The document that governs Mercury Local has eleven rules. I didn't write any of them until the system made the mistake each one prevents.

Rule V8 exists because an article stated an arrest happened "on Tuesday night." The arrest date was April 8, 2026 — a Wednesday. The error made it through drafting, self-critique, and a fourteen-point fact-check table without being caught, because the day was assumed rather than calculated from the date.

Rule V11 exists because a race recap was nearly written for a race scheduled to start at 7:30 PM that evening. The editorial calendar showed the target date. The target date was treated as confirmation the race had already run. It had not.

Eleven rules. Eleven mistakes that were uncomfortable enough to write down. The document is the length it is because the operation has been running long enough to fail in eleven distinct ways.

The Policy Is Not a Setting

Most conversations about AI governance treat it as a configuration problem — parameters, guardrails, output format, test prompt. That is setup. Governance is what happens when the system runs at production scale, when a piece needs to go out at midnight, when the context is stale, when the pipeline is deep in a production run. That is when the real rules surface, because that is when the real failures happen.

A skeptic reading a list of eleven failures does not see a learning system. They see a system that has produced eleven errors serious enough to require documented rules, and they wonder what error twelve looks like.

Eleven documented failures are better than zero — not because the system failed less, but because zero documented failures means the failures are not being recorded. An error that produces a new rule is an error the system will not make the same way again. An error that produces nothing is an error that compounds. The governing document for this operation is not a record of success. It is a record of correction.

One Document, Four Publications

This is the difference between governance and compliance: compliance is checking a box; governance is encoding your failures somewhere the next session can find them.

Mercury Local currently runs The Charlotte Mercury, The Farmington Mercury, Strolling Ballantyne, and Grand National Today. The governing document is the same for all of them. When the system fails on The Charlotte Mercury, the rule that prevents the next failure also governs The Farmington Mercury. The lesson propagates. A separated governance structure — different rules for each publication — would mean errors compounding in parallel, failure histories that never talk to each other, and the same mistake made twice in different cities.

The document does not care which publication the error came from. Neither does the correction.

Who Signs the Correction

When the system produces an error, the question is not technical. It is editorial: who writes the correction? Who faces the source whose name was wrong, the official whose position was mischaracterized, the reader who made a decision based on a fact that was not a fact?

At Mercury Local, the answer is the same as it has always been at any news organization: the publisher.

That answer does not change because the draft was produced by an agent rather than a reporter. It does not change because the fact-check table showed twelve verified rows before publication. The system's failure is the publisher's failure. The rules that prevent the next failure are the publisher's rules. The document that holds those rules is the publisher's document — written, dated, and waiting at the start of every session.

This is the part of agentic journalism that gets lost in the governance conversation. The question is not whether AI output is accurate enough. It is who is accountable when it is not.

The Document Is the Standard

What a multi-agent newsroom produces is not just articles. It produces a failure history — a running record of what the system got wrong and what the correction cost. The governance document is what that history looks like when you make it operational: every rule traceable to a specific mistake, every mistake traceable to a specific piece of writing that someone had to fix.

I am not running eleven rules. I am running an operation that produced eleven failures worth writing down. The rules are what accountability looks like when you make it legible and permanent.

The prompt is the policy. If the policy has eleven rules, it is because the prompt has been wrong eleven times — and someone decided to put that in writing.