Ability to update testing data in existing model (and recalculate BLEU score)
After creating model we found custom testing data were incorrectly paired so the final BLEU score was irrelevant. I suppose custom testing data are independent on model and could be updated without the need of creating new model. To prevent any abuse updating testing data could be allowed only once a day. I really do not plan to invest $300 to fix this cosmetic issue. But now I can't compare the quality with other models reliably.