I'm writing a Discord bot that, among many other things, must engage in question-answer dialogues with users. There may be multiple of the same kind of dialogue ("script") running at once asynchronously.

On program startup, a YAML file is read which contains scripts (not in the programming sense), which in turn contain lists of questions, each of which contain the question text. Both scripts and questions contain accessory attributes such as regex formulas, rejection responses, and display names.

Currently, ChatBotManager contains a list of Script objects which in turn contain Question objects, all of which can error-check the relevant portion of the YAML they are passed on initialization. Since the YAML is user-editable, this initialization is non-trivial, and should throw critical-exceptions when passed bad data.

After all this initialization, I have a very nice hierarchical structure with "has a" rather than "is a" relationships, and each level of data can contain methods relevant to it.

The question is this: which of the below ways of recording user responses is most pythonic, expandable, and readable?

• Record user responses in a separate, partially-mirrored datastructure and refer to the Script and contained Question objects as read-only
• Record user responses in Question objects themselves, then use some way to ensure that data does not contaminate parallel, identical chatbots. Both kinds of objects can have methods relevant to only their function, and other functions need little knowledge of the contents.

Note: When finally processing the responses, I need access to accessory attributes of each question.

I'm not asking this because I am stumped and can't progress: the first option is clearly within my knowledge. I want to know best practice. The second option embraces decoupling and clearly delineated responsibilities, but I don't know how to implement it properly. Perhaps my Googling for a solution lacked the right terminology?

I don't want to re-read the YAML file on every object's creation, nor run the validation code each time. Reading and validation need only happen on bot startup. I'd prefer to keep validation methods in the object that will store the validated data.

One idea I had is to deep-copy a Script object on each chatbot creation, then discard it on completion. I've read this is an expensive operation, and while my program isn't low on resources, it seems bad practice.

Should each Question be able to generate a Response object that contains the response and the data about the question that is needed later? These could then be compiled in a ScriptResponse object generated by the Script which also contains information about the script.