This folder is a proof-of-concept scaffold for a fully automated microlearning content pipeline:

Reddit ingest → topic ranking → human approval → content packet → (video) → publish

The goal is to standardize the handoff contracts between components so Petrarch (scraper) and Quimbot (ranker/generator) can interoperate cleanly.

• schemas/ — canonical JSON Schemas for pipeline artifacts
• examples/ — example JSON/JSONL records
• notes/ — implementation notes / scoring heuristics
• docs/ — comprehensive pipeline design documentation
  • AUTOMATION_BEST_PRACTICES.md — Complete architecture, quality gates, success metrics
  • REDDIT_TO_VEO_PIPELINE.md — Trend discovery → visual storytelling methodology
  • CANDIDATE_TOPICS.md — 10 validated topics with Veo prompts and scripts

1. Topic record (topic_record)

   • One record per candidate topic (typically 1 Reddit post).
   • Produced by the ingest + enrich steps.

2. Shortlist record (shortlist_record)

   • A ranked list of topics for a given day/run.

3. Approval record (approval_record)

   • Captures the human gate decision.

4. Content packet (content_packet)

   • The script + metadata used by video generation and publishing.

• Prefer JSONL for large collections (e.g., ingest output).
• Prefer JSON for single documents (shortlist, content packet).