A starting assumption present among many Arabists is that the ʕarabiyyah -- or Poetic Koine -- was a form of the Arabic language, quite close to what we now know as Classical Arabic, and that it existed as a kind of inter-tribal form of communication for high poetic culture.
The argument for this is in the main part based on the corpus of Pre-Islamic poetry. This is marked, they say, by a remarkable linguistic unity, while the composers of these Pre-Islamic poems are said to come from all over the peninsula. The only way to understand this linguistic unity combined with geographic spread, is to assume a single poetic register shared by all of these poets.
Some scholars have gone as far as to say that it is not just the Pre-Islamic poetry was incredibly unified, but that even the Arabic language itself, was one of remarkable linguistic unity. This final, more extreme position is -- in my opinion -- now thoroughly debunked by the uncovering of remarkable linguistic variation in Pre-Islamic Arabic in the epigraphic record. Forms of Arabic like Nabataean Arabic, Safaitic, Hismaic and Late Nabataean inscriptions are remarkably different from each other. This is certainly, in part, due to a diachronic difference between these different varieties, but it should be noted that not a single one of these varieties could be the ancestor of the ʕarabiyyah by the simple fact that none of them retain nunation. Something that looks close, or identical, to the ʕarabiyyah as defined by the Arab grammarians, is strikingly absent in the epigraphic record (so far).
But also the "light" definition -- that the ʕarabiyyah was an Interdialectal Poetic Koine -- is not without its problems.
First, there is the assumption that the attribution of the Pre-Islamic poetry to the Pre-Islamic period is genuine. As the collection of these poems into writing only takes place in the second Islamic century, this is not exactly a given fact. Nevertheless, some things invoked by the Pre-Islamic poetry require knowledge of the Pre-Islamic period, which are not obviously present by the time the texts get collected, and can sometimes be shown to not have been understood, as classical commentators on the poems would misunderstand certain words, which we now understand because we have a historical record. The mention of zabūr in Muʕallaqat Labīd, for example, makes much more sense in its Pre-Islamic meaning: The South Arabian Zabur sticks than as 'scripture' (thanks to Ahmad Al-Jallad for pointing this out to me). As such, it seems clear that at least part of the corpus of Pre-Islamic poetry can with certainty be projected to the Pre-Islamic period.
Second, there is the assumption that the Pre-Islamic poetry was transmitted accurately and without change from before Islam until the second Islamic century when they were collected. This assumption is far more problematic. By the time the poetry gets collected, there is already a very strong ideology that Classical Arabic is the standard of high culture, and eloquent speaking; The now omnipresent diglossia were -- if not already present -- certainly developing. We simply cannot assume that there were no sociolinguistic pressures present in this period to classicize the language of the Pre-Islamic poetry. In fact, I would argue that there are clear indications of classicization of Quranic Arabic, and it seems at least as likely that the Pre-Islamic poetic corpus underwent a similar process. Within this, I believe valid, assumption the remarkably linguistic unity of the pre-Islamic poetry becomes completely circular. The Pre-islamic poetry looks unified because it was intended to be unified.
Third, there seems to be an assumption that there is no counter evidence in the Islamic Period that there was a unified Pre-islamic ʕarabiyyah. But this is not at all in line with the facts. First, there is considerable linguistic variation even between the reading traditions of the Quran, despite being restrained by the Quranic Consonantal Text, nevertheless no one could deny that these varieties are "Classical Araboid". However, most of the variation that we find in these reading traditions is conspicuously absent in the Pre-Islamic Poetry. It seems unlikely that that was never there. Likewise, Sibawayh describes a remarkable amount of linguistic variation, most of which does not at all surface in the Pre-Islamic poetry; It could however not be argued that Sibawayh was describing anything other than the ʕarabiyyah. Any dialect of Arabic that lacked full case inflection (which must have certainly existed in his time) completely escapes attention in his description. It does not seem likely that 1. All the variation Sibawayh describes developed in a post-islamic period and 2. that the ʕarabiyyah that is reflected in the Pre-islamic poetry just so happened to have all the features that end up being typical of what ends up becoming the "Canonical Classical Arabic". Classicization, must have played at least some role.
Nevertheless, the strict meter and rhyme of the poetry, does not allow for a complete "wild west" of different dialects that have all been merged towards a single unified Classical Arabic. But, it does allow for more variation than one would think. In the following section, I wish to explore some of the linguistic variation that could be possible given the strict limitations of the rhyme and meter; This is purely speculative, but it will bring up some hypotheses which could be tested, but because the questions have never been asked, do not seem to ever be tested.
One basic and unassailable assumption that we must defend in the Pre-Islamic Poetry is that case vowels must have existed. The rhyme of the Pre-Islamic poetry often relies on a constantly returning vowel ū̆, ī̆ or ā̆, and more often than not this rhyming vowel is a case vowel. You can see that the poetry clearly uses unusual syntactic construction just to make sure the right case vowel ends up in the rhyming position. Any implementation of variation that puts the reality of case vowels into question, is therefore by definition wrong. A suggestion like "Metri Causa Epenthetic Vowels" is completely disconnected from reality. However, this is not to say that the Case vowel/Nunation system must have been identical to that of Classical Arabic (on which more below).
The Glottal Stop
The Pre-Islamic poetry is metrical, either having light syllables CV (L) or heavy syllables CVC and CVV (H). However, several common developments that happen to the Proto-Arabic *ʔ tend to have little to no effect in the metrical shape. The Pre-Islamic poetry is transmitted with the retention of the glottal stop, but could easily be the result of classicization of poetry that had originally lost it, for example in a word like raʔsun 'head', the loss of glottal stop causes compensatory lengthening yielding rāsun. This development is almost universal in the modern dialects today (some dialects of Yemeni are someone ambiguous). This development has absolutely no effect on the actual metrical weight of the word, however which remains HH. The same is true for the loss of intervocalic ʔ, which is usually replaced by a glide, e.g. wāʔilun > wāyilun (both HLH). The only position where the loss of intervocalic hamza would probably cause a change in syllabic structure, is in between two low vowels, e.g. saʔala > sāla 'to ask' (LLL > HL), but even this development could go unnoticed in certain Arabic poetic metres. The Kāmil meter, for example has several positions where LLL and HL are metrically equivalent.
It would be extremely interesting to see to what extent certain metrical irregularities in poetry could perhaps be understood as the result of such a shift. However, I have yet to find a discussion of metrical irregularities in Pre-Islamic poetry that tackles such issues in these terms. It should also be noted that the loss of pre-consonantal ʔ does not necessarily also entail the loss of intervocalic ʔ. For example the Quranic reading tradition of Warš ʕan Nāfiʕ has the former but not the latter (whereas Abū ʕamr has both - although not in *aʔa sequences).
Another common loss, and one clearly reflected in the orthography of the Quran is the loss of post-consonantal hamza, e.g. yasʔalu is spelled يسل <ysl>, and not as in Classical Arabic- يسال <ysʔl>. In Classical Arabic, some of these Hijazisms have entered the standard, most notably in the imperative of yasʔal which is sal (besides regular isʔal) and the verb 'to see' which is yarā etc. rather than the expected yarʔā < *yarʔayu. Pre-Islamic poetry as we have it today certainly makes use of these Hijazi forms. But it would be extremely surprising if yarā was never treated as metrically HH, rather than LH as the Classical Arabic reading suggests.
It should, moreover, bother us that in the Muʕallaqat Imruʔu l-Qays we find verses such as ta-RAA ba-|-ʕa-RA L-ʔAR-ʔAA-|-mi FII ʕa-|-ra-ṢAA-ti-HAA|| Where both the loss of post-consonantal ʔ in tarā and the retention in al-ʔarʔāmi are necessary to function in the meter within the same verse. Why would Pre-Islamic Arabic already have incorporated this "Hijazism", and so irregularly?
Whatever is going on exactly, it is clear that a certain amount of variation could be allowed without ruining the meter, while still giving rise to a realistic-looking dialect of Arabic.
A place where we can actually see dialectal variation in the Pre-Islamic poetry is in the Pronominal suffixes. The third singular and plural masculine suffixes in Classical Arabic have the following shapes:
3sg.m. -hu after a heavy syllable, -hū after a light syllable; Harmonizing to -hi and -hī after i/ī/ay
3pl.m. -hum (harmonizing to -him after i/ī/ay) which becomes humu before a CC cluster.
One point of variation which we would, of course, never be able to see is whether the poem originally had the vowel harmonization as we find in Classical Arabic today.
In several of the Quranic reading traditions the treatment of both pronouns, however, is different. Here the 3sg.m. is invariably long -hū and the 3pl.m. invariably has a long vowel -ū a tthe end: -humū/-himū.
It the Muʕallaqat imruʔu l-Qays, it seems that the -hū was always long:
L H X | L H H H | L H X | L H L X ||
yu-ḌII-ʔu | sa-NAA-HUU ʔAW | ma-ṢAA-BII-|-ḥu RAA-hi-BIN||
ʔi-ḎAA ǦAA-|-ša FII-HII ḤAM-|-yu-HUU ĠAL-|-yu MIR-ǧa-LII||
But the Muʕallaqat Ṭarafah requires us to read the pronominal suffix as -hu after a heavy syllable as in classical Arabic:
|la-KAṬ-ṭi-|wa-LIL-MUR-XAA | wa-ṮIN-YAA-|-hu BIL-ya-di||
|ḥi-QAA-FAY-|-hi ŠUK-KAA FII L-|-ʕa-SII-bi | bi-MIS-ra-di||
Both of these Muʕallaqahs however seem to simply have -hum for the 3pl.m. as can be seen in this line that both Muʕallaqahs share:
wu-QUU-FAN | bi-HAA ṢAḤ-BII | ʕa-LAY-ya | ma-ṬII-ya-HUM||
This is not a combination that occurs in the Quranic reading traditions, where any tradition that has invariably long 3sg.m. also has humū, but it is of course not a form that could impossibly exist. Not however that the only -hum form attested in Muʕallaqat imruʔu l-Qays is this one, which occurs at the mid-verse point. This may have triggered a pausal variant.
While a tripartite case-system is absolutely necessary to make sense of the rhyme in the Pre-islamic poetry, this does not mean that the case-system was identical to Classical Arabic.
In the Modern Yemeni dialects of the Tihāmah, nouns that originally had nunation -i.e. Indefinite nouns- now have a suffix -ū, e.g. bētū 'a house' but im-bēt 'the house'. This is presumably the result of a shift of word-final *vn > vv. As this would have no metrical effects, such a development could have easily taken place within any of the Pre-Islamic poems, while this is no longer visible. A Pre-Islamic poem could have had a system very similar to the Tihāmī dialect, without us ever being the wiser:
And in fact, the fact that baytū/baytī/baytā is exactly the shape that these words take on in pause in the poetry as it is read today, might be a reason to consider that such a system existed, in the Pre-Islamic Period.
In Classical Arabic orthography there is a difference between final weak verbs with a final radical *y and a final radical *w, e.g. بنى <bny> banā and دعا <dʕʔ> daʕā. This can reasonably be argued to go back to an ancient phonemic distinction between final /ē/ and final /ā/, and this is in fact what some reading traditions of the Quran have (e.g. Warš ʕan Nāfiʕ). While later poetry does rhyme these indiscriminately, whenever it is not part of the rhyme syllable, it is impossible to tell whether this distinction was present. It is nevertheless today read indiscriminately in all environments.
The same is true for words like al-miqrāh ''campsite" which come from *al-miqrayatu. Words of this type are likewise pronounced with an ē vowel in the reading traditions.
It seems unlikely that the ʕarabiyyah was as it is presented in the post-islamic collections of Pre-Islamic poetry. If we accept that the linguistic variation that we find in the reading traditions of the Quran are indeed shades of varieties of a somewhat diverse Poetic Koine, then we would expect such forms to be equally dominant in the poetry. But they are not.
If we accept that the pre-Islamic poetry was accurately transmitted we would have to conclude that the only reading tradition that is actually in the ʕarabiyyah is that of Ḥafṣ ʕan ʕāsim. This seems to me a problematic conclusion with no explanation why that would be the case.
In this post I've speculated what kind of variation would be allowed without it breaking the metre or case vowel system. My impression is that nobody has seriously looked at understanding metrical irregularities or morphological idiosyncracies in the Pre-Islamic poetry to understand if there is any linguistic variation in the corpus; By not looking at this, the assumption that the Poetic Koine was linguistically unified is completely circular.
This is not to say that it is not ultimately true that there was a Poetic Koine, or something close to it. But, relying on the classical narrative as presented by Arab scholars who collected the texts several centuries after their composition (accepting that that part of the narrative to be trusted) just is not good enough. As it stands, I would say the existence of a Poetic Koine is unproven independently from the Arab tradition. A careful reading of the pre-islamic poetry, could shed light into this issue.
Are there any works that have seriously looked at the linguistic motivations for metrical irregularities etc.? I'm not aware of any, but would love to hear it!
To close off, let me share something cool. In Muʕallaqat imruʔu l-Qays there are quite a few irregularities in the metre one of these is found here:
wa-ʔIN-na šiFAAʔII ʕABraTUN MUHRAAqaTUN
The last section SHOULD be LHLH, but a syllable is missing. If we assume that the Ghawa-syndrome operated in this verse; that is: an epenthetic a is added after a Guttural+Consonant cluster, this verse becomes metrically regular:
wa-ʔIN-na šiFAAʔII ʕABraTUN muhaRAAqaTUN
 One could take the absence of the ʕarabiyyah in the epigraphic record as an argument in favour of an oral high poetic register, much in the same way as Epic Greek must have been that at some point. However, it is of course an argument from silence.
 Me and Ahmad Al-Jallad address the highly problematic nature of this "Provocative Solution" in the second appendix of this article.