I didn't mean to imply Google was covering anything up, but Gemini in this specific conversation clearly was.

imho the best you can say is that the "thinking" trace says it was. thinking tokens aren't infallible indications of what the model's doing