Editorial pipeline with a human gate that can stop a publish

A review gate only counts if it can stop a publish. We build the workflow as states and permissions inside the CMS your editors already open, so a draft cannot reach published until a named person with reject rights clears it, and every rejection stays on the record with its reason. For one content type in two languages that build runs six to ten weeks. Generation is the easy half. The time goes into deciding who is allowed to say no, and making the system enforce that when the calendar is tight. What gets built: Source material first. Drafting runs against your own material, indexed and retrieved at generation time: product documentation, past articles, support transcripts, the price list. Every factual sentence carries a source field. Empty field, the piece cannot leave draft. That one constraint removes most of what makes editing generated copy unpleasant, because the editor stops hunting for where a number came from. Then the gates. Usually three. A structural check that the brief was followed, claims are sourced, and the angle does not duplicate something already live. An editorial pass by a person who can kill the piece outright. A pre-publish check on the things software does better than people: internal links resolve, schema validates, the canonical is self-referencing, and the translation carries the matching hreflang return tag across en, uk and pl. The first and third we automate. The second we will not, and we will not build you something that looks like it. Where it lives. Sanity, Contentful, Strapi, Payload and WordPress all hold workflow state and role permissions well enough to carry this. Airtable and Notion do not, and if your process currently lives in one of them we say so in week one rather than after invoicing. Who does the work: Two people, the same two from scoping to handover. A systems editor who has run a desk against a deadline and knows what a reviewer will and will not do at six in the evening, and an engineer who writes the integrations. We have twice told a client to stop generating for a full quarter and fix their product documentation first, because a pipeline fed contradictory source material produces contradictions faster than a human writer ever could. Both were told in week two. Both stayed. When the answer is inconvenient: The most common inconvenient finding is that tooling is not your bottleneck. If one editor reviews everything and that editor is also the head of content, no workflow improves throughput. The fix is a second reviewer with real authority, which is a hiring decision, not a build. We put that in writing and we would rather lose the standard tier than deliver a machine that stacks a queue in front of somebody with no time to read it. What you receive: A working pipeline in your environment, review roles mapped to your actual named people rather than to job titles, a runbook your team owns and can edit, and the failure modes written down: what happens when the model API returns errors for an hour, when the reviewer is on leave, when a piece is rejected twice by different people for opposite reasons. Handover is a working session with your editors, run in English, Ukrainian or Polish, not a recorded walkthrough nobody opens. Not included: We do not write your content on the build tiers. We do not fine-tune models. We do not migrate your existing archive into the new workflow, and we do not touch templates or front-end code. Translation memory for Polish and Ukrainian is a separate quote once we have seen the volume. No link building, on any tier, at any price. Not for you if: Your goal is more output per euro. A pipeline with a person in it is slower than one without, and we are not the people to talk you out of that trade. You publish under twenty pieces a quarter, in which case a shared brief and a calendar reminder will serve you better than anything we would charge for. You need the reviewer role to exist on paper for an auditor. Or the approach is already settled internally and what you want is a build partner who does not comment on it. Terms: Milestones are named at kickoff and escrow releases against each one on your acceptance. If we conclude in week two that the build should not proceed, we say it, we bill the discovery milestone only, and the remainder returns to you.