Docket tracks federal and state legislative activity and makes it searchable: bills, votes, amendments, riders, spending awards, executive actions, and the procedural moves that rarely reach the news. Every fact links to an official government record.
Neutrality is a build constraint, not a promise
Most trackers claim neutrality in an about page and then let editorial judgment back in through ranking, word choice, and comment sections. Docket pushes the constraint into code so it fails loudly when violated.
Government sources only. Every displayed fact links to an official government URL. No think tanks, no news outlets, no secondhand data.
No adjectives in generated text. Numbers, names, dates, and status. "Reduced by X dollars" passes. "Slashed" does not. A banned-words lint fails the build.
Ranked by official action. Feed order comes from date of legislative action, dollar magnitude, and affected-population estimates from agency data. Never from news coverage or social shares.
No comment sections. Comments are the largest bias-injection surface in a product like this. The only feedback accepted is a yes or no on whether a summary was clear, never on whether a bill was good.
Summaries are field extraction. Prose is assembled from structured fields in official bill text rather than generated freely. High-salience bills are tested against a polarization measure before they go live.
Why it works this way
A neutrality rule a human has to remember is a rule that erodes. A neutrality rule that breaks continuous integration is one that holds. The interesting engineering here is not the ingest pipeline, it is deciding which editorial decisions can be removed from a human's hands entirely and then removing them.