How It Works¶
This repo is a multi-phase pipeline that turns a date window into a newsletter draft,
with a paper trail in workspace/ that shows what was collected and curated along the
way.
If you want to run it, start with: - Start here (Feb 2026)
Pipeline Phases¶
- Phase 1A: URL manifest generation
- Phase 1B: Source content retrieval
- Phase 1C: Discovery consolidation
- Phase 1.5: Curator notes (optional, if notes exist)
- Phase 2: Events extraction
- Phase 3: Content curation
- Phase 4: Newsletter assembly
- Phase 4.5: Polishing + validation
Phase Gates¶
Each phase writes explicit artifacts to disk and must pass existence/quality gates before the next phase.
Examples:
- workspace/newsletter_phase1a_url_manifest_*.md
- workspace/newsletter_phase1b_interim_*_*.md
- workspace/newsletter_phase1a_discoveries_*.md
- workspace/newsletter_phase2_events_*.md
- workspace/newsletter_phase3_curated_sections_*.md
- output/YYYY-MM_month_newsletter.md
Self-Learning Loop¶
Corrections are encoded into skills and references, then validated by scoring and rule checks.
Loop:
Finding -> Root cause -> Skill/rule update -> Regenerate -> Validate -> Record learning
Trust Disk, Not Self-Reports¶
The system relies on file-based proof and deterministic checks rather than agent claims.