v0.3.0
  • PUCT crawler live — 617 filings indexed from Interchange, going back 90 days
  • Haiku pre-pass gate reduces LLM cost by ~60% vs full-Sonnet on every doc
  • Brief composer: 25-item cap with relevance scoring per user market role
v0.2.0
  • Extraction pipeline: Sonnet-based structured extraction with taxonomy tags
  • Eval gate: nightly golden-set run against 300 docs, 95% accuracy threshold
  • Citation enforcement: hallucinated citations dropped before brief send
v0.1.0
  • Initial infrastructure: Railway Postgres, Cloudflare R2, pg_queue job system
  • Brevo transactional email integration with unsubscribe and bounce handling
  • Basic auth skeleton (better-auth, magic link)