Skip to content
Pulci Nella — the Clownware markPulci Nella — the Clownware markClownware Docs

ADR-059: Docs Drift Gate Replaces Push-Sync

Status

Section titled “Status”

Accepted — supersedes ADR-008; amends ADR-006 for cross-repo stack claims

Context

Section titled “Context”

ADR-008 established an automated push-sync: on every docs/** change pushed to master, the sync-docs-to-starlight.yml workflow copies docs/ into the Starlight documentation repository (clownware/astro-starter-docs, which deployed to astro.clownware.org — domain since retired) and opens a PR.

A full audit on 2026-07-02 found the pipeline dead in practice and broken by design:

  1. No sync PR has ever been merged. 22+ sync PRs sat open in the docs repository, the oldest from 2025-09-30. The workflow reports success when a PR is created, so the failure was silent.
  2. The PRs are unmergeable by construction. The workflow deletes everything in the docs repository except .git and copies docs/ into the repository root. Merging any sync PR would delete the Starlight application itself (src/, astro.config.mjs, package.json).
  3. The content sets diverged long ago. This repository’s docs/ (~151 files) and the docs site’s src/content/docs/ (91 pages) have different structures and largely different content. The docs site was authored for Starlight; it was never a mirror of docs/. ADR-008’s replace-everything model never matched the target repository’s shape.

The consequence: the public docs site froze at roughly September 2025 while this repository moved through Astro 5→6, Tailwind 4, the Fonts API, the Content Layer API, and the enforcement-gate layer — so the site now actively documents a stack the starter no longer ships. ADR-006’s review cadence did not catch this because it is a calendar intention for in-repo docs, not a gate on the external site.

Decision Drivers

Section titled “Decision Drivers”
  • Enforcement over intention: consistent with ADR-039, drift must fail a build, not wait for a calendar review
  • Respect the real architecture: the docs site is an independently authored Starlight application with its own information architecture, not a rendering of docs/
  • Maintenance burden: the push-sync requires PAT rotation and produced only unmergeable PRs
  • Silent-failure elimination: “workflow green” must mean “docs correct”, not “PR opened”

Considered Options

Section titled “Considered Options”

Option 1: Rewrite the push-sync to target the Starlight content directory

Section titled “Option 1: Rewrite the push-sync to target the Starlight content directory”

Description: Keep docs/ as the single source of truth; rewrite the workflow to transform frontmatter and write into src/content/docs/ without touching the Starlight application.

Pros:

  • Docs live beside code; one PR updates both atomically
  • Preserves ADR-008’s single-source-of-truth goal

Cons:

  • Requires a real transformation layer (frontmatter, paths, sidebar) — significant new tooling
  • Forces a one-time reconciliation of two long-diverged trees (151 vs 91 files) into one shape
  • The site’s IA (tracks, snippets, ai-prompts) has no home in docs/; either the site flattens or docs/ bloats

Option 2: Independent authoring plus a CI drift gate

Section titled “Option 2: Independent authoring plus a CI drift gate”

Description: The docs repository is authored on its own cadence. Delete the push-sync. The docs repository’s CI fetches this repository’s versions.json (the machine-readable stack manifest already enforced in-repo by version:check) and fails the docs build when the site’s stack claims diverge.

Pros:

  • Matches how the site actually evolved; each repository keeps the shape suited to its job
  • Converts stack drift from a silent condition into a build failure — enforcement, not cadence
  • Removes PAT management and the destructive workflow entirely
  • Minimal new tooling: one fetch-and-compare CI step in the docs repository

Cons:

  • Prose drift (beyond declared versions) is not caught mechanically; it needs the docs repository’s link validation and periodic audits
  • Docs updates require deliberate work in a second repository

Option 3: Monorepo

Section titled “Option 3: Monorepo”

Description: Move the Starlight site into this repository.

Pros:

  • Atomic updates, one CI surface

Cons:

  • Rejected by ADR-008 for the same reasons that still hold (hosting split, template scope per ADR-035); would couple template cloners to a docs application they don’t need

Decision

Section titled “Decision”

We will go with Option 2 because it makes the enforcement philosophy (ADR-039) apply to the docs site with the least machinery, and it stops pretending the site is a mirror when it has never been one.

Implementation Details

Section titled “Implementation Details”
  • Delete .github/workflows/sync-docs-to-starlight.yml from this repository
  • Close the open sync-docs-* PRs and delete their branches in the docs repository
  • Docs repository CI gains a drift-gate step: fetch this repository’s raw versions.json, compare against the site’s version manifest, exit non-zero on mismatch
  • The docs repository’s version manifest sources from this repository’s versions.json rather than from the docs repository’s own installed dependencies
  • docs/development/docs-sync-setup.md is retired alongside the workflow

Consequences

Section titled “Consequences”

Positive

Section titled “Positive”
  • Stack drift on the public site becomes a build failure instead of a ten-month silent freeze
  • No PAT rotation, no destructive automation, no unmergeable-PR queue
  • Each repository’s CI meaning is restored: green means correct

Negative

Section titled “Negative”
  • Content (prose, examples, coverage) must be maintained deliberately in the docs repository; the gate only covers declared stack facts
  • docs/ content that should appear on the site needs manual porting

Neutral

Section titled “Neutral”
  • ADR-008’s Considered Options analysis remains a useful record; its chosen mechanism is retired
  • ADR-006’s frontmatter review dates remain in force for in-repo documentation; for cross-repo stack claims the cadence is replaced by this gate. (Update 2026-07-05: ADR-006 was subsequently Withdrawn — an audit found its mechanism was never implemented, so there were no review dates in force to retain.)

Validation

Section titled “Validation”
  • Gate fires: a deliberate version mismatch in the docs repository fails its CI
  • Queue cleared: zero open sync-docs-* PRs in the docs repository
  • No silent freeze: docs site version claims match versions.json at every deploy

References

Section titled “References”

Notes

Section titled “Notes”

The audit that motivated this ADR found ~10 systemic defect classes on the live site (stale version manifest, legacy Content Collections API, wrong repository slug, malformed code fences, deprecated deploy guidance) and near-zero coverage of the enforcement layer. Those are content fixes tracked in the docs repository; this ADR governs only the sync architecture.

Enforcement

Section titled “Enforcement”
  • Testable consequences:
    • TC-1: no sync-docs-* workflow exists under .github/workflows/ — the push-sync stays retired.
  • Checks:
    • TC-1 → check no-push-sync (status: warn)
  • Not machine-checkable: the drift gate itself runs in the docs repository’s CI and is outside this repository’s enforcement surface.
  • Graduation log: (empty at creation; entries added when a check changes status)

Date: 2026-07-02
Participants: Template maintainers
Outcome: Accepted