Poking around and git bisect point to 7f430f142d7f19cb575f0e665719619c9312ae57.

Comparing the request pattern for strings vs the math datatype highlights the issue.

String (parse then format)

https://test.wikidata.org/w/api.php?action=wbparsevalue&format=json&parser=string&values=23&errorformat=plaintext&uselang=en&options={"lang":"en"}
https://test.wikidata.org/w/api.php?action=wbformatvalue&format=json&datavalue={"value":"23","type":"string"}&generate=text/html; disposition=verbose-preview&property=P95180&errorformat=plaintext&uselang=en&options={"lang":"en"}

Math (only a format with the JS specifically stating the value is unknown as no parse has happened )

https://test.wikidata.org/w/api.php?action=wbformatvalue&format=json&datavalue={"value":"23","type":"unknown"}&generate=text/html; disposition=verbose-preview&property=P719&errorformat=plaintext&uselang=en&options={"lang":"en"}

I was debugging this issue. However, I don't know where to start searching for the problem I was looking at https://github.com/wikimedia/mediawiki-extensions-Math/blame/master/src/MathWikibaseHook.php. However, I was unable to identify the key difference between Math and all the other data types, that are still functional.

I tried testing it on https://test.wikidata.org/wiki/Q516#P717 and it still doesn't work. Is that expected because we have to wait for wmf-18?

In case it helps: I had been able again to add math values for some time last week, but now it fails again.

Moving back to todo for investigation.

As @Lucas_Werkmeister_WMDE mentioned in mattermost, there seems to have been a regression in wmf.16 and we will need to backport the revert in T266671 back up. The permanent fix should be rolling out with wmf.18.

Should be fixed now (T266671#6615909).

It probably makes sense to wait with verifying this until next week, so that we verify not just the backported revert (wmf.16) but also the proper fix (wmf.18).