Skip to main content

Search And Discovery designed

Search should help readers answer "where do I start?" and "which source backs this?" The first implementation should be internal and generated from portal metadata, not a hosted public search dependency.

The first reader-facing index is Discovery Index.

Decision

Use a generated local discovery index first. Defer Algolia/DocSearch until the portal has an approved external publication track.

OptionFit nowUse later
Generated local indexBest for internal review, no external service, can include owner/status/source metadataKeep as source of truth for filters and review packs
Algolia/DocSearchPremature while the portal is internal and visibility tracks are not filteredCandidate for public/customer/partner docs after publication filtering exists
Static navigation onlyAlready useful, but insufficient once deep dives and API pages growKeep as the primary curated path, not the only discovery model

Discovery Index Shape

Each indexed portal page should expose:

FieldWhy it matters
title, slug, descriptionBasic result display
audience, visibility, statusAudience and publication filtering
owner, review_domainMaintainer and reviewer routing
last_reviewed, review_cadence, freshness_riskStaleness and sync triage
source_docsTraceability back to canonical docs
headingsSection-level discovery inside long pages
diagram count / diagram labelsVisual inventory and review-pack assembly
generated referencesAPI, AsyncAPI, source inventory, and future playground surfaces

Reader Workflows

Implementation Direction

  1. Extend the existing source inventory into a portal-page discovery index.
  2. Add a reader-facing discovery page with filters for audience, status, owner, review domain, source doc, and freshness risk.
  3. Add heading extraction so long pages are searchable at section level.
  4. Add diagram inventory from Mermaid blocks and static image assets.
  5. Use the same index for review packs and publication-track readiness.
  6. Revisit hosted search after public/customer/partner filtering is real.

Guard Direction

The checker should eventually validate that:

  • every indexed page has the required portal metadata,
  • every result links to a valid route,
  • source-doc links exist,
  • generated API artifacts are present,
  • stale high-risk pages are visible in the index,
  • internal-only pages do not appear in external-track indexes.