Beginning Again
Starting something over is easier to justify than to do. The reasoning is usually sound: the current version has accumulated problems, the problems are structural, a clean start would avoid them.
The reasoning is sound and the outcome is frequently worse, at least for a while, and the reason is worth understanding before deciding.
The second version inherits the wrong lessons
A rebuild carries forward the problems you noticed and discards the solutions you did not know you had.
The existing version contains a great deal of accumulated response to conditions nobody documented. A check that looks unnecessary was added after something failed. A structure that looks awkward is awkward because a simpler structure did not survive contact with a real case.
When you start over, you keep the parts you understood the purpose of. The parts whose purpose was never articulated get dropped, and they get rediscovered one at a time, usually through the same failures that produced them originally.
The dip is longer than expected
Every rebuild has a period where the new version is worse than the old one. This is unavoidable and generally underestimated.
The estimate is wrong because the comparison is against the current version’s known problems rather than its unknown virtues. You are comparing a list of things that are broken against a thing that works, and the list feels heavier than it is.
Whether to proceed depends almost entirely on how long you can tolerate the dip, and that is a question about circumstances rather than about the merits of the rebuild.
When it is right anyway
None of this is an argument against starting over. Sometimes the structural problems are real and no amount of incremental work will fix them.
The distinguishing question I have found useful is whether the problem is with the accumulated decisions or with the foundation they sit on. Accumulated decisions can be revisited one at a time. A foundation cannot, and if the foundation is wrong then every increment is building further from where you want to be.
The other useful question is whether you understand why the current version is the way it is. If you do not, a rebuild will not be a rebuild. It will be a first attempt with extra confidence, and the confidence is the dangerous part.