Write in public
Most developer worklogs die as notes. The writing track is about the machinery that stops that happening — and the harder problem underneath it, which is publishing regularly with AI assistance without sounding like everyone else who does.
The two courses are companions, and they attack the problem from opposite ends. The voice course starts from the reader's side: why AI prose reads flat, which vocabulary and rhythm tells give a generated draft away, why the blacklist approach fails, and how to run a pre-publish style audit that counts what a prompt cannot hold. Its ground truth is a published preprint on stylistic normalization plus this site's own failed experiments — including the voice iterations that lost a blind test against the original draft, and shipped nothing.
The pipeline course starts from the system's side: a devlog-to-article pipeline that actually runs in this site's repo. A triage rubric for deciding which worklog entries deserve to become articles, a gated propose-draft-ship state machine in local SQLite, an explicit human checkpoint where a person accepts or rejects before anything goes live, and scheduling with an SLA-recovery rule so a missed slot gets recovered instead of silently vanishing.
Together they make one argument: consistency comes from systems, and credibility comes from voice, and you need both. A pipeline without a voice standard ships forgettable posts on schedule. A voice standard without a pipeline produces one great post and then six weeks of silence. The two courses cross-reference each other where their subjects meet — the pipeline course's drafting chapter carries a hard prerequisite into the voice course, because prompting for your own style is load-bearing there.
Everything on this track is documented from production, not invented for the curriculum. The states and gates are the real ones. The failures are real too, and they stay in: the missed 09:00 slot that started the SLA-recovery work, the deterministic check that rotted, the drafts that cleared every mechanical gate and still failed the style guide. If you keep a worklog and wish it turned into published writing, this track is the working answer.
A workflow for technical authors whose voice is their credibility — why AI prose reads flat, the vocabulary and rhythm tells, cutting the throat-clearing, injecting first-hand experience, and a pre-publish style audit you can actually run.
Turn an unstructured worklog into shipped posts — a triage rubric, a gated propose → draft → ship state machine in local SQLite, explicit human checkpoints, voice-preserving LLM drafting, and SLA recovery for missed slots. The system runs in this site's repo.