Rewrite Twitter to Improve Discourse
I realise that hate is good for engagement, ideally any changes they made wouldn't affect this (else they wouldn't make them). I also realise that the average level of "pleasant" isn't a metric you can directly measure - so we could assume average positive sentiment per tweet.
On the second question let's assume you can measure ad hominem (somewhat possible: https://www.aclweb.org/anthology/P19-2028.pdf) and changes which minimise this are desirable.