Dear Harry,

IAST is not an absolute standard in the way ISO-15919 is; it's more like a set of conventions, without a definitive document and hence malleable at the edges. There is, for example, no definite provision in IAST for the upadhmānīya and jihvāmūlīya (the transliterations listed on the Wikipedia page for IAST are just one of the options in use), nor for the Vedic retroflex l, much less for Dravidian retroflexes and alveolars. The same Wikipedia page gives ḻ for the retroflex l, which I have never seen before and which clashes with the convention of using ḻ for the sound in e.g. Tamiḻ.

I personally have never heard of a flavour of IAST that uses an overdot for the anusvāra and agree with you that the IAST anusvāra is with an underdot. The first of two random Google hits agree: https://www.omniglot.com/writing/sanskrit.htm and https://fpmt.org/wp-content/uploads/education/translation/guide_to_sanskrit_transliteration_and_pronunciation.pdf

I dare say that as far as IAST can be considered a standard, the "correct" IAST anusvāra is ṃ, while ṁ is an informal alternative. So, put in so many words, yes, Wikipedia is wrong.

See also the stub on the discussion page for the IAST article: https://en.wikipedia.org/wiki/Talk:International_Alphabet_of_Sanskrit_Transliteration#Anusvara

All the best,

Daniel