(shamelessly kidnapped from the Zeeb, because the Zeeb is shit at doing tables. Also, mèþru et sim.)
Bosporan (spoken in Crimea and the Krasnodar area, not the area around Constantinople) is attested over a time period spanning about 1200 years in numerous texts and inscriptions; it is about as well-attested as Tocharian. Two main stages are attested: Old Bosporan, written with the Greek alphabet, from about 400 BC to around 150 AD. A relative lull of about two centuries follows; Late Bosporan is attested from around 350 AD and shows clear signs of Gothic influence, including in the alphabet. The corpus fizzles out around 800 AD, and the last speakers probably lived around the turn of the millennium.
Bosporan comprised a branch of its own in IE, and seems to be somewhere in between Balto-Slavic and the Greek/Armenian/Phrygian group (unsurprising given its location). In line with the former, it maintains no trace of interconsonantal laryngeals, merges common IE *ō with *ū, and possesses instrumental/dative/ablative endings in *m. In common with the latter, Bosporan possesses an augment, a thematic 3sg in *-ei, and (in common with Armenian and maybe Thracian) a stop shift *T *D *Dʰ to Tʰ T D. Grassmann's Law is also operable, but it applies to contemporary aspirates (e.g. those from PIE *T.) Bosporan distinguishes *a from *o and is a centum language.
A brief overview of the historical phonology of Old Bosporan follows. Late Bosporan gets a little weird.
Initial *e-epenthesis. This accounts for the lengthened reflexes of word-initial laryngeals before a vowel; so where Rix's Law turns *HC- into *HeC- into Greek, the Bosporan reflex comes via *eHC-.
À la Western Romance, initial *sC- also gives *esC-—so this was probably just a general rule when a fricative was found word-initially before a consonant.
Common IE changes. The usual, but (as usual!) with some dialectal flavor.
*h₁e *h₂e *h₃e > ĕ ă ŏ *eh₁ *eh₂ *eh₃ > ē ā ō
Initial *H treated as above, with epenthesis. Interconsonantal and word-final *H seem to have disappeared without a trace. Schwa secundum, however, merges with *e as in Greek.
*iH *uH give *ī *ū, but (as in Tocharian) *ih₂ *uh₂ > *ya *wa. Cf. ⲍⲁⲙⲟⲥ 'smoke', which must have had an intermediate *dʰwamos.
Short *o in open syllables gives ō, apparently by Brugmann's. It's entirely unsurprising that Bosporan would share some dialectal features with Tocharian and Indo-Iranian.
Szereményi's Law: Word-final *VRs gives V:R
Siebert's Law: *CCw, *CCy > CCuw, CCiy.
Thorn clusters metathesize as in Greek and II.
*ḱ ǵ ǵʰ > k g gʰ.
Glide fortition. Word-initially and after *s, *w seems to have given intermediate *b—these may not have happened simultaneously for obvious reasons. *y gives /z/ word-initially only.
*w > *b / #_, s_.
Palatalization chain shift.. Dental stops become /s/ or /z/ (this has to have been before the stop shift because *d becomes /z/) before *i, *y, *u or *w; plain velars give dentals before front vowels *i or *e.
*t *d *dʰ > s z z / _i, _y, _u, _w *k *g *gʰ > t d dʰ / _i, _y, _e
*T *D *Dʰ > Tʰ T D; this includes *b from older *w (which is reflected as /p/: ⲡⲓⲉⲭⲟⲥ 'wolf'.)
Treatment of syllabic consonants
Syllabic consonants gain epenthetic e. If they're in a word-initial syllable as *CṚ and CRe is a permissible syllable, they give that. Otherwise, they give eR. There seems not to have been any difference between regular syllabic consonants and lengthened syllabic consonants before a laryngeal.
A few vowel changes.
First, post-consonantal *w and *y metathesize with a preceding consonant, unless it's word-initial; in word-initial syllables they just drop. (After consonant clusters, Siever's Law had already applied.)
*Cw *Cy > *w(C)C *y(C)C / V_
Following this, *l becomes *y after a consonant, and *w after a vowel.
*u(:) and *o merge, then split again, based on length.
*o(:) > *u(:) *ŭ *ū > o u
Short *i merges with *e, à la Cree/Ojibwe, as ĭ. Long *ī then joins to give plain /i/.
*oy *ey merge with *ē to give e. (As you can see, the length distinction is slowly but surely collapsing into oblivion).
*ew *ow *aw give o u u.
Short and long *a merge.
A nasal before a stop voices the stop and then disappears.
*NTʰ *NT *ND > D
Remaining *ns gives *z. *s becomes z also before nasals and remaining voiced stops.
Loss of *w
Remaining instances of *w between vowels drop. Since vowel length is no longer phonemic, this can creates hiatuses: ⲟⲩⲓⲥ [u.is] 'sheep' (from *h₂ówis.)
*y generally stays between vowels, but original *-eye- ended up as hiatused -ⲓⲓ-, e.g. ⲟⲩⲓⲓⲥ 'sheep (pl.)' [u.i.is].
The first of two aspirated stops deaspirates. Since Bosporan aspirates come from plain PIE voiced stops, this applies in entirely separate words than in Greek and Sanskrit. As in Sanskrit, this also occurs in clusters: ⲡⲑⲉⲣ 'father'.
Tʰ...Tʰ... > T...Tʰ...
Some other, minor developments
*sr gives dr; *sn *sm metathesized, merged as *ns, and became /z/. (E.g. ⲓⲍⲟⲥⲁ 'daughter-in-law', from *snusós.)
Word-final *m becomes /n/ as in Greek, Gaulish and Phrygian.
Accent is neutralized, though it's not clear what the details are. Since Greek without accents looks fugly, the standard transcription uses Greek miniscule, written here with a Coptic typeface.