Skip to content
Arxo Arxo

ML Architecture

ML Architecture

Section titled “ML Architecture”

The ml_architecture metric evaluates architecture quality for machine learning systems across training, evaluation, serving, and MLOps operations.

Last verified against engine metric version 1.2.0.

Why It Matters

Section titled “Why It Matters”

Production ML systems often fail due to architecture gaps rather than model quality:

  • Train/serve skew creates silent prediction drift.
  • Weak reproducibility and lineage make incidents hard to debug.
  • Missing eval and validation controls allow regressions to ship.
  • Serving and monitoring gaps increase outage and staleness risk.

ml_architecture surfaces these issues as detector-level scores plus evidence-backed findings.

What It Measures

Section titled “What It Measures”

Core ML Architecture Integrity

Section titled “Core ML Architecture Integrity”
  • ml_architecture.train_serve_skew_score
  • ml_architecture.skew_test_absence_score
  • ml_architecture.pipeline_complexity_score
  • ml_architecture.reproducibility_score
  • ml_architecture.train_inference_boundary_score
  • ml_architecture.data_lineage_integrity_score
  • ml_architecture.experiment_isolation_score
  • ml_architecture.eval_integrity_score
  • ml_architecture.serving_maturity_score

Operational Readiness and Controls

Section titled “Operational Readiness and Controls”
  • ml_architecture.drift_monitoring_score
  • ml_architecture.data_validation_score
  • ml_architecture.ci_integration_score
  • ml_architecture.fairness_audit_score
  • ml_architecture.ab_testing_score
  • ml_architecture.shadow_canary_score
  • ml_architecture.monitoring_alerting_score
  • ml_architecture.model_staleness_score
  • ml_architecture.serving_ops_score

Advanced Governance and Resilience Controls

Section titled “Advanced Governance and Resilience Controls”
  • ml_architecture.model_validation_gates_score
  • ml_architecture.calibration_uncertainty_score
  • ml_architecture.feature_store_consistency_score
  • ml_architecture.progressive_delivery_analysis_score
  • ml_architecture.provenance_attestation_score
  • ml_architecture.responsible_ai_governance_score
  • ml_architecture.attestation_enforcement_score
  • ml_architecture.model_registry_governance_score
  • ml_architecture.lineage_schema_fidelity_score
  • ml_architecture.adversarial_resilience_score
  • ml_architecture.post_market_incident_readiness_score
  • ml_architecture.genai_telemetry_semconv_score

Composite and Diagnostic Outputs

Section titled “Composite and Diagnostic Outputs”
Metric KeyRange / TypeDirection
ml_architecture.overall_score0..1Higher is better
ml_architecture.overall_score_extended0..1Higher is better
ml_architecture.gpu_file_countNumberInformational
ml_architecture.database_file_countNumberInformational
ml_architecture.env_config_file_countNumberInformational
ml_architecture.graph.*Graph entriesInformational

No ML Files Detected

Section titled “No ML Files Detected”

If no training or serving ML files are detected:

  • detector scores are emitted as neutral (0.5)
  • ml_architecture.overall_score and ml_architecture.overall_score_extended are both 0.5
  • diagnostic counts (ml_architecture.gpu_file_count, ml_architecture.database_file_count, ml_architecture.env_config_file_count) are emitted as 0
  • no findings are emitted in this neutral mode

Config Quick Reference

Section titled “Config Quick Reference”
metrics:
  - id: ml_architecture
    enabled: true

Policy Quick Start

Section titled “Policy Quick Start”
metrics:
  - id: ml_architecture


policy:
  invariants:
    - metric: ml_architecture.overall_score
      op: ">="
      value: 0.70
      message: "ML architecture health baseline not met"
    - metric: ml_architecture.overall_score_extended
      op: ">="
      value: 0.70
      message: "Extended ML architecture health baseline not met"
    - metric: ml_architecture.train_serve_skew_score
      op: ">="
      value: 0.75
      message: "Train/serve skew controls are insufficient"
    - metric: ml_architecture.reproducibility_score
      op: ">="
      value: 0.70
      message: "Reproducibility baseline not met"

For production-ready profiles, see Policy and CI Gates.

Runtime and ID Compatibility

Section titled “Runtime and ID Compatibility”
  • Documentation route: /metrics/ml-architecture
  • Stable metric ID: ml_architecture
Section titled “Read Next”