About
What this site is and how it works.
SignalDispatch is a field guide to the UAP record: the cases, the programs, the claims, and the evidence behind them. It takes no side. Statuses attach to individual claims, never to people or camps, and every assessment links the sources it rests on so readers can weigh the evidence without borrowing this site's judgment.
The sections
The daily dispatch covers the last 24 hours plus one piece of history, checked before publication. The field guide holds the big cases as individual assessed claims, with a timeline view of the whole record in date order. Open files hold the unsettled questions, with every serious explanation scored against the evidence. The observatory is real radio astronomy — survey data today, a backyard hydrogen-line telescope next. The source register lists every source the site relies on.
How daily publication works
A writing model receives a bounded evidence packet: selected field-guide material, new primary documents, and current headline metadata. It cannot add source URLs. A separate model family then checks each externally verifiable statement against the same packet.
Publication fails closed. Unsupported or uncheckable claims, invented links, non-verbatim quotations, overconfident attribution, or a malformed review block the post. Passing this process reduces risk; it does not turn an AI system into an infallible authority.
How statuses work
Statuses attach to individual claims. A government conclusion is recorded as an official finding. A witness statement is recorded as testimony. Neither label proves the underlying proposition. "Unresolved" and "unsubstantiated" are different: one describes an open investigation; the other says the public evidence does not presently establish the claim.
Corrections
Knowledge entries carry a review date and this visible revision log. A materially changed claim keeps its earlier context in version control and receives a new revision entry here.
Opened the public field guide with eight cornerstone topics and claim-level evidence statuses.
Models
The current writer is deepseek/deepseek-chat-v3.1. The independent checker is anthropic/claude-sonnet-4.6. The pipeline refuses to use the same model for both roles.