To get a better image of the model of pre-indo-european that I'm currently trying to reconstruct, I've been attempting to compile a PDF with a bunch of reconstructions and explanations in them. Of course, this brought me to some lovely problems that I hadn't foreseen before I started working on the PDF. One of these, and one of the major ones at that, is the prehistory of the *s-stems; But I now believe that I have quite a nice explanation.
First let me bring up a nice and proper word from Latin, which I hate bringing up because I know very little about it. But, I'll quickly switch to Sanskrit after that.
opus 'work', genetive operis, it has all the traits of a *s-stem, which it also is.
In Sanskrit we have this word as well ápas- अप॑स्. 'work'. It's easy to see that this word goes back to PIE *h₃épos. What is important to know though, is that Sanskrit besides the noun, also has an adjective that in Classical Sanskrit has become homophonous with the noun, but in Vedic Sanskrit was still distinct: apás अ॒पस् 'active'. Only the accent is different, but this is essential. It is also a feature which is definitely indo-european as it is also found in Greek. For example γένος 'one's descent', besides εὐγενής 'of good descent'. Most Greek examples are less ideal, since it doesn't have tonal minimal pairs such as Sanskrit, it usually has a prefix like εὐ- or δυσ-.
Continuing with the word *h₃épos, let's look at a bit of both the adjectival and nominal inflections:
Why the *e in the root doesn't disappear when unaccented is a bit of a mystery to me. Especially in the adjectival form. But of course playing around with analogy will yield us this *e eventually anyway.
So, I think that possibly, the Adjectival form isn't a derrivation of the Nominal forms, but the other way around. Why? Well if we go back into pre-indo-european, the Adjectival forms actually look more correct.
Let's have a look:
Nom. M/F *h₃əp-ə́sə=sə
Nom. N. *h₃əp-ə́sə
I have some doubts whether the sequence *-ésə=sə would yield *ḗs As per Szémerenyi's law, but it would easily be explained by other similar forms which do have a long accented *e, like the *r-stems.
In the next stage Syncope took place, but not yet schwa differentiation in unaccented position (those that were left anyway).
Nom. M/F *h₃(ə)p-ə́s-z
Nom. N. *h₃(ə)p-ə́z
This is the stage where the accent was retracted to the previous syllable to create a noun of the same stem. There was no longer a rule for the accent to be on the penultimate syllable, and was happily jumping around in quite a few forms.
Much like how the noun *uĺkʷos may have been formed from adjective *ulkʷós around that time, the same happened with the *s-stems. Thus creating forms like the following (Note how the possible syncopated *ə definitely returns here, because there's nothing else to give the accent):
Nom. N. *h₃ə́p-əz
This late innovation might even explain why we find the heterodynamic genitive *-os, on a proterodynamic word.
This is all still pretty rudimentary but I think it works pretty well. But I've probably overlooked something, and everything that I built up here might subsequently be horribly slaughtered. But hey that's what I blog for, so I don't end up as Edo Nyland reconstructing all languages back to conlangs constructed from Basque.