Skip to content

FAIR vs QBER in CRML

This page compares FAIR-style models and QBER-style models, and shows how CRML can represent both.

1. FAIR-style (non-Bayesian Monte Carlo)

Characteristics:

  • Inputs: TEF, VF, LM (primary/secondary)
  • Frequency: deterministic or PERT/traditional distributions
  • Severity: lognormal (or similar)
  • Independence assumptions by default
  • No explicit Bayesian updating

In CRML, this maps to:

  • simple poisson frequency
  • single lognormal severity
  • no dependency section

See: Example: FAIR Baseline.

2. QBER-style (Bayesian hierarchical + MCMC)

Characteristics:

  • Hierarchical Gamma–Poisson for frequency
  • Mixture severity with heavy tails
  • Copula dependencies across components
  • Entropy-based criticality indices
  • Bayesian posterior inference (MCMC)

In CRML, this maps to:

  • hierarchical_gamma_poisson frequency
  • mixture severity
  • dependency.copula defined
  • assets.criticality_index defined

See: Example: QBER Enterprise.

3. Key differences

Aspect FAIR-style QBER-style
Frequency TEF, VF → point or simple MC Gamma–Poisson, hierarchical
Severity Single lognormal Mixture (lognormal + Gamma, etc.)
Dependencies Often ignored Explicit copulas
Updating Typically static Bayesian updates (MCMC)
Tail modeling Limited by LM params Rich heavy-tailed, multi-modal

CRML is agnostic: it provides the language to describe either style (or hybrids) and lets the runtime implement the appropriate math.