Diagnosability is the degree to which a system provides clear, accurate, and timely information to identify and localize the cause of a failure from its observed symptoms.

IEEE 1232: Standard for Artificial Intelligence Exchange and Service Tie to All Test Environments (Context: System Diagnostics)

Diagnosability enables efficient fault identification and localization—directly impacting Mean Time to Repair (MTTR) and reducing operational costs. It bridges the gap between raw Observability and effective Recoverability.

Key Performance Indicators: MTTD vs MTTI

To measure diagnosability, organizations typically distinguish between detection and identification:

  • MTTD (Mean Time to Detect): The average time between the occurrence of a fault and its detection (e.g., an alert firing). High detection capability is driven by monitoring coverage and threshold tuning.
  • MTTI (Mean Time to Identify/Isolate): The average time between detection and the successful identification of the root cause or specific faulty component. Diagnosability is the primary driver of MTTI.

Reducing MTTI requires a system that provides actionable failure signatures rather than just generic “error 500” states.

  • Analysability: ISO 25010 defines this broadly, including impact analysis for changes. Diagnosability is more narrowly focused on the fault diagnosis aspect during operations.
  • Debuggability: While debuggability often emphasizes the process and tools (e.g., interactive debuggers, dumps), diagnosability emphasizes the system’s inherent property to signal its internal health and failure reasons autonomously.

Diagnosability Criteria

A highly diagnosable system meets several criteria:

  1. Observability: Sufficient data (logs, traces, metrics) is emitted.
  2. Contextuality: Signals include correlation IDs, tenant context, and transaction history.
  3. Accuracy: Low false-positive/false-negative rates in error reporting.
  4. Localization: The system can pinpoint the failing component or service boundary (Isolation).
  5. Interpretability: Error messages are human-readable and map to specific corrective actions.