task 4892 doubts

lichess.org/training/4892
seems like 3. RE3 instead of QC1 wouldn't led to loss of quality for white. But at the same time 2. ... QB2 instead of RB8 will finally led to win of black. Mistake or my misconception?

3. Re3 Qe2 is crushing for black :) But I agree the training exercise might not be ideal.

How to deal with this non-ideal exercises? How to report them?
For example there are lots of questions where one side already has huge advantage but exercise still asks for forced win, i think this is not great for educational value. (But in this case problem looks appropiate) chesstempo.com style seems better to me (2 pawns up is enough to win), what do you think about it or is there a topic to discuss this?

lichess.org/training/18591
1. ... Qxh5 leads to win in three moves (the same as Qg1), but the task says it's wrong. While Qg2 is marked as a "good move", despite it leads to win in four moves...