>>we have *d > *h1
> >and *h1k > g, which is more readily explained by /?/.
>
> So... maybe *h1 was a voiced glottal stop.
No, it doesn't have to be. The glottalic theory fits neatly here. If the
"voiced" consonants of PIE were in fact pre-glottalised, then a combination
of ? + voiceless consonant would become very similar to its voiced
equivalent. So h1k > g works well. If ....!