The only data that they have, based on transparency reports and dissections of their source code, is the time you created your account and last connected to the servers.
Messages themselves are essentially only relayed, with sealed sender, and anything that would be actually useful to identify who was at a protest and who wasn't encrypted.
Things like, e.g when messages arrive at the server would have to be monitored live on compromised servers, which reasonably unless you assume* it is wiretapped already prior to a protest, isn't realistic.
*: of course, I am saying this because making an assumption and portraying it as truth (e.g assuming something is already wiretapped based on no evidence at all) is not the smartest of moves when it comes to threat modeling...especially if you wanna stay sane whilst having a threat model