About tags, consider some possible cases and the consequences for various systems.
'x ala x', at least for main verbs, has a nice transformational history but nothing that makes sense to a linear parser (nor does introducing the modal Y/NQ help much) and it doesn't seem to impinge on other structures or functions. So it seems to be a paradigm case.
'anu seme' is so ambiguous from a technical standpoint and the ambiguities are so irrelevant generally that there seems to be little point in sorting it out. At worst, you get (very) occasional odd (maybe even funny) answers to 'sina wile tawa sitelen tawa anu seme': beyond 'wile' and 'ala'. you might get 'tomo' or 'moku e kala' or (in your dreams) 'unpa' or 'li weka e sama'. If the parser is primed for all of these, then it has to produce a lot of strings, but better to just take the 'anu seme' as a floating chunk of some sort, and explain the odd ones if they arise. Not a paradigm case, but a clear case of a practical solution. Occasionally someone will challenge the Y/N nature of the question, but, if they give an answer, we can come up with an alternate analysis.
Number strings, at least those with the number words in non-rising order (I don't know what we do with 'wan tu luka', which still seems to add up to 8) seem another paradigm case. But there are possible conflicts as in "The general needs two centuries" (a Roman general, of course) 'jan lawa li wile e kulupu pi jan utala 100 tu' . Of course, we would intuitively solve this by moving the 'tu' to attach directly to 'kulupu', but it is not clear this is always possible or feasible. Still, for now, this looks like a safe case.
Compound colors, like 'laso jelo' or even 'loje jelo walo' work on the notion of many words just meaning one thing. That it is one thing is, of course, L1 chauvinism for the most part. And it does conflict with reasonable alternatives. I had no sooner thought that 'laso jelo' was not a problem than I remembered my wife complaining about how yellow the broccoli had become 'kasi laso jelo li jaki. taso kasi pi laso jelo li pona mute'. Now, in the preparser, you may want to stick in a hyphen or whatever in the latter case (but you don't need it) but I don't think this will work as a tag in the language generally.
Repetitions. There seem to be several slightly different cases here. Even the main cases, 'mute mute (mute)', 'pona pona' and 'ike ike', of intensification, seem to work differently. The non-count words (those except for 'mute' and sometime 'lili') seem to rarely give problems in meaning:
'jan pona pona' does differ from (official) 'jan pi pona pona') but only because of the idiomatic nature of 'jan pona'; 'soweli pona pona' is not a problem. But problems might arise in more complex cases, just as with numbers: 'kulupu pi mani pona pona': I assume that there is occasionally a difference between a herd of very good sheep and and good herd of good sheep. But, as in the case of numbers, shifts seem to come to the rescue.
The problems with 'mute' and 'lili' (and some other considerations) also doesn't arise at the first level: 'jan mute mute' is going to mean quite a few people however it parses, though you may have different ideas about how they are grouped. But 'soweli pona mute mute' goes off in three ways (at least) if we allow reduplicating tags: "very many good critters", "many very good critters" or just "very good critters". But, this goes too far in ignoring 'pi'; keeping serious use of 'pi', the last two are properly 'soweli pi pona mute mute' and the second would, following the now familiar pattern, be changed to 'soweli mute pi pona mute' (with a little philosophical violence, but none practical). So, the final enumeration position can stand repetition (and presumably also 'mute'-'lili' mixes) without needing 'pi'. And the same would be true for the "very" (and "barely"?) uses at the end of string, provided we always move enumerations forward to head (where the repetition can also be used). [5/20 But I just had occasion to suggest 'jo e linja lili mute' for "fuzzy" and that clearly has to be distinguished from 'linja pi lili mute' "nearly bald" and no shifting is available that doesn't make thing worse. I am now of the mind that the 'pi' rule had better be rigidly adhered to.]
Repetition for verbs (intense or repeated or repetitive actions) don't see to make much of a problem, since the modification structure around verbs is markedly less complex (or less well studied or used). The case of modals is possibly tricky semantically, and could even introduce extra levels structure if tags were not used, so tags seem appropriate here (but there aren't many examples yet . but i had a vision of a guy dancing around saying 'mi wile wile pana e telo jelo' "Man, i really gotta go!")
I can't think what reduplicated prepositions would do.
If 'mute lili' and the other combinations are allowed as tags in some places, one wants to ask about structures with 'ala', but the history of philosophy tends to say that the scope of "not" has to be kept clear everywhere or all Hell breaks lose. And similarly for 'ali'.