Evaluation

Evaluation notes

Every public metric must point to an evaluation note. Notes include dataset/task definition, protocol, sample size where applicable, system version, and limitations.

Mavaia v1.02025-11-24

Hallucination detection

94%

accuracy

FocusOS v0.12025-10-12

Intent classification

94%

accuracy

ACL v1.02025-11-24

Local routing success

96.7%

success rate

ACL v1.02025-11-24

Confidence ↔ accuracy correlation

0.73

pearson r

Mavaia v1.02025-10-30

Emotional memory retention

78%

retention score

ACL v1.02025-11-05

Processing overhead

823ms

p99 latency

FocusOS v0.12025-10-18

Response latency

<2s

p95 end-to-end