Kath beat me to it. I took IE phonology with Lubotsky and Kroonen two summers ago, and the word in question when this came up was Skt.
avis--by Brugmann's, we should see
āvis, but we don't. A reconstruction of
*h₃éwis suffices to explain the Sanskrit reflex as well as the rest of core IE and Hittite (
ḫawiš, with a short
a, right?)
Now my personal pet theory is that laryngeal coloring may not have been as extensive in Indo-Iranian as it was elsewhere.
*h₁e,
*h₂e, and
*h₃e all merge as
a, of course, and likewise for
*eh₁,
*eh₂ and
*eh₃ as
ā.
Ollet 2014 points out that for
*eH, there is good reason to believe that coloring did occur, because roots of the form
*Keh₂C- show an unpalatalized initial consonant in Sanskrit. E.g., PIE
*gʷeh₂- 'go' clearly had an e-grade root aorist (Greek
ἔβην 'I went'), but the Sanskrit cognate is
ágāt 'he went', not
**ájāt. So there must have been coloring of
*e by a following laryngeal. However,
*h₃éwis and its like suggest that, perhaps,
preceding laryngeals didn't cause coloring in Indo-Iranian.
Of course, as was pointed out to me by a helpful Twitter account, Lycian has a word 'χawa-' (is it just me, or does the board provide no graphical distinction between Roman <x> 'eks' and Greek <χ> 'chi'?)--
*h₃éwis and
*h₃ówis should both have given **χewa-. I think
*h₂ówis~h₂éwis is the usual synthesis? But that's an unusual-looking ablaut pattern.