Why is it thought that T resisted assibilation after another T?

Question

It's well known that past a certain point, Latin "t" developed an assibilated pronunciation when followed by "i" and then a vowel, as in the word grātia. Sources agree that there are some exceptions, most of which make intuitive sense to me; e.g. -tīV- is supposed to have resisted this development, and in -stǐV- and -xtǐV- the -t- is supposed to have remained non-assibilated.

But one thing that puzzles me is that various sources say that -ttiV- was immune to this development. From a theoretical standpoint, I can't see why a preceding -t- would be expected to prevent assibilation. The sequence -ttiV- seems to be pretty rare, so I'm also not sure about whether we actually have much evidence to support this generalization about the natural development of the pronunciation of -ttiV-. (E.g. is there compelling inscriptional or Romance-based evidence?)

Walter Blair's Latin Prounciation (1873) says "For manifest natural reasons it is commonly agreed as improbable that the T could have been hissed when it was immediately preceded by another T, or by a hiss, as S or X. (See Zumpt's Lat. Gr., p. 7" (p. 119).

I confess that I haven't sought out Zumpt's grammar yet, as I thought that the members of this site might be able to tell me more quickly and in more detail whether there are actually solid reasons for supposing that -ttiV- was a special environment where the t > [ts] sound law did not apply. If anyone wants to do more digging and post an answer summarizing the results of their research, I would appreciate that as well.

Answer 1

Since posting the question, I found a small amount of relevant information that I thought I'd put as a partial answer. I'd still appreciate answers from other members of this site!

Evidence from other languages showing that [t] may assibilate while [tt] does not

Assibilation of singleton [t] but not geminate [tt] seems to have occurred also in Finnic. "Towards a Typology of Stop Assibilation," by T. A. Hall and Silke Hamann (2003), says

An argument that the length of the frication phase is perceived in relation to the overall duration of the sequence is illustrated by assibilation in Finnic. Here the /t/ was fricativized to [s] before a high front vowel, but the geminate /tt/ remained unchanged (Posti 1954: 51). This can be explained by the fact that the friction phase is perceived as longer if it follows a singleton than if it follows a geminate plosive.

(p. 125)

This explanation seems relatively satisfactory to me. I still would be interested in learning more details about how we know that -ttiV- did not palatalize in Latin.

Evidence from Romance suggesting that [tt] might have assibilated in vulgar speech

On the other hand, I found that the Italian name of the island Brazza is supposed to come from Latin Brattia. Similarly, mattiana seems to have developed to Asturian mazana and Mirandese maçana, although Wiktionary says that matianium has been suggested as an alternative source.

If the pronunciations of these words developed naturally, and if they are from Latin forms with [tt] rather than coming from forms with non-geminate [t], it suggests that the rule about not assibilating -ttiV- may have developed artificially, rather than as a natural rule of the spoken language.

In some other sources, I've seen it suggested that -ti- in words from Greek might not assibilate, which seems similar and also seems like it might be an artificial rule.

Overall, I have the impression that there is not enough evidence to clearly establish whether [tti] before a vowel naturally developed differently from [ti] before a vowel in Latin. There weren't many Latin words with [tti] before a vowel.