i'am not using auto-delegation and i have written a logic that solves the whole dialog from code using elicit slot directives. Also error handling is processed in die code. For that to work, i'll need to know, if a user did response with a wrong slot-type. For Example in an calculator-skill an amazon-number-slot gets filled up with the word "Cat" (something instead of a number).
In the past, the slot then gets filled with an '?' to show up there was something wrong (so this worked just finde and absolute reliable).
Right now this behavior have changed without me changing code. And because i'am testing this in the developer console i'am thinking, no code is involved at this stage (so maybe there is something change / wrong with Alexa itself?)
Instead of the '?', Alexa isn't be able to assign "Cat" to the slot. Alexa just reacts with an error-sound. Not even the intent gets resolved. Seems like the request get interrupted right away. There are also no JSON-Output / Input in the Test-Area at this point.
First step in dialog: Katze ('cat') in response to a number-slot doesn't trigger anything at all.
Second step in dialog: BUT if the first number is filled correctly (by saying a number like '4' in the first step of the dialog), and than answering the second question for a number with a wrong slot-type like 'cat' instead of a number, the old (and correct) behavior happens. 'Cat' gets "resolved" to '?' because it isn't an number. So, why isn't that working correctly anymore with the first number (first step of dialog)? Seems like alexa isn't be able to assign the response to the right intent in the first step of dialog.
(All number-slot are set-up the same way. No slot-validation is used / because i'am doing that kind of stuff in de code).