Highlights
What this release added
- Added one persistent, privacy-safe recommendation-quality contract across chat, Today/Journey, and voice, including accepted-as-is, edited, rejected, dismissed, created, started, snoozed, completed, and handoff outcomes with normalized source attribution.
- Added an executable recommendation-quality evidence report and `v0.52.0` readiness gate that refuse product conclusions below the documented live sample floor.
Fixes
What this release corrected
- Separated explicit recommendation rejection from low-intent dismissal and removed duplicate preference-feedback submissions while preserving edited recommendations as useful learning input.
- Aligned production validation with the authenticated AI insight-extraction contract instead of treating its intentional anonymous `401` response as a failed deployment.
- Restored production Public API token issuance by applying the existing owner-binding migration and expanded database verification to fail when required release columns are missing.
Known issues
What remained constrained
- The initial recommendation-quality report has zero normalized production events. This release installs the evidence contract; it does not establish recommendation-quality lift or provide enough data for policy training.