ROADMAP - Pattern Metadata Cache

Status: Phase 2 in progress (Phase 1 completed)
Updated: May 2026

Goal

Build a local, build-time pattern intelligence layer for 400+ patterns that reuses the current Gemini integration, existing Notion props, and safe sync behavior back to Notion.

Current Direction

  • Runtime: Next.js App Router + Node build scripts
  • Inputs: Notion API, public/search-index.json, public/.notion-cache/*
  • Existing version stamping already happens in scripts/build-search-index.mjs
  • Existing Gemini env model already exists in the chatbot server code
  • Metadata generation stays as a standalone local build script aligned with current prebuild workflows

Core Flow 

Notion DB rows + search-index.json
  -> fingerprint delta check
  -> Gemini metadata extraction (JSON only)
  -> safe Notion sync for Description / Version
  -> public/metadata-cache.json

Cache Contract

  • Primary artifact: public/metadata-cache.json
  • Base pattern fields remain in public/search-index.json
  • Metadata cache stores AI, sync, and delta data keyed by pattern id
  • Default sync rule: do not overwrite manual Notion description edits unless explicitly allowed

Phases

Phase 1 — Local Metadata Cache + Safe Notion Sync

Completed:

  • metadata builder script
  • build:metadata commands
  • env docs
  • compact cache artifact
  • safe Notion sync for Description and Version
  • fingerprint delta checks and force/limit controls

Phase 2 — Visual Interaction Extraction + Feedback Capture

Completed:

  • media-aware fallback for Gemini failures
  • media block fetching from Notion children
  • retry generation with visual context
  • private thumbs up/down voting on pattern pages
  • persistence of votes to Notion

Remaining:

  • reduce remaining failure cases from complex nested media pages
  • verify failure count reduction in the next full run

Phase 2.1 — Description Quality

Focus:

  • improve actionable search-ready descriptions
  • combine media parsing with Notion props like flows, components, devices, and languages
  • prioritize usefulness over tight length limits

Validation:

  • dry-run sample builds
  • check previously empty descriptions
  • rebuild search index and confirm propagation

Phase 3 — Multi-Framework Snippet Engine

Next:

  • generate implementation-ready snippets and mapping hints with durable storage
  • extend metadata output without breaking current search/runtime consumers