Generate in bulk, never rubber-stamp in bulk

The content state machine covers the full lifecycle:

• Draft → pending review → published: generated content rests in a reviewable state by default; only human-confirmed content enters the publish queue.
• Unlisted / rejected / archived: published content can be unlisted, failing content rejected and old content archived — each step recorded.
• Review and publishing separated: approving content and executing the publish are independent; approval never means instant go-live, because the publishing schedule still governs timing.

• Title-library dedupe: bulk imports automatically catch empty and duplicate titles; duplicates never enter the task queue.
• Topic-opportunity flags: every candidate topic is marked unique / possible duplicate / duplicate, and filtering supports the duplicate status directly.
• Duplicate-article interception: when the same article is submitted twice, the earliest valid record is kept and later duplicates are skipped with a recorded reason.

This serves two goals at once: your own site never competes with itself, and cross-site reuse surfaces an explicit warning.

Every generation task freezes its full configuration at submission (client, categories, titles, model, template, pricing):

• Later template or model changes never affect the explainability of delivered tasks — when something goes wrong you can replay exactly what was used.
• Publish times, results and final links are tracked end to end, forming a complete operation record.
• Client delivery or internal retrospectives no longer depend on memory and spreadsheets.

To be explicit: quality control here means a controlled process, not a promised outcome.

• The platform does not promise search rankings — search performance depends on content quality, site authority, competition and search-engine strategy.
• Mass-producing low-value content to manipulate rankings violates search-engine spam policies; the review step exists precisely to keep human judgment ahead of the go-live button.
• An AI review model can pre-screen before human review, but final approval responsibility stays with people.

• Can review be skipped entirely? The pipeline is designed as review-then-publish; you can tighten or loosen the flow per site in site configuration.
• Does duplicate detection produce false positives? Items flagged "possible duplicate" leave room for human judgment; only clear duplicates are intercepted automatically.
• Can review records be exported? Task records and publish results support export for reconciliation — see the quota notes in Multi-Site Management.