For the research studies, it's important that only participants who have been recruited and consented into the study have access to the chatbot (and the research team). We need the ability to specify a set of participants that should be allowed to interact with the chatbot.
For failures, we should either fail silently or allow an optional static reply. (I could imagine a future where we want to use an LLM to generate a reply, but let's wait until there's a clear use case before doing that.)
For the research studies, it's important that only participants who have been recruited and consented into the study have access to the chatbot (and the research team). We need the ability to specify a set of participants that should be allowed to interact with the chatbot.
For failures, we should either fail silently or allow an optional static reply. (I could imagine a future where we want to use an LLM to generate a reply, but let's wait until there's a clear use case before doing that.)