Skip to content

RAG Architecture

RAG Architecture

The rag_architecture metric evaluates retrieval, safety, and evaluation readiness for production RAG systems.

Last verified against engine metric version 5.0.0.

current Breaking Notes

Legacy previous keys are removed.
Risk/gap keys use 0..1 where lower is better.
New readiness composites replace previous readiness/coverage helpers.

What It Measures

Retrieval readiness risks

rag_architecture.retrieval_latency_risk
rag_architecture.index_freshness_risk
rag_architecture.context_budget_gap
rag_architecture.reranking_gap
rag_architecture.query_planning_gap
rag_architecture.retrieval_adaptation_gap
rag_architecture.retrieval_topology_maturity_gap
rag_architecture.retrieval_thresholding_gap
rag_architecture.contextual_retrieval_gap
rag_architecture.cache_layer_gap

Safety and boundary risks

rag_architecture.tenant_isolation_risk
rag_architecture.retrieval_scope_filter_gap
rag_architecture.prompt_injection_guard_gap
rag_architecture.retrieved_content_sanitization_gap
rag_architecture.embedding_poisoning_guard_gap
rag_architecture.retrieval_trust_boundary_gap
rag_architecture.claim_support_gap

Grounding and evaluation risks

rag_architecture.eval_harness_gap
rag_architecture.eval_metric_coverage_gap
rag_architecture.grounding_citation_gap
rag_architecture.citation_precision_gap
rag_architecture.citation_coverage_gap
rag_architecture.answerability_gate_gap

Composite Outputs

Metric	Range	Interpretation
`rag_architecture.retrieval_readiness`	`0..1`	Higher is better retrieval posture.
`rag_architecture.safety_readiness`	`0..1`	Higher is better safety/boundary posture.
`rag_architecture.evaluation_readiness`	`0..1`	Higher is better eval/grounding posture.
`rag_architecture.overall_health`	`0..1`	Higher is better overall readiness.

Configuration

rag_architecture should run with centrality for best evidence weighting context.

metrics:
  - id: centrality
    enabled: true
  - id: rag_architecture
    enabled: true

Read Next