as I recall you can do the ATD nut up comfortably tight and then you apply quite a bit more force to start it acting as an extractor.
If what you have is machined a bit 'off spec' then it could possibly start undoing itself very slightly too soon, if that makes sense.
It won't threaten to self extract by being tightened - it's when it's being UNdone that the centre bolt comes up against 'something'. That 'something' is the first turn of a raised left hand thread half way up the bolt, which is a fit with a female LH thread in the yoke of the ATD. As you continue to unwind anti-clockwise, the ATD is pushed off. The idea then being that if you pull and turn the centre bolt
clockwise when the unit is in yer 'and, the bolt will come out. Er, sometimes.
I rant about these ATD things too often, I know.
But for a ha'pence worth of tar a
very good idea from the magneto's
electrical performance angle was made, in the flesh, into a sow's ear. Also on the downside the question of additional stress on armature spindles, with possible consequences as in Julian's pic from a few days back, can't be ignored. A one-piece machined drive end would have been much better - with the benefit of hindsight, yes - but they already knew there were good reasons to go 'solid' from their experience with the racing instruments. An ATD imposes greater strain than a plain gear, and further out to boot.
I wonder - in my complete ignorance - whether back in the day it would truly have been much more costly to make one piece items? I dunno, but as RD pretty much said, 'that's no way to make things'.
The 'best' auto advance units I have ever seen on a magneto were 'inboard' at the drive end, clock spring and rollers (like BTH bike ones) in a cast housing, with an outrigger bearing . . . Bullet proof but big and pricey. With the likes of Edward Turner demanding magnetos at about £1 or 25/- a unit, it is said, it's a wonder Lucas were able to provide anything at all frankly.