...and if Mitch's suggestion also isn't enough, I'll go back to Pat's suggestion, with this twist: Removing the cage, and adding balls, as Pat recommended, only I use full-compliment as in, iir, 24 balls. Not "full, then remove one" as I've heard many times over the years, but full. I tested this approach for needed clearance under load, and even with a disc-shim between a pair of balls, the loaded assembly rotated just fine. I used cardboard from a paper match to fashion the semi-round shim. My theory is, the less "degree of freedom" that the balls have, the less that any of re-spaced balls can align with the race's dent features, as they try to do under load.
Lastly, adjustment is critical, and one will ideally make final headset adjustments on the headset with but a naked stem's quill fully torqued in place. It's been quite a while since I've worked on a headset that didn't respond favorably to these, Pat's and Mitch's assembly tips. I certainly wouldn't ride very long on a bike that didn't self-center continuously with hands off of the bars (the acid test of a smooth-running headset).
David Snyder Auburn, CA usa
Mitch Harris wrote:
> If Pat's suggestion isn't enough, you can rotate the headset to one
> side. Mark the position and pop the upper race, lower cup, and crown
> race and re-install in a few degrees (30 or so at least) offside.
> When they get brinelled, headsets seem to do this on the fore-aft axis
> so this usually give you undamaged bearing surface to start with on a
> well-faced properly installed headset.