v1.1.0 — 2026-07-04
Empirically calibrated humanization (4 blind-judged rounds, 100+ adversarial agents):
- Seams≠markers architecture:
markers:budgets + context gates in the card; kaue.md §5b identity anti-triggers (biography exempt, named receipts win) - themes.md substance bank (receipted stances/anecdotes; no hyperspecific third-party names)
- profiles/kaue/LESSONS.md — calibrated writing lessons, writer-loaded and binding for ALL routed blends (ROUTING guard + per-secondary guards)
- Pipeline v2: mood + re-outline-in-voice, coordinator palettes, fresh-eyes reviser, long-form ledger, course mode (context-gated closers/openers)
- Validator v2: density (marker budgets/gates), intra-doc audit, warmth telemetry, batch twin-scan, QA-machine/epigram/clipped-sentence checks
- Golden regression outputs in testbed/accepted/ (
check_round.py) - Calibration working data moved out of the repo (gitignored); knowledge absorbed into pack/rules