Measure model routing at spawn time and separate batch workloads from human prompting

To audit heterogeneous subagent use, compare the child's first observed model with the parent's last observed model at or before that child's start. Comparing both sessions' current model fields can turn later parent model switches into fictitious routing decisions. Keep missing observations explicit, separate guardian/system review from delegated work, and classify repeated evaluation/annotation runs before interpreting an increase in session count as human adoption.

Applicability and limits: this method describes observed model differences, not necessarily explicit user overrides, lower cost, parameter size, improved quality, or causal productivity. A model-resolved provider subset cannot represent missing collectors or other providers. Prompt audits also need human-versus-generated origin checks and cohort-boundary truncation; inherited context can make a short prompt appear falsely incomplete.