On 27/09/2015 15:14, Dieter Gunkel wrote:
(3) a detailed study of the distribution of word- vs. compound boundaries at caesurae in stylistically comparable poetry, e.g. that of Mātṛceṭa or Aśvaghoṣa?

Although this is not stylistically comparable poetry, I've studied caesura use in the Mudrārākṣasa, and since nothing closer seems to have come up, I'll summarise my results here. I'm afraid I've only published about this in Hungarian, and web translation services don't handle the language very well, but in case you're interested, the paper is online, https://www.academia.edu/10906817/
All my figures are for syllabo-quantitative verse only, āryā types are excluded.
The play (as edited by Hillebrandt) includes a total of 500 yatis in 96 stanzas composed in 10 different metres. A compound boundary at a yati seems to be perfectly acceptable and occurs in 42 instances or  8% of the total cases, as opposed to 0 instances at a morpheme boundary such as between a prefix and a verbal root or nominal stem, and 0 instances within a morpheme. (There are a number of special cases involving vowels merged in sandhi, which are the main topic of the linked paper and about which I'll put an English paper on academia.edu in a couple of weeks, but this is not related to your question.) The distribution of yatis at compound boundaries is rather uneven across the metres, but this may be just random scattering; I have not tested for statistical significance. They occur in as many as 25% of the yatis of the single mandākrāntā verse in the play and as few as 0% of the yatis in the 3 hariṇī stanzas. What may be more significant is the great difference between the frequency of such yatis in śārdūlavikrīḍita and sragdharā,  both of which are represented by a fair number of samples (39 and 24 respectively, i.e. 156 and 192 caesurae): yati at compound boundary occurs in 6% of cases in śārdūla and 13% in sragdharā.
For comparison, the vasantatilaka, in which Viśākhadatta does not use a caesura (nor does anyone else as far as I know, but one is prescribed e.g. in Apte's dictionary appendix on prosody) also has cpd boundaries at the alleged yati point in 7% of cases, but also has a morpheme boundary in 28 and an indivisible morpheme in 33 percent of the total cases.
One more thing: a compound boundary at a yati is explicitly said to be acceptable in the Yatyupadeśopaniṣad ( Weber, Albrecht. 1863. Ueber die Metrik der Inder. (Indische Studien 8) Berlin: Harrwitz und Gofsmann. pp. 462–466 )
All the best,