During some casual investigation of Karjalan kielen sanakirja, I appear to have stumbled on something interesting.
One of the more distinctive innovations among the Karelian dialects is the reflexation of Proto-Finnic *s. In Northern Karelian, and in the northernmost dialects of Southern Karelian (including the Tver, Tihvin and Valdai dialects, spoken by Karelians displaced much further south), this is by default retracted to postalveolar š: see e.g. kešä ‘summer’, oštoa ‘to buy’, šilmä ‘eye’.  The development, though, is blocked by a preceding *i, as in e.g. aisa ‘shaft (in harness)’, pisteä ‘to stick’, muistoa ‘to remember’, viisi ‘5’.
In Livvi aka Olonets Karelian (likewise also in Ludian and Veps), funnily enough *s is inversely shifted to š only after *i. This probably indicates that Old Karelian & Old Veps had no **š, and this sibilant split initially produced an allophonic contrast of palatalized [ś] versus unpalatalized [s]; in the northern part of the dialect area, the latter was eventually retracted, and in the southeastern part, the former.
There are several complications to this textbook picture, though. First, as is common, dialect loaning etc. seems to have generated a rather messy border for the sound changes… The Paatene, Mäntyselkä and Porajärvi dialects appear to be particularly inconsistent cases. (Later palatalizations like *-si > *-śi could maybe explain some cases.)
But more interestingly, there are also some other cases where s is found thruout almost all of the Karelian varieties. And it appears that sometimes these can be archaisms of a sort: the distribution goes regularly back to Proto-Finnic *c, rather than *s. So far I have located the following cases:
- *acja > asie ‘matter, thing, errand’
- *acrajin > asrain ‘trident’
- *kecrädä- > kesrätä ‘to spin thread’
- *käci > käsi ‘hand’
- *ocra > osra ‘barley’
- *suci > šusi ‘wolf’
- *toci > tosi ‘true’
- ? *vacara > vasara ‘hammer’ (South Estonian vassar does not suggest *c, but the word’s origin from Indo-Iranian *vadźra does.)
- *veci > vesi ‘water’
- *ükci > yksi ‘1’
Possibly also *vooci > vuosi ‘year’, which has š (< ? *s) in “central Karelian” (Repola, Rukajärvi, Paatene, Mäntyselkä, Porajärvi), but s (< ? *c) in “Northern Karelian proper” (Uhtua, Vuokkiniemi, Kontokki).
It’s notable that the cases here do not only include cases of *c resulting from the Proto-Finnic assimilation *ti > *ci; they also include all cases known from Karelian of the somewhat anomalous PF cluster *cr — which therefore appears to confirm Petri Kallio’s recent proposal that it should indeed be reconstructed as *cr, not as *str.
How to understand this correspondence? It does not seem to be possible to assume that the development has been *c > *ś. Not only would this be phonetically awkward (especially since the geminate *cc is reflected as non-palatal čč across all of Karelian), it also seems to be the case that the Tunkua dialect has *c >> š, in contrast to *ś > s. On the other hand, assuming the retention of PF *c as an affricate all the way ’til the full disintegration of Karelian would also be problematic, for starters e.g. because the weak grade of PF *cc would most likely have already also become a short affricate *c by this time. Yet a gradation pattern ttš : s has not been attested from anywhere in Karelian (nor elsewhere in Finnic), AFAIK.
So I would suppose that the reflex of *c in Old Karelian must have been a second sibilant phoneme; which might be simplest denoted *s₂. Precedents from elsewhere (e.g. Castilian Spanish, Old French, High German) suggest though that the most likely phonetic value for this would be a laminal sibilant [s̻], contrasting with *s₁ < *s as an apical sibilant [s̺]. This would additionally explain why the shift *s₁ > š exists in the first place: to enhance the contrast between this and *s₂ (just as also occurred in German).
A hypothesis of *[s̻] in Old Karelian also seems to offer some new possibilities for explaining the later development of PF *cr. I have already for quite some time wondered about its development to tr in Eastern Finnish dialects on one hand, in Ingrian on the other; which seems to be an odd contrast to sr in Karelian (and also Livvi/Ludian/Veps). In Ingrian it could be easily attributed to contact with Estonian/Votic, but this is further complicated by *sr > zr appearing after all in the Soikkola dialect. Moreover, Western Finnish has *sr > hr. All this makes the Eastern Finnic tr-area look more like a secondary innovation than an isogloss shared with Southern Finnic. Perhaps a development along the lines of *sr > *θr > tr could be assumed? — And at this point it becomes quite handy that *[s̻] > [θ] is also a fairly common innovation (cf. Castilian Spanish again; or, Old Persian). If we only had to assume the fronting of this phone in particular, the scenario here seems to become a little bit better-grounded.
Not every instance of PF *c yields this Karelian *s₂, though. The usual reflexation of *s₁ appears in e.g. the following cases:
- (intervocalic:) *kaca > kaša ‘corner (of ax)’, *keüci > keyši ‘rope’, *kuuci > kuuši ‘six’, *täüci > täyši ‘full’, *uuci > uuši ‘new’
- (postconsonantal:) *kakci > kakši ‘2’, *küpci > kypši ‘cooked’, *lapci > lapši ‘child’; *künci > kynši ‘nail’; *hirci > hirši ‘log’, *orci > orši ‘perch’, *varci > varši ‘shaft’, *virci > virši ‘hymn’
There are also three cases yielding *ś after *i (going per the Tunkua reflexes): *niici > niisi ‘heddle’, *raici > reisi ‘thigh’, and the abovementioned *viici > viisi, which suggest that *c > *s in this set of words is altogether probably quite old, and they’ve after this simply developed as any inherited PF *s. At least *rc > *rs and *Uc > *Us could be simply regular.
Another question is word-initial *ci-, *si- (e.g. *cilta ‘bridge’, *silmä ‘eye’). A brief scan suggests that these seem to be reflected identically; though the Paatene dialect now consistently seems to have s (likewise word-initially before any other front vowels).
If any additional traces of this *s₁ / *s₂ contrast could be found elsewhere in Eastern Finnic (e.g. in Upper Luga Ingrian, which also has the change *s > š) will have to be left for further study.
 Though I could ask if this isogloss might make a better boundary between Southern and Northern Karelian than medial voicing; while pretty easy to locate, the latter is also a relatively trivial feature that seems likely to have just rubbed off on Southern Karelian as Russian influence. — KKS actually includes the dialect of Jyskyjärvi as a part of Northern Karelian even though it does have medial voicing. I also wonder what’s the motivation behind this choice.