How Dispatch works

The model

Dispatch is AI-assisted research with human editorial review. Software does the reading at scale — dockets, sanctions lists, NGO reports, regional press. Humans decide what is true enough to publish. Nothing reaches a page without a named primary source behind it.

Confidence grades

A — Documentary. The claim rests on a primary government or institutional document: an indictment, a sanctions notice, a regulatory rule, a published UN or major-NGO report.

B — Corroborated reporting. The claim rests on reporting by established outlets or investigative consortia, consistent across sources but not yet confirmed by a primary document.

C — Indicative. The claim is plausible and worth tracking — single-source reporting, unverified testimony, or pattern inference. We publish grade-C material only when we say plainly that it is grade C.

What we will not do

We do not publish numbers we cannot count or cite. We do not present allegations as convictions — charges are charges, designations are executive findings. We do not name victims. We do not list secure channels that do not exist yet. And when we get something wrong, the correction is published on the page where the error lived.

Status of every claim

Each dispatch ends with what we cannot yet verify. The register in the Compound Atlas labels its statuses as editorial judgments from the cited record, not live intelligence. The Source Library holds every primary document we rely on.

Corrections

Found an error? Tell us. Corrections are acknowledged in place, with the date and nature of the change. Corrections are published, not buried.