docs(adr): ADR-002 content system of record

[nwalker85]

Binds Phase 1 doctrine for the content pipeline. Decision only — no code, no behaviour change.

What this decides

• Git is the system of record for content. No CMS database, no vendor-hosted content.
• Plain markdown, not MDX. Deferred until a post genuinely needs an inline component.
• Two persistent repos: drafts (Forgejo, private) → published (GitHub, public).
• Promotion is the ratification act. A Forgejo Action opens a PR here carrying the source commit SHA; the merge is the receipt that content became canonical.
• The CMS points at drafts only. This repo is written by the promotion Action and nothing else.
• Editorial gates on the promotion PR. Leak / name / number / link checks fail; voice conformance and prose review warn.
• @nwalker/token-forge is the visual source of truth.
• n8n automates around publishing, never stores or moves content.

Release notes

Added — docs/adr/ADR-002-content-system-of-record.md.

Why it matters — the site has no content pipeline. /writing exists only on an unmerged branch and its single post is a hand-authored 234-line React component, so publishing a second post means writing a second component. This ADR sets the model that replaces that.

Breaking — none. ADR-001's develop→staging and tag→production model is unchanged; content PRs ride the existing release path.

Migration — none required by this PR. Implementation follows as WP0–WP3.

Operators should know:

1. main is stale and slated for retirement. 53 commits behind develop and not the default branch, so git clone -b main or any tool assuming main gets a months-old tree.
2. The Checkout Custody Standard was waived for this work under explicit break-glass authorization, scoped to this repo and this phase. Receipt and follow-up: ~/docs/30-projects/web-estate-restructure/BREAK-GLASS-2026-08-18-custody-waiver.md.
3. Publishing will require a commit and a deploy, not a web editor button. Deliberate — it is what makes content diffable, greppable and reviewable.

Context

Prior art: walkerwire.com on WordPress at 20k followers. The failure modes were structural — content in a database that could not be diffed, no grep across the corpus, no review path, structure owned by the CMS.

The 2026-08-17 launch produced four near-miss defects caught only by human attention at the last moment. All four are mechanically detectable, which is what the editorial gates in decision 6 exist for.

Not in scope

Assistant demo (own PRD), source recovery for ravenhelm.ai, infrastructure repairs, canonical-URL migration off Substack.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 997b75fd-54ff-4fa7-adfd-601e16402c33

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}