For my PhD work, I've been working on preparing an experimental branch of StrIoT for merging down to the main branch. This has been a long-lived branch (a year!) within which I've been exploring some ideas. Some of the code I want to keep, and some I don't.
The history of the experimental branch is consequently messy. Looking it over and considering what a reviewer needs to see, there's a lot of things that are irrelevant and potentially distracting. And so, I've been going through an iterative process of steadily whittling down the history to the stuff that matters: some strings of commits are dropped, others squashed together, and others re-ordered. The resulting branch is a historic fiction.
This is common practice. Joey Hess ruminated about it 5 years ago in "our beautiful fake histories", pointing out that the real history is also useful, and perhaps worth preserving.
After a recent conversation with my supervisor I realised the situation was analagous to writing a research paper (or a thesis): the process of getting to the conclusion which the thesis documents is messy, with false starts, wrong directions, and plenty of roads-not-travelled. The eventual write-up focusses on the path that lead to the conclusion, and a lot of the side-quest stuff disappears.
The "true history" then, is captured elsewhere: in lab books, diaries and the like, and these have their own value. So do my messy exploratory branches, before they've been cleaned up for merging.