Feedback + Evals

Assistant feedback is a product/privacy boundary, not just a thumbs button. The current architecture keeps raw review context bounded and separates user feedback from committed eval fixtures.

Implemented

Assistant Feedback covers feedback capture, bounded review handoff, and customer/Leverage privacy boundaries.
Feedback-Driven Evals covers the reviewed path from bounded feedback context to committed synthetic or redacted eval artifacts.
The old feedback-to-eval candidate sync path, Source Deployments, generated feedback-eval workflow, and direct customer-cluster Jira credential model are retired.

Current Limits

Central handoff/Jira failure notification and controlled replay still need remaining production hardening.
Production customer-cluster mTLS enforcement and certificate rotation automation remain planned release work.
Web-visible feedback notifications wait for a shared web notification feature.

Read Next

Canonical Sources

docs/features/assistant-feedback/spec.md
docs/features/assistant-feedback/decisions/ADR-001-feedback-storage-review-notification-and-privacy-boundaries.md
docs/features/feedback-driven-evals/spec.md
docs/features/feedback-driven-evals/decisions/ADR-001-reviewed-feedback-to-eval-prs.md
apps/api/src/mastra/evals
apps/cms-customer/src
apps/cms-leverage/src