Lesson 85.1 defined the editorial job: real questions, fitting artifacts. Now the pipeline that carries a question from research to a published, reused asset — without losing evidence, voice, or accountability along the way.

The central rule of this lesson: each handoff is a file or structured record, never chat context. Chat forgets, mixes versions, and cannot be audited. Files can.

The stages and their artifacts

• A research packet is the collected evidence: sources, quotes with locations, product facts with dates, open unknowns. It is the only thing allowed to support claims downstream.
• A brief turns the packet into a plan: the question, the reader, the argument or structure, the format, the gaps to close. The brief is where the editor says "this is worth drafting" or sends research back.
• Source notes are the working annotations: what each source actually says, what it does not say, where sources conflict.
• A draft is the full piece written against the brief, with every material claim traceable to the packet.
• Editorial review is the gate: accuracy, voice, originality, usefulness, publishing readiness — Lesson 85.3 makes this a checklist.
• An asset is the finished companion material: diagrams, code, screenshots, downloads.
• Metadata is the findability layer: title, description, links, tags, dates, related pieces.
• Publishing is release through the gate, recorded and announced.
• Reuse is the piece's second life as derivatives — Lesson 85.4's subject.

Each stage transforms a named input into a named output. Research produces the packet. The brief consumes the packet and produces the plan. The draft consumes brief plus packet. Review consumes the draft and produces either approval or a correction list. No stage works from vibes about what an earlier stage "probably found."

The Classroom workflow as the real example

ByeBuy's lesson pipeline is this pipeline running:

outline approved
  → research packet (sources + prior-lesson context)
  → brief (question, reader, structure)
  → draft (claims traceable to packet)
  → editorial review (standard gate)
  → assets + metadata (diagrams, links, tags)
  → publish (recorded release)
  → reuse + update cycle (derivatives, corrections)

AI assists at several stages — organizing the packet, structuring the draft, preparing metadata — but the brief approval and the review gate are human decisions, exactly the amber marks Lesson 84.4 taught. The pipeline is also the operation map Lesson 84.1 asked for, with the contract boundaries of 84.3 at the AI-assisted steps. Class 85 is Class 84 applied.

Expect the pipeline to pay for itself at the review stage first. Teams that adopt file-based handoffs usually report the same surprise: review gets dramatically faster and calmer, because the reviewer finally sees the brief the draft was written against and the packet its claims rest on. Corrections become specific ("claim three is not in the packet") instead of atmospheric ("this doesn't feel right"). That specificity is what lets a small editorial team hold quality while volume grows — the throughput gain comes from less rework, not faster typing.

Exercise: write your CONTENT-PIPELINE.md

Map your pipeline with an owner and a completion artifact at every stage:

# CONTENT-PIPELINE — [publication name]

| Stage | Owner | Completion artifact | Done means |
|-------|-------|--------------------|----|
| Research | […] | research packet file | [e.g. all claims sourced, unknowns listed] |
| Brief | […] | brief file | [editor approved question + structure] |
| Draft | […] | draft file | [complete, claims traceable] |
| Review | […] | approval record or correction list | [checklist passed] |
| Assets | […] | finished files | […] |
| Metadata | […] | filled record | […] |
| Publish | […] | live URL + release log | […] |
| Reuse | […] | derivative list | […] |

Adapt stage names to your reality, but keep the properties: files, owners, done-definitions. A stage with no artifact is a conversation pretending to be a process.

Finish line: a pipeline document where any piece's current stage — and what it needs to advance — is unambiguous to a stranger.

Verify: take one in-progress piece and walk it through the table. If its location or next step is unclear, the pipeline needs repair, not the piece.

Common failure mode: letting drafts skip the brief ("just write something and we'll fix it in review"). Review then becomes rewriting, queues balloon — the bottleneck Lesson 84.2 warned about, self-inflicted. The second failure mode is the packetless draft: writing from the model's memory with sources hunted afterward to decorate conclusions already reached. Sources found to justify a draft are ornaments; sources gathered before the brief are foundations. The pipeline order is the quality control.

Check your understanding

1. Why must handoffs be files rather than chat context? 2. What does the brief approve that makes drafting and review cheaper? 3. Which pipeline stages are natural amber (AI-prepared, human-approved) marks, and why?

Next, Lesson 85.3: AI's real production strengths — structure, variants, summaries, metadata — inside evidence discipline that keeps the machine from pretending to know.