We had two live skills successfully working with our programmed sample utterances, and seemingly overnight both skills began failing. As we looked at our logs, we see that where once we were getting an exact match on our custom slot values we are now getting multiple matches - even when the spoken "value" for the slot should match exactly one of our slot values. In addition, other slot values that should match a synonym defined for a slot value are coming back with no match.
So the question is more along the lines of what changed? We were working smoothly for a number of weeks with multiple beta testers. Then we successfully got the skills certified and went live. It is only within this week that the skills began failing - first confirmed yesterday - 05/31.