When sensor noise is actually a compliance signal

Telemetry teams love smoothing algorithms. Compliance teams hate them when they hide the first tremor before a batch recall. We document smoothing choices the way we document accounting policies—who approved them, when they change, and what downstream tests still run on raw feeds. We pair engineers with risk analysts for a single afternoon to replay three historical incidents. The goal is not blame; it is to show how an ignored spike would have appeared if the current filters were applied retroactively. The output is a simple toggle in the monitoring UI: "smoothed" versus "raw guardrail." Supervisors pick the lens they need, and both states are logged for audit.