What role and/or influence do anchors have in the internal calculations of KO?

[brianbrubaker]

I've noticed several times when messing around with different parameters that the anchors in a glyph affect the calculations of the auto-kerning. Is that true? If so, would it be better to eliminate them as a variable altogether when developing ones KO models?

For instance, I came across an HM pair that was being kerned in by -2 points. I knew this was wrong because HH was a model and the /H and /M have the same left side spacing and construction. As is so often the case, Kern On helped me find an inconsistency: the baseline and cap height point of the left most vertical line on the /M were out of alignment from each other by two points.

However, when I fixed my error, the auto-kerning stayed at the -2 place. It wasn't until I reset the anchors of the /M that the auto-kerning updated.

Just wondering why that might be as I attempt to learn the variables that affect how Kern On makes its "decisions".