TDD Newsroom

The method running in public

A working project

TDD Newsroom is an experiment in news produced with AI from public records and official sources, under the Directed User Interface. Its aim is a news service that follows the record wherever it leads, labels how strongly each thing is known, and shows its work.

It is also a live application of the method: every item is sourced, labeled for certainty, placed in time, and logged when it is released.

How a story moves

1. From the recordItems begin in the public releases of government agencies and international organizations, and each links the sources it rests on.
2. Drafted, then checkedAn AI reporting pass drafts the item; a separate AI editing pass checks it and judges its certainty on its own.
3. Released by a personA human editor releases every item. Nothing reaches readers any other way.
4. Corrected in the openA correction is dated on the story itself and announced in the feed as its own entry, as visible as the original.

An item is Confirmed only when its key facts were read in the primary record; otherwise it is Reported, with the source named. Each says what the record does not yet show. Every release is logged with the time and a fingerprint of the published text. There is no advertising.

What it is not, yet

• Not autonomous. A human editor releases every item. Automated agents are planned as a later phase, each to be tested against the editors' own record before it runs on its own.
• Not reporting from the scene. It interviews no one and works only from what public sources publish, so it is as good as those sources, and it says which ones it used.
• Not infallible. It makes mistakes. The corrections are part of the record, and the record is the point.