Recension of a Three-Witness Hexameter Poem

Only shared errors, not shared readings, define a family, so begin by testing each shared error for monophyly. Take ferox for feros first: if ferox is metrically impossible at that point, two scribes could have made the same correction independently, which makes it a polygenetic error and useless as a family marker. If ferox scans, the agreement is meaningful. Suppose it scans; then B and C share an error, and A and B share a different one (trisulca). No three-branch stemma accommodates that pattern, and the simplest hypothesis is (A, (B, C)): A and B descend from a lost common ancestor, and C from B or from B's exemplar. Three consequences follow. First, B and C count as one witness, so their agreement cannot outvote A on numerical grounds. Second, A standing alone remains a full witness and its idiosyncratic readings must be examined individually, not dismissed. Third, because the stemma has only two branches, recensio eliminates nothing and every variant must go to examinatio on grounds of metre, idiom, sense and the poem's register. Say so explicitly in the preface - an editor who claims to have applied a stemma here has not understood what a two-branch stemma permits.

At line 6 the decision is grammatical and the metre is only a check. The subject is plural (monstra), so saeviunt is correct and A's saevit is a mechanical slip, easily produced by the loss of a syllable at line-end or by assimilation to a neighbouring singular. Print saeviunt, record saevit in the apparatus, and note that A's error confirms rather than disturbs the stemma. At line 30 the decision cannot be made from the grammar: both ac and atque are possible Latin. The rule that ac is avoided before a vowel is a generalisation about classical verse, and you do not yet know the poem's register. So the method must be different - establish the register from uncontested lines first: does the poem elsewhere admit post-classical or colloquial forms, does it use dein for deinde, does it tolerate elision of atque? If the uncontested text is uniformly classical, B and C's atque is right and A's ac is a lectio difficilior produced by a scribe copying a late source; if the register is late and mixed, A may preserve the original and B and C have normalised. Write the reasoning into the apparatus commentary, not just the decision.

Begin from the presumption that transposition in a twelfth-century copy is mechanical, and require positive evidence to overturn it. Look for the mechanics: does the couplet begin or end with the same word as the line it now follows, which would indicate eyeskip and reinsertion? Does the sense run continuously in both orders, or does one order leave a pronoun without an antecedent and a forward reference dangling? If only one order is coherent, the coherent order is authorial and the other is the accident. Check as well whether the resulting passage preserves the poem's structural rhythm - a catalogue of labours usually admits variation, so structural arguments will be weak here, whereas a broken syntactical link would be decisive. If both orders read, adopt A and B, record C's order in the apparatus as 'versus alio ordine C', and say in a note that the variation is most economically explained by scribal displacement. Resist the temptation, common in editing verse, to invoke authorial revision: for an anonymous late text that is the least economical hypothesis available, and transposition should be the editor's last resort.