As a researcher, I make heavy use of personas:

1. I get the whole team together and we create the assumption personas (Mary Sues) you describe.

2. In the same workshop, I pull out all the bullet points and get the team to position them in a 2x2 matrix of "importance" and "confidence" - do you have actual hard data to support any of this? If so, how important (relevant) is it to the product itself? (So, no stuff about 'owns a cat'.)

3. I use the "riskiest assumptions" (important, but not confident) as the core of an interview script and have the team give feedback on the planned question list.

4. I set up interviews and invite members of the team (including engineers) in rotation. They stay on mute until near the end, but can ask questions at any time. They can DM me questions during the interview and I rephrase them to strip out bias, and then pose them to the participant. I share the recordings with the whole team.

5. We then play spot-the-difference, comparing the assumptions persona with what they actually told us, and then create revised "qualitative" personas that summarise the real things seen and heard.