also useful to have a "chinese wall" between research that knows what went into the models vs marketing/eval models as a third party would