Hello from Chile,
The plate recognition engine sometimes fails because of plate pattern (for instance, an "8" instead of "B" where should be only letters).
An API response with the top N results can be very useful to compare the candidates against the pattern and improve accuracy.
In general terms, the API works ok with chilean plates, but it never achieves the high perfomance showed in the white paper related to LPR. I suspect that the accuracy could improve with a custom training using local images . Do you plan to add this feature in the near future?
Buen día,
Felipe