Here we are looking at the regular and hairline masters and the extralight interpolated instance. As you see, the red glyph in the master weights has its bowl aligned over the right stem of the blue base consonant, using the topright anchor. In the interpolated instance, the vowel sits too far to the right.
Then the tonemark. It’s anchored on the top anchor, but in the interpolated instance it has abandoned its correct position though the anchors are all correct.
Interestingly, decomposing the iVowel-lao.right solves the mark positioning bug. But decomposing all the marks isn’t exactly what we want to do, and it doesn’t quite explain why the compiler gets the interpolations wrong.
Intermediate/alternate layers in composite glyphs in variable fonts are hell, there have been bugs for ages (and the workarounds seem to fundamentally change from version to version).
Is your setup completely “normal”? No special layers anywhere?
For the glyphs here, the only weird thing is the mark uses another mark as a component. From the incorrect positioning, I suspect the anchors of the component are being inherited just in the interpolations although I have ‘propagate anchors’ disabled in font info.
@GeorgSeifert could that be related to the issue with the /tcaron that I send you via mail a few days ago? Maybe not, but there I also got the caron “flying away” from the t, only in the non-master instances of the exported variable font. All anchors are in proper place and no special layers are used.