Day 97 | Let OC Review First, Then Allow Script Writes

Today is June 11, 2026, the 97th day of the lab.

The focus today is shifting content moderation from "discovering issues after publication" to "intercepting before writing." Duplicate articles have proven that automated publishing, without content moderation, easily pushes posts online that are format-correct but lack sufficient value.

We broke the process down into several steps: first, list the slugs that need fixing; second, schedule new topics; third, write replacement drafts; and finally, submit them to OC for content review. Only after passing the review are we allowed to run the update script in dry-run mode. The dry-run confirms that the original trilingual records exist before proceeding to production writes.

This sequence may seem conservative, but it ensures that every step has evidence. Topic review answers "Has the subject changed?"; the draft checklist answers "Which files were written?"; OC review answers "Can it proceed to update?"; dry-run answers "Will it modify the correct original record?"; and the production report answers "Did the online update actually succeed?"

Today also confirmed one thing: Agents cannot simply be asked to "write articles"; they must also be required to "prove there is no duplication." Otherwise, they will naturally continue writing on the same type of topic along the lines of the recent context, especially in consecutive daily posting tasks. Context helps writing, but it also creates inertia.

On Day 97, the lab added a gate to the content pipeline. This isn't about distrusting the Agent, but ensuring that the Agent's output must pass through a verifiable checkpoint.