Results

Pending Experimental Runs

This section will be populated after the evaluation harness completes benchmark runs across all 22 domains.

Macro-Average F1 Across 25 Domains

System	F1	EM	Tokens/q	RDS	CPCA ($)
RAG	---	---	---	---	---
GraphRAG	---	---	---	---	---
CKG	---	---	---	---	---

F1 by Query Type

System	T1	T2	T3	T4	T5
RAG	---	---	---	---	---
GraphRAG	---	---	---	---	---
CKG	---	---	---	---	---

Token Cost and RDS

Table 3: Token counts and RDS ratios --- pending.

Figure 6: Token composition by system (estimated mean per query). RAG and GraphRAG consume 4,000--5,000 tokens per query, dominated by retrieval context. CKG uses only ~300 tokens---a 13x reduction vs RAG. Values will be updated with experimental measurements.

F1 by Hop Depth

Table 4: F1 at k=1,2,3,4,5+ for each system --- pending.

Figure 3: Hop-depth F1 degradation curves --- pending.

Structural Metrics

Table 5: RP, HNR, BC, HR for each system --- pending.

Efficiency Curves

Figure 1: F1 vs token budget curves (3 systems) --- pending.

Figure 2: RDS by domain scatter plot (25 points per system) --- pending.

Structure Premium Correlation

Figure 4: RDS ratio vs DAG richness with correlation coefficient --- pending.