.
Its along the lines of this.
Instead of forming the involute curve we can generate it. One way of doing this is to use the involute form that has an infinite base circle diameter i.e. a rack; where the form is a straight line at an angle; much easier to manufacture the cutter.
If we then produced an end mill with this form, it would be just like a truncated āVā tool or a single tooth from a rack of the correct modulus.
Now the first pass of this cutter, on the centre line of the gear, would produce a like for like tooth form but if we then rotated our work piece a small amount and our cutter the correct linear distance away from the centre line; as if the gear and the rack where messing, then the gear tooth would become more accurate. The accuracy depending on how many increments we decided to do. There would also need to be some allowances many for clearances etc.
I know this would be time consuming and the process would have to be repeated to generate the other side of the tooth but potentially a very accurate gear could be produced.
I am not expert enough with Gcode or Mach3 to know how to control the tool path but the geometric relationship between the part rotating and the tool moving is relatively simple.
As side note I remember going on a factory tour of Borg Warner, here in South Wales, back in the late 70s, before CNC was invented :-) it was part of my T6 City & Guilds college course in Production Engineering. I remember them doing this but with mechanical controls and the tooth form was being ground. I remember they had more ways of producing gears than you could shake a stick at.
Brian