# v0.103.0: Low-interference capture and workload control > **Status:** Planned > **Scope:** Large > **User promise:** *"IVM stays within a visible database budget while pursuing an achievable freshness target."* > **Blocked by:** [v0.102.0](v0.102.0.md) > **Source:** [September 2026 assessment](../plans/pg-trickle-assessment-and-pre-1.0-roadmap.md) ## Theme Reduce capture and refresh interference with application work after the WAL receipt invariant is proven. Reuse the v0.96 resource contract for every entry point and controller decision. ## Work - Make WAL receipt durable before acknowledgement. Preserve complete source transactions, replay identity, and a committed high-water mark. - Measure and improve decoding cost across many sources. Evaluate shared typed decoding only if it preserves source isolation and transaction completeness. - Shorten source write-conflicting locks only after a replacement snapshot and capture-boundary proof covers late commits, old snapshots, DDL, and decoder lag. - Apply resource admission and cancellation to graph, manual, scheduled, rebuild, and output-consumer work. - Protect fairness and backlog age so one hot view, rebuild, or stalled consumer cannot silently consume the full budget. - Give each automatic freshness action evidence, bounds, hysteresis, override behavior, and a stable reason. ## Exit criteria - [ ] Fault injection proves that each committed source transaction remains accounted for across receipt, persistence, acknowledgement, retry, and restart. - [ ] Paired foreground-write benchmarks meet the named workload budget and retain all attempts. - [ ] Many-source WAL workloads stay within the declared decoding-amplification bound. - [ ] Late commits, large transactions, decoder lag, and concurrent source DDL preserve capture completeness. - [ ] Pressure causes a documented throttle, admission reduction, freshness breach, or suspension without losing committed changes. ## Scope control Do not promise low latency under overload. Keep the conservative lock or capture path when a lower-interference replacement is not proven.