Skip to content

Scoring and Keys

Scoring and Keys

This page documents current scoring behavior for rag_architecture, aligned to:

crates/arxo-engine/src/metrics/ai_observability/rag_architecture/plugin/mod.rs

Scoring Formulas

Normalization:

inv(s) = clamp(1 - s, 0, 1)
all gap/risk metrics are 0..1 (lower is better)

`rag_architecture.retrieval_readiness`

Computed from weighted retrieval gaps in plugin:

retrieval_readiness = inv(weighted_retrieval_gap)

`rag_architecture.safety_readiness`

safety_readiness = inv(weighted_safety_gap)

`rag_architecture.evaluation_readiness`

evaluation_readiness = inv(weighted_evaluation_gap)

`rag_architecture.overall_health`

overall_health =
  (retrieval_readiness * 0.40 +
   safety_readiness * 0.35 +
   evaluation_readiness * 0.25)

Range: 0..1 (higher is better).

Emitted Key Contract (current)

Risk/Gap keys (`0..1`, lower is better)

rag_architecture.retrieval_latency_risk
rag_architecture.index_freshness_risk
rag_architecture.context_budget_gap
rag_architecture.reranking_gap
rag_architecture.eval_harness_gap
rag_architecture.grounding_citation_gap
rag_architecture.chunking_coupling_risk
rag_architecture.tenant_isolation_risk
rag_architecture.retrieval_scope_filter_gap
rag_architecture.embedding_compatibility_risk
rag_architecture.cache_layer_gap
rag_architecture.ingestion_resilience_gap
rag_architecture.query_planning_gap
rag_architecture.retrieval_adaptation_gap
rag_architecture.retrieval_topology_maturity_gap
rag_architecture.eval_metric_coverage_gap
rag_architecture.claim_support_gap
rag_architecture.prompt_injection_guard_gap
rag_architecture.retrieved_content_sanitization_gap
rag_architecture.citation_precision_gap
rag_architecture.citation_coverage_gap
rag_architecture.answerability_gate_gap
rag_architecture.retrieval_thresholding_gap
rag_architecture.contextual_retrieval_gap
rag_architecture.embedding_poisoning_guard_gap
rag_architecture.retrieval_trust_boundary_gap

Composite keys

rag_architecture.retrieval_readiness (0..1, higher is better)
rag_architecture.safety_readiness (0..1, higher is better)
rag_architecture.evaluation_readiness (0..1, higher is better)
rag_architecture.overall_health (0..1, higher is better)

Informational keys

rag_architecture.detected_frameworks (JSON array)
rag_architecture.graph.* (graph entries)

Version Note

This contract is documented against metric version 5.0.0.

Read Next