A dispatch on a working branch is not yet a dispatch. It is a file with the right shape. The transition to canon happens at a gate, and the gate is the editorial pipeline’s most consequential surface.
The gate is not a button. It is a sequence of small certifications that have to align at the same moment. The schema passes. The Reviewer reads. The DLDS lane is attested as accurate. The visual system inherits cleanly. The publish merge lands on main. CI builds. Deploy ships. The reader, somewhere, refreshes a feed.
What the agent owns and what the agent does not
An agent drafting an entry owns the prose, the structure, the kicker, the dek, the tag set, and the provisional DLDS lane. The agent does not own the publish decision. The brand library is explicit on this: in Era 1, the Mayor signs the gate. The Reviewer role is held by the human until an Auditor pattern has been certified on enough volume to delegate.
This is not a temporary scaffolding. It is the lifecycle as designed. An AI-led dispatch can be drafted in a session, but it cannot be canonized in a session. The publish gate creates the asymmetry that makes the editorial surface trustworthy. Draft cheap, certify slow.
Draft cheap, certify slow.
What the gate refuses
The gate refuses three things, in increasing order of severity. It refuses structural failures: a missing field, a malformed schema, a broken link. It refuses voice failures: a sentence that drifts into LLM-hedge, a kicker that performs instead of names, a dek that summarizes the article instead of opening it. It refuses identity failures: a dispatch that misrepresents its provenance, that treats the surface as a SaaS marketing slot, or that quietly imports a tone from somewhere outside the seven voice anchors.
The first refusal is automated. The second is the Reviewer’s qualitative pass. The third is the Mayor’s, and it is the rare one — it is the refusal a structural Reviewer cannot make on its own, because identity is not a checklist.
Why the gate matters more than the draft
A draft is a hypothesis. A canonized dispatch is a commitment. The brand library will, in a year, be read by an agent who has no access to the draft cycle that produced this entry, and that agent will infer Prime’s voice from what got through the gate. Every dispatch that crosses the gate becomes training data for the next operator, human or otherwise.
This is the structural reason the Reviewer role is held tight. The gate is not protecting the reader from a bad article. It is protecting the corpus from a slow drift the corpus cannot self-correct.
A dispatch is not finished when the draft is written. It is finished when the Mayor signs the gate and the deploy lands. Everything before that is rehearsal.