Document Metadata

• title: Submission and Enrichment
• description: End-user submission pipeline and AI-assisted metadata/image generation
• status: evolving
• lastUpdated: "2026-03-13 11:46 ET (America/New_York)"
• owner: Product

Submission and Enrichment ## Entry Point - Route action: app/(public)/submit-new/action. ts - Validation schema: `app/(public)/submit-new/schema.

Submission and Enrichment

Entry Point

• Route action: app/(public)/submit-new/action.ts
• Validation schema: app/(public)/submit-new/schema.ts
• AI prompts: app/(public)/submit-new/prompt.ts

Flow

1. Validate submitted URL/name and optional metadata.
   • required at submit: website + trust signal
   • most metadata remains optional for low-friction contribution
2. Scrape website metadata/content (lib/scraper.ts).
3. Run AI enrichment for:
   • display name
   • slug
   • category (required)
   • tags (preferred) and labels (legacy compatibility; optional)
   • punchline and description
   • image-fit recommendation
4. Process feature image:
   • use uploaded file, or
   • evaluate scraped image candidates, optionally with AI selection, or
   • generate fallback image
5. Persist quality + provenance metadata:
   • products.enrichment_meta with metadata_flags and enrichment_fallback
   • products.enrichment_meta.facet_suggestions with accepted/rejected/source facet audit data
   • product_submissions with trust signal and submitter context
6. Insert product facet links.
7. Revalidate public product/taxonomy caches.

Approval Gate

• Submissions can be incomplete at creation time.
• Admin approval requires publish-ready minimum fields:
  • title
  • website
  • punchline
  • description
  • feature image

Taxonomy Behavior

• User-provided and AI-provided taxonomy values are merged and normalized by facet policy.
• Accepted facet links are created in product_taxonomy for:
  • resourceType, topic, audience, lifeStage, format, location, cost, trust
• Canonical persistence policy now applies by facet:
  • closed-set facets reject unknown terms (resourceType, topic, audience, lifeStage, format, cost, trust)
  • location: controlled-open-set (unknown values allowed + review metadata)
• Audience guardrails:
  • audience is optional and should stay empty unless the source explicitly targets a canonical audience segment.
  • Do not default to intended-parents.
  • Use intended-parents only for conception/fertility-planning resources clearly aimed at prospective parents.
  • For broad general-health or mental-health resources, prefer an empty audience array unless the source makes the audience explicit.
• Resource-type and topic guardrails:
  • Use directory only when the resource primarily lists or matches multiple external providers/resources.
  • For a single clinic, health system, hospital, practice, or treatment center, prefer clinic or provider instead of directory.
  • Do not assign mental-health or community-support unless the source explicitly centers those themes.
• navigationCategory suggestions are retained in audit metadata only and are never persisted as term links from this flow.

Failure Handling

• Scrape failures fall back to user input and safe defaults.
• AI enrichment failures degrade gracefully to non-AI values.
• AI enrichment now attempts a strict-format retry before fallback.
• AI fallback state is persisted via enrichment_meta.enrichment_fallback.
• Image failures fall back to generated feature artwork.

Provenance

• Source file: DOCS/features/submission-and-enrichment.md
• Source URL: https://github.com/maggielerman/smc-directory/blob/main/DOCS/features/submission-and-enrichment.md

Dataset Preview

• Raw CSV row/table content is available in the source artifact.