Cascading verification escalation: degraded signal → deeper sampling → structural audit → delegation modification

Why This Is a Rule

The three verification layers (Three verification layers: continuous signals (daily), periodic samples (weekly), structural audits (quarterly) — each catches different failures) — continuous signals, periodic samples, structural audits — are independent detection mechanisms, but they must be connected by escalation logic to function as a system. Without escalation, each layer operates in isolation: a degraded signal triggers no deeper investigation, a pattern discovered during sampling doesn't trigger an audit, and audit findings don't modify the delegation. The layers observe but don't act.

The cascading protocol connects each layer to the next through if-then escalation: Signal → Sample: when continuous metrics show degradation (error rate increasing, quality scores dropping), escalate from automated monitoring to deliberate sampling — manually review a selection of outputs to determine whether the signal reflects a real quality problem or measurement noise. Sample → Audit: when sampling reveals a consistent pattern (not a one-off but a recurring theme), escalate to structural audit — examine the delegation's design, the delegate's capability match, and the process architecture. Audit → Modify: when the audit reveals a structural problem, modify the delegation itself — change the scope, adjust the specification, redesign the process, or reassign.

Each escalation is a proportional response: you don't trigger a full audit for a single degraded signal (overreaction), and you don't ignore persistent patterns because audits are expensive (underreaction).

When This Fires

• When implementing the three-layer verification architecture (Three verification layers: continuous signals (daily), periodic samples (weekly), structural audits (quarterly) — each catches different failures)
• When verification data shows concerning trends but you're unsure how to respond
• When you need a systematic escalation protocol rather than ad-hoc investigation
• Complements Three-tier error budgets: green (no action), yellow (investigate), red (halt and redesign) — with pre-committed responses per zone (green/yellow/red error budget tiers) with the specific inter-layer escalation logic

Common Failure Mode

Flat response: every verification concern gets the same level of investigation. A single missed deadline triggers a full structural audit (overkill). Persistent quality degradation gets only a metric review (insufficient). The cascading protocol matches investigation depth to signal severity: mild signal → mild investigation, persistent pattern → deeper investigation, structural finding → structural change.

The Protocol

(1) When continuous signals degrade → escalate to sampling: manually review 3-5 recent outputs against quality standards. Is the degradation real? (2) If sampling confirms a pattern (not a one-off) → escalate to structural audit: examine the delegation specification, the delegate's capability, the process design, and the working conditions. What structural factor is producing the pattern? (3) If the audit identifies a structural problem → modify the delegation: adjust the specification, provide additional training, redesign the process, reallocate the work, or reclaim the task (When reclaiming a non-delegable decision, retain only the judgment core — re-delegate research, analysis, and execution to preserve leverage). (4) After modification → return to normal monitoring and observe whether the cascading signal resolves. (5) If the signal continues despite modification → the root cause wasn't identified. Re-audit with broader scope.