How It’s Made

Produced by AI. Directed by humans. The method behind every desk, shown in full.

Why we’re showing our work

Noise Removed — the daily brief and its fifteen-minute spoken edition — is produced by artificial intelligence under human direction. We could be vague about that. We’re not going to be. A reader’s trust in any news product depends on knowing what produced it: who chose the stories, who wrote the words, who checked the facts, who decides what publishes, and who answers when it’s wrong. This page says all of that.

If what you read here doesn’t meet your standard for trustworthy news, the right answer is to read and listen skeptically, or not at all. If it does, the trust is earned rather than assumed.

Journalists provide the signal. We remove the noise.

We do not replace original journalism. We could not exist without it. Every story in the brief draws on reporting done by working journalists at named outlets; every story’s Sources line names them. Their work is our raw material.

The brief’s job is to take reporting from across the political spectrum and strip the framing, the spin and the partisan packaging that drowns it out — and then show you the framings, named, beside the record, so you can see what was removed. The signal is theirs. The discipline of removing the noise is ours. Both are required; neither alone produces what you get.

The Spin Index

Every story carries a Spin Index — a 0–100 gauge of how much framing it arrived wrapped in: the loaded language, the framing, the agenda. 63 · Heavy Spin  The number is the coverage before we cleared it; what you read is the story with it removed. The three most spun stories become the ones told in full on the air. How the index is scored →

Who does what

Produced by AI. Directed by humans. That credit closes every episode and every brief, and it is literal.

Directed by Chad Kaul. Chad designed the method and directs the operation. He wrote the editorial mission, the criteria stories are scored against, the structure every story is built on, the disciplines that govern verification and tone, and the rules for what the AI may and may not do. He makes the calls the rules don’t cover, he listens to every episode before it publishes, and he answers for every word.

The desk is the newsroom’s editorial voice: short declarative sentences, concrete diction, attribution over adjectives, the framings laid out and the record laid beside them. It is not a person. It is the register everything here is written and spoken in, and it is the desk that answers for the product: republicdesk@signalmedianews.com.

Produced by Claude. The daily work — the research, the fact-checking, the scoring, the drafting, the self-review, the script — is done by Anthropic’s Claude. The production model is named in every byline; when it changes, the byline changes, because different models produce different work.

Hosted by an AI host. Each desk’s episode is read by a host designed by that desk — for America, James Calder. He is not a person. The words he speaks are the day’s brief, rendered into a script — he adds nothing to it, holds no opinions about the news, and reads no endorsements. A person listens to every episode before it goes out.

The method, every morning

1. Reads the day’s reporting from across the political spectrum — a hundred articles or more, in full, left, center and right — so the pool of candidate stories is built from every side’s front page, not one.
2. Selects stories by what they mean for you, scored against a fixed set of civic criteria, with a diversity rule so no single topic dominates the day.
3. Separates the record from the framing. For every story: what happened, why it matters, the background, who is affected, what is disputed — then each side’s framing, named and attributed, then the factual record with a plain confidence rating. We do not manufacture symmetry: where the evidence is one-sided, we say so.
4. Measures the spin. Every story carries its Spin Index.
5. Traces every fact. No name, number, date or outcome comes from the AI’s memory. Each is traced to a cited source or a primary document — or flagged. What can’t be traced doesn’t publish.
6. Checks itself for bias as it writes, under a roster of named disciplines, and counts every application.
7. Writes the brief, then the script. The full brief by 6:40 Eastern. The spoken edition by eight.
8. Shows the work. The receipt at the foot of every page and in every spoken sign-off: articles read, sources, facts checked, biases unspun — each counted by a tool from a written ledger, never asserted.

The episode

Fifteen minutes, weekday mornings and Sundays. Six segments, the same every day: The Day in a Minute · The Top Ten · The Next Five · The Lead and the Loudest — the lead story and the two most spun, told in full with every side’s framing read and named · The Questions — four things the record can’t answer yet, addressed to the people who could · Worth Discussing Today — three facts to take with you. Then the receipt, spoken. Then the credit.

A material error is corrected out loud at the top of the next episode, in a fixed slot, and in the notes of the episode that carried it. No episode is replaced or quietly edited.

What we don’t do

• No opinion columns, no editorializing. The brief and the host report; neither has a take.
• No anonymous sources. Every claim traces to a named outlet or a named primary document.
• No AI-generated images of real people or places, ever. The brand’s images are instruments — a spin meter, a chart of the record — never a synthetic photograph.
• No manufactured symmetry. A well-documented claim is never presented as the equal of a thin one for the sake of balance.
• No endorsements. The host endorses nothing — not a product, a candidate or a cause. If a sponsor is ever read on the air, it is read as a sponsor’s message and labeled as one; political buyers are not accepted.
• No undisclosed AI. Every brief and every episode carries the credit and the model’s name.

How to verify any claim

Every story closes with a Sources line. Find the story, read its Sources line, look up the cited outlets’ coverage, compare. If a claim doesn’t match its sources, that is an error, and we want to know: republicdesk@signalmedianews.com.