This is not a matter of truth or logic.  It is merely a matter of convention, like CSX before it.  IAST certainly defines anusvara as m+underdot.  Someone must have edited the Wikipedia page.  I've looked through a few pages of that page's history, and I didn't spot who made the change, but it doesn't really matter.  I've edited the Wikipedia IAST page back to normal (ṃ).  Thanks for spotting this, Harry.  When I edited the page, I added the comment,

"IAST anusvara is dot below.  IAST follows the conventions of written scholarly communication on South Asian studies as established since ca. 1900.  This differs in small ways from Unicode and the Library of Congress standards. "

The document headers of files in the SARIT library explain this difference between IAST and the ISO 15919 encoding.  Note that Unicode names dot-above only for indigenous Indian writing systems like Devanagari, but it does not legislate for Indic transliterature into Latin script (so my Wikipedia comment above was slightly wrong).  As far as I know, it was the Library of Congress that started promulgating diacritics like under-circle for vocalic ṛ.    (The LC documentation for Sanskrit is the same as IAST except for vocalic r and l).  This usage was then changed a bit more and codified by the committee members of ISO 15919 "romanization of Indic scripts". 

I think the main differences are as follow:
——
IASTISO 15919
ṃm-overdot
ṛr-under circle
ḷl-under circle
ee-macron
oo-macron

Best,
Dominik

--
Dominik Wujastyk, Professor Emeritus, Classical Indian History
University of Alberta

"The University of Alberta is committed to the pursuit of truth, 
the advancement of learning, and the dissemination of knowledge 
through teaching, research and other scholarly and creative activities and service."
-- Collective Agreement 3.01