PRESCRIPTION 002

Salience-Proportional Observability

Type: PRESCRIPTIONAuthor: Jason CrittendenDate:
SALIENCE-PROPORTIONAL OBSERVABILITY

Observability systems that suppress persistence signals produce manufactured silence — the appearance of stability without the structural property. This silence corrupts governance input, delays escalation, and creates false baselines.

THE PROBLEM

The problem is not noise reduction. Noise reduction is legitimate. The problem is that noise reduction mechanisms cannot structurally distinguish noise from persistence, and default to treating both as noise.

DESIGN REQUIREMENTS

DR-1: SALIENCE ESCALATION ON PERSISTENCE Unresolved conditions must increase in visual and operational salience over time. Salience is a function of persistence duration, not alert frequency. A condition that persists for one hour is more salient than one that fired ten times and resolved.

Invariant mapping: ZTG-2 (Stasis — graduated containment) applied to the observability layer.

DR-2: SILENCE REQUIRES VERIFIED RECOVERY Silence — the removal of a condition from active salience — is permitted only following verified recovery. Verification requires:

Deduplication, correlation, and suppression may reduce duplicate salience for a single condition. They may not reduce salience to zero. Only verified recovery produces silence.

Invariant mapping: ZTG-0a (Observability) + ZTG-4 (Evidence-Coupled Execution)

DR-3: FAILED RESOLUTION ESCALATES THE CONDITION When automated resolution is attempted and fails, the condition escalates. The resolution attempt does not consume salience. Failed resolution is evidence that the condition is resistant to automated remediation. Resistance increases risk. Risk increases salience.

Invariant mapping: ZTG-2 (Stasis) — when automated processes cannot resolve a condition, the system moves toward human intervention.

DR-4: EXPLICIT COMFORT-FIDELITY TRADEOFF Every suppression, deduplication, correlation, or aggregation operation that reduces salience must declare and record:

Invariant mapping: ZTG-0b (Replayability) + ZTG-4 (Evidence-Coupled Execution)

  • Initial detection: Standard alert at declared severity
  • Persistence beyond automated resolution window: Salience increases one tier
  • Persistence beyond first escalation window: Salience increases again; human notification required
  • Persistence beyond declared SLA: Maximum salience; governance intervention triggered
  • The condition that triggered the alert is no longer present
  • The verification is mechanistic, not inferred
  • The verification is recorded as constitutive evidence of recovery
  • What signal was reduced
  • Why — the suppression rule or correlation logic
  • What recovery verification is pending
  • Whether the suppression is operator-requested or system-automated
INTERACTION WITH GOVERNANCE

Salience-Proportional Observability is not a governance mechanism. It is an observability requirement that governance depends on. When the observability layer manufactures silence, governance operates against a false picture. This Prescription ensures that the observability layer does not corrupt the governance input.