Reports

Policy briefs in the reports library should answer a bounded decision question with named evidence, not offer a general position paper. RDNA5 gives a dated reconstruction-needs baseline through 31 December 2025, OCHA's 2026 plan gives humanitarian need and planned reach, IMF programme documents give macroeconomic assumptions, IAEA updates define nuclear-safety preconditions, and UN Charter Article 33 gives lawful peaceful-settlement routes. A brief can connect those records, but it cannot treat one of them as proof of everything.

A concrete report should therefore state the decision it addresses, the source period, the affected population or sector, the uncertainty and the policy options that follow. For example, a reconstruction brief should not reuse RDNA5 as a final war-cost number; a nuclear brief should not turn IAEA restart conditions into a meltdown forecast; and a diplomacy brief should not cite Article 33 as proof that any specific settlement is workable.

Evidence updates should be short, dated records of what changed in the evidence base. OHCHR's first-half 2026 casualty update, OHCHR's winter energy-attack report, IAEA power-line reporting, Kiel's August 2026 aid tracker update and the IMF Article IV assessment each use different reporting cycles and definitions. A useful update explains those differences instead of blending them into one trend line.

The update format should show the new figure or finding, the previous baseline if available, the source limitation and the content pages affected. A casualty update should remain a verified minimum. An aid update should distinguish commitments, allocations and deliveries. A macroeconomic update should keep IMF assumptions visible, because financing, attacks, energy conditions and policy implementation can change the interpretation.

Datasets should publish structured records only when the field definitions are clear. IOM displacement and return figures depend on a collection window, Eurostat temporary-protection data is a dated EU snapshot, Kiel aid data separates financial categories, IAEA PRIS lists reactor units and facility status, and OHCHR monthly casualty records are verified minimums. Those datasets cannot be merged safely without preserving their original units and definitions.

A downloadable dataset should include source ID, publisher, publication date, accessed or checked date, geography, period, value type, confidence and limitation notes. If a table mixes people, euros, reactors and casualties, the schema must make the difference obvious. The purpose is reproducibility: a reader should be able to trace every row back to the public record that supports it.

Review status should tell readers how far a report has moved from collected evidence toward publication. High-confidence official assessments such as RDNA5 still require date and scope review. Medium-confidence methodology sources such as Mediazona's named-loss database need careful wording about incompleteness. Policy research from Chatham House and scenario work from S&P Global require argument-boundary review, while WHO health-care reporting needs checks against overgeneralising facility or clinical claims.

The status label should therefore identify the review task: source boundary, data category, legal attribution, translation, sensitivity, update date or publication approval. A report should not be marked ready because it sounds polished. It is ready only when the reviewer can point to each factual sentence, the source that supports it, and the limitation that prevents overclaiming.