Have your students sometimes refused to work with a weaker-level partner?
I’ve recently read this interesting post by Betty Azar responding to a teacher’s worry that students will acquire wrong models if you put mistakes on the board during a delayed error correction stage. She explains why fossilisation is not going to happen: error correction sessions develop the learners’ abilities to self-monitor, reinforce target language and generally help them become more aware of grammar cognitively. Good arguments that can be shared with students, I think! (But I’m still going to mark those incorrect examples with an asterisk or a different colour, just in case :)).