Scaling Pipelines Across Teams

A pipeline that works for you may not work for your team. You know the implicit assumptions — which extraction template to use, what "good" looks like in the draft, which gate criteria are strict versus advisory. Your teammates do not. Scaling a pipeline requires making every implicit assumption explicit.

The pipeline package has four components: the prompt sequence (every stage prompt with placeholders for variable inputs), the artifact specifications (what each inter-stage artifact looks like), the gate criteria (every pass/fail condition at every transition), and the usage guide (when to use this pipeline, what inputs it needs, what outputs it produces, and known limitations). If any of these four components is missing, the pipeline works for the author and fails for everyone else.

1. 1. Document the Prompt Sequence Write every stage prompt as a template with clearly marked placeholders: {{CLIENT_NAME}}, {{RESEARCH_QUESTION}}, {{SOURCE_LIST}}. No prompt should require the user to guess what goes where.
2. 2. Specify Every Artifact Provide a sample of each inter-stage artifact. Show what the research output looks like, what the draft looks like, what the review report looks like. Samples are more useful than descriptions.
3. 3. Publish Known Limitations Be explicit about where the pipeline breaks. "This pipeline does not handle documents longer than 50 pages. The review stage misses formatting errors in tables. The merge step occasionally duplicates paragraphs when sources overlap significantly."