Step 1 — Prepare a corpus

Option A - Sample corpus (fastest): Use docs/gdf-eval-kit/sample.gdf.json from your Guides repository checkout (emergency-ops scenario). Import it to your account:

export GDF_API_KEY='gdf.{account_id}.{key_id}.{secret}'

curl -sS -H "Authorization: Bearer $GDF_API_KEY" \
  -H "Content-Type: application/json" \
  -X POST --data-binary @sample.gdf.json \
  "https://guides.co/api/gdf/v1/guides?ai_readable=1"

Note the returned slug. Publish + public if you want unauthenticated URL fetches.

Option B - Client content: Build a GDF document (see GDF v1 Specification) with sections and pages. Import via the same POST /guides?ai_readable=1 or create in Studio.

Option C - AI-assisted: With MCP or Custom GPT connected, ask the agent to scaffold and fill a guide from client docs, then export via GET /guides/{slug}.

Checklist:

[ ] At least 3-10 pages with distinct facts (one fact per page improves item-level scoring)
[ ] Stable titles and section groupings
[ ] ai_readable enabled if using public URLs