KX Results — Harness-as-Knowledge-Retrieval (CKG-benchmark replication)

Token-level F1 (SQuAD-style) and RDS = F1/tokens, per Yarmoluk & McCreary. Tokens = input + output + cache_creation + cache_read per query session.

Macro (all query types)

Condition	n	Macro F1	Tokens/q	RDS	RDS ratio	Cost/q	Turns/q	Errors
monolith	45	0.6106	100,237	6.09e-6	0.59×	$0.5606	1.0	0
cnt	45	0.8080	78,603	1.03e-5	1.00×	$0.1017	2.8	0
bare	45	0.4310	233,583	1.85e-6	0.18×	$0.2432	8.7	0

F1 by query type

Condition	T1 entity	T2 obligation	T3 path	T4 aggregate	T5 cross-link
monolith	0.833	0.607	0.458	0.909	0.135
cnt	0.813	0.672	0.642	0.909	1.000
bare	0.875	0.040	0.544	0.006	0.946

Tokens/query by type

Condition	T1	T2	T3	T4	T5
monolith	100,804	100,838	100,873	98,258	101,002
cnt	78,936	27,673	142,511	92,229	59,287
bare	62,649	492,445	105,995	294,462	124,818