ACAPS Access Constraints — and let users see the disagreements between them rather than averaging them away.
Where a single picture is genuinely useful — the country-profile snapshot tab — the composite is built on demand from the constituent scores, with the contribution of each clearly attributed in the UI. There is no PRISM Score that exists outside the canonical inputs that produced it.
A platform that integrates 50+ sources will inherit every limitation of every one of them. The five below are the ones most likely to matter when you're interpreting a number on a PRISM page.
PRISM is released under CC BY-SA 4.0. Charts, deep-dive analyses and derived figures may be reused and adapted with attribution; downstream work must carry the same licence. Underlying datasets retain the licences of their issuing institutions — a chart of OECD DAC ODA is CC BY-SA when published in PRISM, but the source data is governed by the OECD's terms of use.
Provenance for every published figure is traceable to a named, dated source on the page where it appears. The methodology log behind any specific number is one click away — cards link to the issuing API, badges identify the data class (Live / Snapshot / Cycle), and timestamps record the last successful refresh.
PRISM exists because humanitarian decision-making deserves better than fragmented dashboards and one-source narratives. But the platform also reflects on its own embeddedness in the data architectures it analyses. The categories — who counts as 'refugee' versus 'migrant', what constitutes 'protection risk', which actors produce 'legitimate' knowledge — are not neutral inputs. Surfacing canonical scores side by side, and refusing to flatten them into a single composite, is itself a methodological choice with political consequences.
Every number on this platform is as reliable as the source it came from, no more and no less. PRISM surfaces the disagreements between sources rather than averaging them away.
Questions, corrections, or a source PRISM should ingest? [email protected].
PRISM uses large language models in four narrow, disclosed places. None of them produces a number. The platform is fully functional with the models switched off, and every AI-assisted output is grounded in data PRISM already holds.
The bright line: no quantitative figure on PRISM is generated or modified by a language model. Every number flows deterministically from the issuing API or dataset through open-source processing code to the chart that displays it. Where a model is unavailable, the platform degrades gracefully — the summary block is simply absent, and everything else renders unchanged.
PRISM is created and edited by Umutcan Yüksel and began, in October 2025, as a research project under the Policy Leader Fellowship at the School of Transnational Governance, European University Institute. It is owned and operated by Integrity Consulting and published as a free, open-access platform. The editor's full profile is on the about page (ORCID 0000-0002-3565-3568); editorial correspondence goes to [email protected].
Editorial standards. Every published figure carries a named, dated source. Estimates and modelled ranges are labelled as such; provisional figures are flagged and revised as sources consolidate. Analysis of ongoing crises follows a fixed editorial specification covering sourcing, tone and sensitivity.
Corrections. Errors are corrected in place as soon as they are established, and material corrections are noted on the affected page. Anyone can flag an error at [email protected].
Independence and funding. PRISM is self-funded through Integrity Consulting. It accepts no donor funding, carries no advertising, and has no institutional stake in the crises it monitors. Source data is used under the terms of the originating agencies, and access to the platform is free.
Methodology v4.2 · last reviewed 4 July 2026 · maintained by the chief editor
PRISM data and analysis already anchor peer-reviewed and institutional research. The platform is designed for citation: stable URLs, per-dataset citation guidance, and machine-readable provenance on every page.
Publications built on PRISM data:
Citing PRISM. For the platform: PRISM (2026). Protection, Risk & Impact Severity Monitor. https://www.prismonitor.eu — each dataset in the data catalogue carries its own suggested citation with attribution to the originating agencies, and every deep dive is citable via its stable URL, publication date and author metadata.
The full country dashboards and the public country hubs no longer require registration — severity, funding, displacement, food security and protection for 230+ countries.
An AI analyst whose answers are retrieved from PRISM's live datasets and report archive at the moment you ask — never from memory. Every figure carries a linked citation to its exact source, every answer ends with a “Reproduce this” trail of the queries it ran, and when the data can't answer, it says so.