A quantity problem that I puzzled over for a long time: was the a of prandium long or short? In the book on pg. 374 I indicate my uncertainty by putting a macron in parentheses over the vowel. Since prandium is from *prāmo-ediom 'what is eaten first', there is little doubt that the a must originally have been long. But since we also know that a long vowel that ends up before n plus a consonant by syncope appears to have been shortened by a second round of Osthoff's Law (pg. 126), there would be good reason to assume that the first vowel of prandium was short in Classic Latin. Since long and short a are never distinguished in Romance, and the Celtic evidence which is sometimes useful for determine the quantity of a, isn't in this case (OIr. proinn, MW prain), we have no direct or indirect evidence for the quantity of the first vowel.
If we compare other vowels before -nd- that are distinguished in the Romance reflexes we find ambiguous evidence. On the one hand, the French and Spanish reflexes of undecim 'eleven' (Fr. onze, Sp. once) suggest that the vowel was shortened. On the other hand, the reflexes of vendere 'sell' point to the one-time presence of a long vowel:
Ital. vendere (with a close e), OFr. vendre, Sp. vender, Sicil. vinniri
Cf. the reflexes minta 'mint'
Ital. menta (with a close e), OFr. mente Sp. menta, Sicil. minta
This pair shows that the first vowel of vendere had reflexes identical to those of i, and therefore must have been a high mid-vowel reflecting a Classical Latin ē. I suspect the shortening before -nd- was phonological and that vēndere had its length restored on the basis of vēneō, vēnīre 'to be sold.' So all in all, prandium most likely did have a short a.