Bot deflection rates skewing WFM adherence calculations

Trying to understand why our bot deflection metrics are causing discrepancies in the /api/v2/wfm/scheduling/v1/adherence endpoint. When agents handle overflow from the AI bot, the wrap_up time isn’t syncing with the skill group assignment in Architect. This leads to false adherence violations in the Chicago hub during peak hours. Is there a specific interaction_id mapping we need to enforce for hybrid bot-human handoffs?