Access — inside the guide
The Review pipeline.
Something needs to keep checking pages after they're live, forever, without a bill that scales with how much content you've published. That's why this one runs locally instead of on an API.
Four steps
01
Install Ollama
Free, runs locally, no API key, no rate limit. This matters specifically because this workflow runs continuously — a metered API would get expensive or throttle you fast at nine-site scale.
02
Pull a small model
Something that fits your machine — a 3B-parameter model is enough for scoring and flagging, not writing from scratch. Quality here trades off against being free and unlimited.
03
Score on a schedule
Once a week per site is usually enough. The model reads a live page and scores it against a short rubric — is it still accurate, still relevant, still readable.
04
Escalate, don't auto-fix
A low score writes to the escalations table. A human decides whether it needs a rewrite, an update, or nothing. The model flags; it doesn't get to edit anything live on its own.
The full quality rubric, not a vague scorer
Checks for fake statistics, exaggerated claims, thin pages, and stale freshness language — with a hard rule that invented numbers or false claims always score below 40, no exceptions.
Download review-pipeline.jsonEscalations pile up if nobody looks at them. Next: how you find out without logging in. Approvals without babysitting.