ScaleSense v0.2.0 adds a workload-first database decision framework for evidence-backed architecture choices before implementation.
What changed
- Diagnose why the current design truly falls short by tracing logical operations into physical read, write, and storage work.
- Compare the no-change baseline and smaller corrections before introducing another technology, while still adopting complexity when evidence requires it.
- Clarify component roles: PostgreSQL as transactional authority where it owns business invariants, Redis as cache or derived state unless a stronger durability contract is proven, and Kafka as a retained log or stream when replay, independent consumers, or ordering justify it.
- Make data loss and disagreement explicit through acknowledgement boundaries, authority and reconciliation rules, stale-data behavior, repair, quarantine, rejection, and rebuild paths.
- Require production-shaped benchmarks and measurements instead of deriving exact resources from user count or QPS.
- Keep migrations reversible through checked backfills, shadow reads, idempotent propagation, bounded cutover, freshness visibility, and tested rollback.
Verification
- 135 deterministic tests passed.
- Repository and release checks passed.
- Official Codex plugin and Agent Skill validators passed.
- The complete 28-case live suite ran with zero executor errors and was manually reviewed; focused reruns validated flagged behavior. A mandate-versus-technical-verdict defect discovered by the suite was fixed and its final targeted eval passed.
Install
codex plugin marketplace add umit/scalesense --ref v0.2.0
codex plugin add scalesense@scalesense