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.