Here’s my train of thought so far;
Theory:
Gears 1-4 and 2-3 share forks so:
When going 3-4, both forks are activated. If one were they were offset a little too far to one side, as it appears mine are, then the engagement and disengagement will occurs at slightly different times, getting worse over time as the fork abnormally wears.
As a result there’s a potential for a side load applied to the third gear circlip as things move. Eventually with wear, both gears will try and be engaged at once, generating a massive axial force against the circlip as third output tries to move out of the way.
Once the circlip moves third wobbles and is engaged at an angle, and you rip teeth off one side like I did.
Anyway, that is my theory with what I’ve found so far.