Evaluation note

Local routing success

96.7%success rateACL v1.02025-11-24

Dataset / task: Routing benchmark across 500 mixed-intent prompts (analysis, execution, recall)

Protocol: ACL routing evaluated on ability to select correct path (local vs hybrid) with human adjudication.

Sample size: n=500

System version: ACL v1.0

Measured: 2025-11-24

Limitations

  • Prompts are English-only; multilingual routing not measured
  • Hybrid path performance not reported in this note

Sources

TR-2025-08TR-2025-28
Thynaptic | Cognitive AI Research