Initial public release (fresh clean history).
Highlights of the 0.4–0.5 line:
- Personal voice layer (
learnverb): deterministic voice profile from machine-local corpora — no LLM calls, relax-only personalization; honesty rules never personalize. - Scanner–catalog sync test: every scanner phrase list must appear in
references/text-tells.md; drift fails CI. - Load budget: one reference hop per pipeline; scanners are executed, not read.
- Cold-touch vs active-thread correspondence guidance and batch-level campaign scans.
Full details in CHANGELOG.md.