full-content-pipeline.yaml is MAEL’s flagship workflow. This is what actually runs, step by step.

​
The steps

  1. keyword-research → serp-analysis / topic-cluster — find and cluster the opportunity (see First Agent Run)
  2. competitor-intelligence → content-brief → outline → image-brief — turn research into a concrete plan
  3. writer — generates the draft against the brief and outline
  4. Four parallel reviewers — seo-reviewer, readability-reviewer, medical-fact-checker, eeat-reviewer — each scores independently; the DAG’s fan-in barrier waits for all four
  5. internal-linking → schema → qa — mechanical correctness passes
  6. content-evaluation — the real gate (ADR-011): composes the ScorerBackedEvaluator’s 7-dimension stack with the four reviewers’ outputs into one stored verdict across nine dimensions. Medical compliance and hallucination risk are hard, fail-closed — a missing fact-check on medical content fails the gate regardless of composite score.
  7. PASS → approval — human approval is still mandatory. The gate picks which queue the article lands in; it never approves or publishes. FAIL → review_required — approval and cms-publish are condition-skipped.
  8. cms-publish — only reached after explicit human approval

​
Where cost and evaluation live

Every step’s cost_usd is visible via GET /workflows/runs/{run_id} (see Workflows API). The content-evaluation step’s full nine-dimension verdict is stored on content.article_evaluations and surfaced via Content API.

​
Non-negotiable rules for medical (YMYL) content

For any tenant flagged as medical/YMYL:
  • medical-fact-checker and eeat-reviewer run on every article
  • The approval gate is mandatory on every publish — no auto-publish, ever, for the entire pilot
  • Every published article carries the practitioner’s review byline

Multi-tenant setup