Skip to content
Arxo Arxo

Scoring and Keys

Scoring and Keys

Section titled “Scoring and Keys”

This page documents how openclaw_architecture computes composite scores and how to configure scoring behavior.

Composite Outputs

Section titled “Composite Outputs”

openclaw_architecture emits these composite keys:

  • openclaw_architecture.config_security_score
  • openclaw_architecture.skill_governance_score
  • openclaw_architecture.observability_score
  • openclaw_architecture.supply_chain_score
  • openclaw_architecture.overall_openclaw_health

All composite values are 0..1, where higher is better.

Scoring Formula

Section titled “Scoring Formula”

For each axis, Arxo computes a weighted average of detector scores in that axis:

axis_score = weighted_mean_applicable(detector_scores, detector_weights)

Applicability rule:

  • Detectors marked not_applicable (for example via disable config) are excluded from axis averaging.
  • If an axis has no applicable detectors, the axis score defaults to 1.0.

Overall formula:

overall_openclaw_health =
  config_security_score * w_config_security +
  skill_governance_score * w_skill_governance +
  observability_score * w_observability +
  supply_chain_score * w_supply_chain

The overall score is clamped to 0..1.

Default Weights

Section titled “Default Weights”

Default axis weights:

  • config_security: 0.35
  • skill_governance: 0.35
  • observability: 0.15
  • supply_chain: 0.15

Default detector weights are defined per detector and normalized within each axis.

Config Surface

Section titled “Config Surface”

metrics[].config supports scoring controls:

  • scoring.metric_weights
  • scoring.axis_weights
  • scoring.disabled_metrics
  • scoring.disabled_axes

Example:

metrics:
  - id: openclaw_architecture
    enabled: true
    config:
      scoring:
        metric_weights:
          gateway_auth_gap_score: 2.0
          tool_poisoning_susceptibility_score: 2.5
          openclaw_architecture.skill_virus_scan_absence_score: 3.0
        axis_weights:
          config_security: 0.40
          skill_governance: 0.35
          observability: 0.10
          supply_chain: 0.15
        disabled_metrics:
          - shadow_mcp_server
          - openclaw_architecture.reasoning_trace_capture_absence_score
        disabled_axes:
          - observability

Normalization Semantics

Section titled “Normalization Semantics”
  • Negative weights are clamped to 0.
  • metric_weights are normalized independently per axis.
  • axis_weights are normalized across all four axes.
  • If all axis weights are 0, defaults are restored.
  • If all metric weights in an axis are 0, that axis falls back to default detector weights (normalized).

Detector Key Contract

Section titled “Detector Key Contract”

For the full detector list (result keys, rule IDs, impact/effort, recommendations), see:

Section titled “Read Next”