Stumbled on a weird bug today with our AppFoundry integration that handles dynamic workforce scaling across multiple tenant orgs. The PUT /api/v2/routing/queues/{queueId}/agents endpoint consistently returns 403 Forbidden when attempting to assign agents to AI-driven queues during peak load. We are deploying a custom AppFoundry integration to a BYOC edge cluster in us-west-2. The API gateway returns a 502 bad gateway when calling the internal health check endpoint, but the core issue seems to be related to how the AI Bot intent confidence scores are being evaluated before the handoff to human agents occurs. Specifically, we have set a strict confidence threshold of 0.85 in our Architect flow, yet the system is routing conversations with confidence scores as low as 0.62 to the final queue. This is causing a significant spike in handle time and CSAT degradation for our clients. The logs indicate that the intent classification service is returning a valid response, but the subsequent decision node in the flow appears to be ignoring the confidence attribute entirely. We are using the @genesyscloud/ai-bot-sdk v2.1.4 and have verified that the model version is the latest production release. The environment is Genesys Cloud v2024.07, and we are utilizing multi-org OAuth with PKCE for authentication. The issue persists across all tenant orgs, suggesting it is not a configuration error specific to a single client. We have attempted to reset the flow cache and re-deploy the bot configuration, but the behavior remains unchanged. The error logs show no explicit failures, only a warning about deprecated intent mapping fields, which we have already updated. We need to understand if there is a known issue with the confidence threshold evaluation in the current release or if there is a specific API parameter we are missing to enforce the threshold strictly. Any insights into the internal routing logic for AI bots would be appreciated. We are under pressure to resolve this before our next quarterly business review with key clients. Thank you.