Papers
Planned studyv0.0

Autonomous Rollback for Safe Postgres Tuning

This planned study measures whether autonomous tuning can apply bounded Postgres changes, detect regressions, and return to baseline through verified rollback contracts.

Summary

The first thing serious buyers will ask is simple: if QueryRook is wrong, can it undo itself safely?

Claims

QR-CLAIM-2026-001

Evidence-native control plane

Every production-facing recommendation should be backed by inspectable evidence before it can become an operator action.

QR-CLAIM-2026-003

Trust-minimized autonomous operations

Autonomous Postgres operations are safer when the hosted control plane issues bounded, signed action capsules that a customer-owned Conduit verifies locally.

Methods

  • Exercise propose, dry-run, approve, apply, verdict, and rollback lifecycles.
  • Simulate bad recommendations, lock pressure, duplicate actions, and worker restarts.
  • Measure post-rollback return-to-baseline probability.

Metrics

Rollback success rate
Mean time to rollback
Regression detection latency
Durable job recovery rate
Production-impact envelope

Limitations

  • DDL rollback safety depends on action class and customer-specific lock/WAL/replication posture.
  • Some destructive changes should remain permanently outside autonomous authority.

Reproducibility

Current status: Planned bundle. Reports generated by ResearchOps include commit hashes, run IDs, artifact paths, required disclosures, and a SHA-256 integrity hash.