Highlights
What this release added
- Added quest time tracking and privacy-safe labels as personalization primitives, including Trials create/edit fields, telemetry payloads, sequencing inputs, and personalization dataset exports.
- Deepened personalized recommendations with current quest time/label inputs, opt-in safe memory signals, accept/reject feedback loops, and visible source attribution.
- Added CDN and edge-cache policy for static assets, public release metadata, app metadata, and public OG images, plus release promotion verification and live bug-bash evidence gates for major release windows.
- Added a unified observability dashboard at `/admin/observability` that consolidates release health, runtime telemetry, admin analytics, AI/ML evidence, system context, alerts, and cache signals with source-level failure handling and release-health drilldowns.
- Added chat tool-menu and tool-run conversion telemetry so impressions, selections, invocations, cancellations, completed tool results, and downstream chat/quest follow-through are captured as privacy-safe product events.
- Added a Public API playground so pilot partners can mint tokens, test Calendar/Notion/Slack endpoints, and inspect live rate-limit quota headers without owner-side manual log review.
- Added memory-backed recall evidence measurement with repeated-context prompt telemetry, baseline cohort comparison, user override tracking, and privacy-exclusion counts for the v0.49.0 release gate.
- Added voice-first Journey evidence measurement for active-user engagement, transcript completion, send-to-chat, chat handoff conversion, follow-up quest creation, and return-session retention.
Fixes
What this release corrected
- Recorded hosted GitHub Actions recovery evidence after the account budget/spend-limit block was lifted, and moved the completed release-operations blocker out of the active backlog.
- Excluded PII-flagged memory entries from chat recall while recording aggregate privacy exclusion counts without exposing memory values.
- Consolidated duplicate ML/admin and design-system docs into canonical references under `docs/development/`, and refreshed docs indexes so product, API, deployment, backlog, ML/admin, and design-system guidance each have a clear source of truth.
- Normalized the Drizzle migration-history baseline with a forward-only journal marker, safe migration-table baseline stamping, and stricter audit coverage for accepted historical journal gaps.
- Completed release-research decisions for browser automation and Microsoft 365 connector scope in `docs/development/BROWSER_AUTOMATION_SANDBOX_DECISION.md` and `docs/development/MICROSOFT_365_CONNECTOR_SCOPE.md`, keeping both as future gated implementations instead of unscoped `v0.49.0` code.
- Retired legacy ensemble-training and achievement-prediction ML surfaces as explicit `410` compatibility endpoints, while preserving Labs and legacy voice analysis only behind documented feature-gated rationale.
- Added Stripe tier readiness for Free/Pro/Team, upgraded Stripe SDK to `22.1.1`, fixed Team accounts seeing Pro upgrade prompts, and added `pnpm validate:billing-tiers`.
Known issues
What remained constrained
- _None yet_