Skip to content

Phase 2 AI visibility: design review — sharing mechanism for IA migration #7233

Description

@jstirnaman

Goal

Decide the sharing mechanism for the job-led IA before committing engineering effort. The IA's "engine docs live once with thin overlays" model needs some way to share content across Core / Enterprise and deployment variants — but the two candidate approaches (transclusion-default vs. conventional split) have different cost/benefit profiles and the original plan's LLM-benefit claims need empirical grounding.

This issue is a design decision, not implementation. Implementation issues follow once the decision lands.

Parent: #7230

Background

Two routes deliver the same IA outcome:

Route 1 — Transclusion default. Engine docs live at /enterprise/... (canonical). Core's engine pages use a ref-include shortcode to pull engine content inline. Pages render as if the content were local.

Route 2 — Conventional split. Engine docs live at /enterprise/... (canonical). Core's engine pages are short stubs with canonical_url: /enterprise/... and link to the Enterprise page for the bulk of the content. Deployment pages cover only what genuinely differs.

Decision inputs

Factor Route 1 (transclusion) Route 2 (split)
LLM retrieval Duplicate embeddings at multiple URLs (canonical isn't honored by most retrievers) One embedding per concept
Reader UX Everything inline, no navigation Some reader hops between products
Engineering cost New shortcode + fragment library + reverse-index tooling + CI lint Uses existing source: and canonical_url: frontmatter
Failure mode Fragment sprawl, drift across consumers Core feels like a stub product
Existing pattern Parallel to current content/shared/ mechanism Builds on current pattern

A hybrid is also viable: Route 2 by default, Route 1 reserved for small frequently-reused snippets (e.g., a 60-word "how to pass a token in the Authorization header" fragment).

Tasks

  • Empirical test: index a page at two URLs where one declares the other canonical (the existing Core→Enterprise canonical situation gives a free test target before any cleanup). Query ChatGPT search, Claude search, Perplexity, Gemini search. Record which URL gets cited and whether canonical is honored by each retriever
  • Pick a primary route (Route 1, Route 2, or hybrid)
  • If Route 1 or hybrid: scope the ref-include shortcode, fragment design discipline, reverse-index tooling
  • If Route 2 or hybrid: scope per-page audit and stub-conversion approach
  • Pilot: convert top-N show-in/hide-in pages using the chosen route. Identify candidates with grep -rE '\{\{<\s*(show-in|hide-in)' content/ ranked by occurrence
  • Open follow-up implementation issues based on the decision

Out of scope

  • Migration guides per competitor (separate content workstream)
  • Phase 1 IA skeleton (predecessor)
  • Phase 3 editorial discipline (successor)

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions