On pg. 190 I write "Old Latin also permitted gn- as in gnārus [ŋnārus]. (Incidentally, I should have use a point IPA colon to indicate length in a phonetic transcription). This is stated somewhat unclearly. It's true that OL had initial gn- clusters, whatever their exact realization, but all instances of initial gn- did not end up with the same standard orthography in Classical Latin. There are seven native stems that exhibit initial gn- in Old Latin. These are:
Gnaeus (a praenomen), cf. naevus 'birthmark' < gnaevus according to Inc. de praenom. 5)
gnārus 'knowing' (with the derivatives gnārigō 'I make known', gnāritās 'knowledge' (not actually OL, but attested in Sal. Hist. 3.84), gnāruris 'having knowledge', and ultimately probably also narrō)
gnātus 'born' or 'son'
gnītor 'I lean' (Paul. Fest. p. 96 M)
gnōscere 'to know'
gnōbilis (Andr. Com. 3, Acc. Trag. 283)
There is no doubt that Latin had a sound change ultimately eliminating whatever preceded n-, but not all forms converge on that spelling
in Classical Latin. nītor, nōscere, nōbilis are the standard CL spellings, but in the case of Gnaeus and gnārus the spellings with gn are almost exclusively used. A form Naeu(u)us is said to exist by Varro (gram. 330) and Varro also quotes a form narus in an etymological discussion of narrāre (L. 6.51). In the case of gnātus as discussed on pg. 169 fn. 9 there is a tendency to specialize the old spelling for 'son' and the new spelling for the participle 'born'. In the case of gnāvus the g-less spelling predominates: gnāvus is found in homo gnavus et industrius (Cic. Verr. 2, 3, 53) and virum... gnavum (Vell. 2, 105, 2). It is probable that in the case of gnārus and gnāvus the gn- spelling was supported by the negative compounds ignārus 'ignorant' and ignāvus 'lazy' which both appear to be a lot more common than their antonyms. Gnaeus may have been supported by the official abbreviation Cn.
For an interesting attempt to explain the somewhat messy behavior of originally gn- initial words when compounded see the pdf posted by András Cser. Cser argues that at the mid-stage n was preceded by a floating feature [+back] which was realized on the surface in a number of different ways.