Detail #9: Unusual ordinals (again)

Getting back to the ordinal-vs-cardinal bit, how about a language which forms its cardinals a bit like Finnish:

1-7 [unanalyzable things]

8-9 [seem related to 2 and 1, with a suffix not found anywhere else]

10

I am not sure this is how eight and nine have been formed in Finnish, but a simple hypothesis would be that their meanings originally have been something along the lines of two away from (ten) and one away from (ten). In a language where the situation is more transparent, say something like base-12, sub-base 6:

1 em
2 da
3 tchi
4 fij
5 sa dun
6 dun
7 kha
8 ssem
9 dutchi
10 dufij
11 sa lop
12 lop

Notice how 9 and 10 are formed using the sub-base dun, whereas 7 and 8 are unanalyzable; 5 and 11 are formed from the base and sub-base by a morpheme that marks incompleteness (and also has verbal and adjectival uses). Now, ordinals are formed by simple applying the adjectival suffix -((a)r)a

ema, dara, tchira, fira, ..., duna(ra), khara, ssema, duratchira, durafira, ... lopara.

sa dun and sa lop instead are formed by a different formulation, applying a different prefix and different adjectivization:

ereduni, erelopi

Of course, first and second may also differ, along the lines of first, second, so something like
toma, seva or whatever. Once lop is passed, larger numbers are formed by juxtaposition: lop em, lop da, ... lopara-toma, lopara-seva, lopara-tchira, lopara-fira, ...

24 is formed as dalop, thirty as tchilop, etc.

This seems a more likely and realistic thing than having the entire ordinal system offset by one, even though I have a soft spot for that weird construction.