Note: For the purposes of this question, please do not assume that the Tongues of Ice and Fire Wiki is correct1 unless it references somewhere else.
The High Valyrian [grammatical] number page of the Tongues of Ice and Fire Wiki contains the statement that
Collectives often acquire a special meaning (e.g. muña "mother" → muñar "parents.") Sometimes this results in them being reanalyzed into entirely new words, with their own plural (e.g. azantys pl. azantyssy "soldier" → azantyr pl. azantyri "army").
High Valyrian has few irregularities (it is a conlang after all) and generally, when they exist, they're clearly noticed and pointed out. The issue I have is that, while there is a rule where collectives keep the gender of the word they're derived from, that 'rule' doesn't state whether or not completely new words do the same.
To give two opposing examples:
n. 6col.1lun. life. (Relexicalized collective from ābra)
In the above, ābrar is listed as a lunar2 ('lun.') noun, as it is derived from a lunar noun (ābra). However, if it was reanalysed into a completely new word, it could potentially instead be listed as an aquatic ('aq.') noun, such as:
n. 6col.1aq. year (col. of jēda time.)
If the rules are applied as consistently here as they are in the rest of the language, I'd either expect ābrar to be aquatic or jēdar to be lunar, so is there some rule that dictates if/when a reanalysed collective keeps/looses the gender of the word it's derived from?
1 It often has errors and part of the reason for asking this is to find and fix those errors
2 High Valyrian has 4 genders: lunar, solar, aquatic and terrestrial. Relevant to this question is that lunar nouns generally end in a vowel and aquatic nouns generally end in '-r'