Pushed the virtual assistant training payload through the nightly pipeline yesterday. Console throws a hard 400 on the /api/v2/bots/{botId}/training endpoint. Error payload looks like this:
{
"code": "invalid_training_data",
"status": 400,
"message": "Entity regex mismatch for locale en-AU. ACMA mandatory disclaimer not detected in fallback response."
}
Tenant runs on mypurecloud.com.au. Sydney queue won’t clear because the bot rejects the new intent set. ACMA rules force that exact opt-out phrase in the first 15 seconds. Added the phrase to the training data JSON. Parser keeps stripping it out anyway.
Australian mobiles are formatted as 04XX XXX XXX in the entity config. Validation engine seems to expect strict E.164 +61 format. Switched the regex to ^\+61[24]\d{8}$. Training job still fails. Logs show the parser dropping the plus sign before it hits the validation step. Latency to the US training service sits around 340ms. Doesn’t cause a timeout, but the retry logic is doing jack all.
Architect flow handles the handoff to human agents fine. Problem is strictly in the AI ingestion step. Forced locale: "en-AU" in the request header. Same 400 response. Payload size is 480KB. Well under the limit.
Doesn’t look like a network issue. Parser’s just eating the prefix. Local edge should be catching the format before it hits the central validator. Not sure where the regex is getting mangled.