Most AI failures aren’t mysterious; they’re unobserved. This ships a tiny, open receipt per run—κ for stress, Δhol for drift—so you can diff, audit, and keep the path, not just the answer. It’s just JSON; drop it in CI. Looking for feedback on setup friction and where the signals break.

Links to run it COLE (Coherence Layer Engine): github.com/terryncew/COLE-Coherence-Layer-Engine- OpenLine Core: github.com/terryncew/openline-core

Small receipts. Big accountability.