On pg. 110 I give a few simple statements about calculating the position of the word stress in Latin and in fn. 21 I quote a Cicero passage (Orat. 58) which is normally taken as explicit testimony for the antepenult limitation—although it should be noted that it is not entirely clear that Cicero is talking specifically about Latin, the context could suggest a reference to Greek. I should also have quoted statements of other native speakers supporting the other parts of the system.
Quintilian (Inst. 12.10.33) notes that there are (with a few exceptions) no oxytones in Latin:
ultima syllaba nec acuta umquam excitatur nec flexa circumducitur.
"The last syllable is never accented either as an acute or a circumflex."
Donatus (Keil 4.371) remarks that in bisyllabic words the penult is stressed no matter whether it is short or long (the description is complicated by the grammarian's theory that long vowels in the penult could be either circumflex or acute as in Greek):
in disyllabis, quae priorem productam habuerint et posteriorem correptam, priorem |syllabam circumflectemus, ut meta, Creta; ubi posterior syllaba producta fuerit, acuemus |priorem, siue illa correpta fuerit siue producta, ut nepos, leges; ubi ambae breues fuerint, |acuemus priorem, ut bonus, malus.
"In bisyllables which have a long first syllable and a short final syllable we give the first syllable a circumflex, as in mêta and Crêta; When the second syllable is long, we give the first syllable an acute, whether it is short or long, as in népōs and l'ēgēs; when both syllables are short we give the first an acute as in bónus and málus."
Donatus also provides explicit testimony about the importance of the weight of the penult (Keil 4.371):
In trisyllabis et tetrasyllabis et deinceps, si paenultima correpta fuerit, acuemus |antepaenultimam, ut Tullius Hostilius; si paenultima positione longa fuerit, ipsa acuetur |et antepaenultima graui accentu pronuntiabitur.
"In three- and four-syllables and so on, if the penult is short we the give the antepenult an acute, as in Túllius and Hostílius; if the penult is long by position, it gets an acute and the antepenult not accented."