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