I was surprised by the following portion of "Exceptions to rhotacism", by Kyle Gorman (2012):
Latin has a bimoraic minimal word requirement, implemented by a process of Subminimal Lengthening (Mester 1994:20f.). As word-final s is not moraic in Latin, this produces the quantity alternation in cōs-cotis.
[...] word minimality distinguishes between geminates which are (by hypothesis) derived by Assibilation and those which are underlying. The former are subject to Subminimal Lengthening, but the latter are not: compare cōs-cotis and os-ossis.
The first thing that puzzles me is the data given as evidence: I can't find another source that gives cōs-cotis, with alternating vowel length. Lewis and Short, Gaffiot, and de Vaan instead give cōs-cōtis, with a long vowel throughout the paradigm (and so no evidence of lengthening before word-final s).
The other thing that puzzles me is the general motivation for the idea that word-final /s/ is not moraic in Latin. Mester 1994 is not a source for that claim, as it only discusses subminimal lengthening in the context of monosyllables ending in a vowel. Mester gives the example of the imperative dā vs. other forms like the infinitive dăre (page 22), and notes that there are no monosyllabic words ending in V̆ in Latin. We do also see a long vowel in the form dās, which seems like one piece of evidence in support of the rule Gorman states.
But on the other hand, there are multiple monosyllabic words ending in V̆s in Latin. Even setting aside Gorman's exception for words with underlying geminates such as as, bes, os, there's bĭs, cĭs, quĭs, ĭs, and most problematically, văs-vădis "surety". Or at least, all the dictionaries that I've looked at give the nominative singular form as văs, with a short vowel. I don't know how well the length of the vowel in văs is actually attested, though?
There are vowel length alternations in certain nouns with a monosyllabic nominative singular form ending in s like pēs-pĕdis and mās-măris, which might be viewed as further support for Gorman's rule. However, de Vaan says that pēs is probably an example of Lachmann's Law-type lengthening (which would not apply to văs-vădis because the d there is from PIE *dʰ) and that mās-măris is ablaut. There are similar alternations in words ending in other consonants, too, like sāl, sălis and pār, păris ("Latin sāl, pār, mās, and lār", Douglas G. Kilday), even though Latin had monosyllabic words ending in a short vowel followed by /l/ (vel) or /r/ (per, vir). So I don't think any noun paradigm I've seen so far clearly supports Gorman's proposed rule of subminimal lengthening in monosyllables ending in /s/, and I'm inclined to dismiss Gorman's formulation of this rule.
Ultimately, Latin just doesn't have that many monosyllabic words, so it might not ever be possible to definitely figure out what phonological rules applied to them. But can anyone provide any further data or analysis?