The [nd] > [nn] change would only happen postvocallically (i.e. after a vowel), as far as I know. Or, perhaps, it simply doesn't happen initially and following a morpheme boundary? My notes are conflicting... Could look into this some other time, it doesn't change anything here. What you have here is this:
✶skalla·ndak·ndēr > sxalla·ndak·ndēr > xalla·ndak·ndēr > xalla·ndak·ndīr > halla·ndak·ndīr > halla·ndag·ndīr > hallndagndīr
Now here we come to the application of rule S54, the simplification of long consonant clusters, and things are a bit less trivial. Hypothetically we would at this point expect Haldagdir, and for that matter, we would generally expect ✶ndak·ndēr > dagdir; so why is it dagnir? We don't have a straightforward answer, but it may be proposed that what we see here is a dissimilatory process.
Sindarin seems to disfavor multiple [d]'s in a single word, and we have evidence for that in Finrod (instead of Findrod, whereas -ndr- is a very common cluster), for example. Thus, since the [d] here can be avoided by dropping it in favour of [n], we get dagnir instead of dagdir. And so:
hallndagndīr > haldagnīr > Haldagnir.
Athelas is a very good counter example to what I said about [ae] > [e]! However, if you search through the Sindarin corpus, you will actually find many many exceptions to the [ae] > [e] rule, e.g. Aeglos, Aegnor, Aeluin, Daeron, Annael, Dirhael, Gilraen, Maedros...
What follows now is less of an answer and more of my own thought process:
Let's suppose [ae] > [e] does not apply to disyllabic words, only longer ones (thereby actually ignoring some important examples of the rule, e.g. niphred, negra-, mewrim, but let's leave them for now). Now we find examples such as Aerandir, Belegaer, Agarwaen, Naeramarth, Crissaegrim, Gwahaedir, Faelivrin, Hithaeglir, Linaewen (not the full list). These can be divided into three groups:
-
[ae] in the stressed syllable: Crissaegrim, Gwahaedir, Hithaeglir, Linaewen.
-
[ae] in the first pretonic syllable (i.e. the syllable preceding the stressed syllable): Aerandir, Naeramarth, Faelivrin.
-
[ae] in the final syllable: Belegaer, Agarwaen.
The first group even tolerates superheavy syllables, and these are the examples I was thinking of when I said, earlier, that the reduction shouldn't happen in a stressed syllable. Particularly, had it happened in a stressed syllable that is not superheavy, it would turn the syllable into a light one, and obscure the position of the stress, like in athelas, the sole counter-example - more on that later.
The second group may (or may not...) suggest that [ae] should not reduce in the initial syllable. I would posit that the forces at play are, primarily, stress, the weight of the syllable, and whether or not it is initial; therefore I would posit that an unstressed initial [ae] in a superheavy syllable might reduce after all - but we have no examples. Either way, a counter-example to this group would be negra-, which is even stressed initially in most conjugations...
The final group suggests that a final (and thus inherently unstressed) [ae] might stay as it is; however, we then see niphred as a counter-example. A closer look at the corpus reveals that (unless I missed something) all attested cases of [ae] in the final syllable of a polysyllabic word (including 2 syllables) end in a sonorant consonant (= not an obstruent = not a stop or a fricative), with the following exceptions: rovaed, úthaes, pelthaes, Helcharaes, and the deleted harvaed (deleted by Tolkien, that is; so we shall ignore it). Let's review them one by one:
-
rovaed is a superlative of maed "skillful", so perhaps it can be excused due to possibly being a late morphological form.
-
úthaes, itself derived through dissimilation from úthaeth, is prefixed with the negative element ú-, so perhaps can also be treated as a morphological form.
-
pelthaes is Noldorin (=draft Sindarin), derived from ✶peltakse; this, with mature Sindarin phonological development, becomes pellach, thus not relevant.
-
Helcharaes is also Noldorin, and a cognate of Q. Helcaraxe; thus in mature Sindarin it should be Helcharach, and also irrelevant.
Based on this data I would be inclined to conclude that [ae] in the final syllable of a polysyllabic word is tolerated before sonorants and in cases of morphological prefixes applied to words that already have it there; before obstruents, however, it would reduce in polysyllables (e.g. niphraed > niphred).
The case of mewrim, a class plural of maew "gull", is interesting, because here the [ae] is reduced in a morphological form, in a stressed initial position; it does however simplify a superheavy cluster. I would guess that the presence of the semi-vowel [w] is the deciding factor in this one.
And that leads us back to the case of athelas -- and I really have no answer for this one. Given clear examples of similarly structured words where the reduction didn't happen (Gwahaedir, Linaewen), athelas seems aberrant.
A further thing to consider is the question of stress. This topic definitely requires more research, and I am not sure there's even enough data in Sindarin to go on... But I would somewhat cautiously assume that, at this relatively late stage of the phonological development of Sindarin, no segmental process would normally occur that would turn a stressed heavy syllable into a light syllable, with one sole notable exception: [mm] > [m], which is a late universal change in Sindarin, not always evident in spelling, which Tolkien explains specifically, and mentions that it does not affect stress, e.g. Naeramarth is pronounced with stress on the second syllable (and this is one of those instances where Tolkien says one thing and does another; if he followed his own convention, he'd spell it Naerammarth, to clarify the position of the stress - though it would still be pronounced with an ungeminated [m]).
In other words, what I am saying is that I would be cautious and assume that [ae] > [e] in athelas is an irregular/aberrant change that we shouldn't repeat ourselves.
Which is all to say that I am still in favour of Athaedor rather than Athedor.
Final note: dolen is from ✶ndulina, derived directly from the root, thus no geminated [l] to deal with in the first place; at least that is how I would reconstruct it.
