Can someone judge the Priority of this task?

It seems that this is a problem which appears only in PNG rendering mode.
See http://en.wikipedia.beta.wmflabs.org/wiki/User:Physikerwelt/sandbox

Change 272818 had a related patch set uploaded (by Physikerwelt):
Only use parallel rendering in MathML mode

https://gerrit.wikimedia.org/r/272818

Change 272818 merged by jenkins-bot:
Only use parallel rendering in MathML mode

https://gerrit.wikimedia.org/r/272818

Should this still be happening on ptwiki? See https://pt.wikipedia.org/wiki/Axioma_da_escolha?oldid=45562403

In T127738#2275723, @He7d3r wrote:

Should this still be happening on ptwiki? See https://pt.wikipedia.org/wiki/Axioma_da_escolha?oldid=45562403

Does that page show with 'UNIQ's to you? Where? (They don't for me.)

@He7d3r the problem you are seeing... does it also occour here
http://de.wikipedia.beta.wmflabs.org/wiki/Benutzer:Physikerwelt/sandbox1

This is what I see on ptwiki, with MathML enabled at Special:preferences#mw-prefsection-rendering:

Na matemática, o axioma da escolha é um axioma da teoria dos conjuntos equivalente à afirmação "o produto de uma coleção não-vazia de conjuntos é não-vazio". Mais explicitamente, diz que para toda família indexada '"UNIQ--postMath-00000001-QINU"' de conjuntos não-vazios existe uma família indexada '"UNIQ--postMath-00000002-QINU"' de elementos tal que '"UNIQ--postMath-00000003-QINU"' para todo '"UNIQ--postMath-00000004-QINU"' Foi formulado em 1904 por Ernst Zermelo.[1] Até o início do século XX era um axioma controverso, mas graças ao trabalho de Zermelo, Hilbert[carece de fontes] e outros matemáticos, o axioma da escolha foi satisfatoriamente modelado em lógica simbólica, resultando na teoria de conjuntos padrão da matemática contemporânea, a teoria ZFC - Zermelo-Fraenkel-Choice.

It does not happen at Benutzer:Physikerwelt/sandbox1 (but I'm not logged in)

mh. I get the impression that a wrong command in the {{cite template kills the whole rendering process. So that the postMath hook is never called.

See my workaround for that particular page
https://pt.wikipedia.org/w/index.php?title=Axioma_da_escolha&type=revision&diff=45620774&oldid=45562403

However, in general this can not be a solution.

Same problem on
https://pt.wikipedia.org/wiki/Princ%C3%ADpio_da_n%C3%A3o_contradi%C3%A7%C3%A3o?oldid=45736369

O princípio da não-contradição é representado do seguinte modo:
'"UNIQ--postMath-00000001-QINU"'
Exemplo: Não ("a bola é redonda" e "a bola não é redonda")

This problem does not seem to be related to math. I think it's about the {{Citation needed}} template (uppercase) template in ptwiki. Since I do no speek portuges I am not able to help there.

Minimal test case:

<math>B</math>{{Citation needed|date=April 2009}}

https://pt.wikipedia.org/wiki/WP:Página_de_testes/1?action=edit&oldid=45739340&preview=yes&uselang=en

I used the special page expand templates and pasted the result to https://pt.wikipedia.org/wiki/Wikip%C3%A9dia:P%C3%A1gina_de_testes/5
which looks good.

Looks like the problem does not happen on special pages. E.g. I copied the example above to MediaWiki:Intentionallyblankpage and loaded Special:BlankPage. The bug appeared on the MediaWiki namespace, but didn't happen on the Special namespace.

Here is another test case, which does not use "Template:Carece de fontes" to call the code which depends on the namespace:

<math>B</math>{{Manutenção/Categorizando por assunto|que carecem de notas de rodapé|data=April 2009}}

https://pt.wikipedia.org/wiki/WP:Página_de_testes/1?oldid=45741658&action=edit&preview=yes&uselang=en
and finally:

<math>B</math>{{#invoke:datas|ISO_8601|April 2009}}

https://pt.wikipedia.org/wiki/WP:Página_de_testes/1?oldid=45741679&action=edit&preview=yes&uselang=en

Notice that if we change the order of <math> and {{#invoke:, the math is displayed correctly:

{{#invoke:datas|ISO_8601|April 2009}}<math>B</math>

Another report https://pt.wikipedia.org/w/index.php?diff=46637523.

@Krinkle @Izno I'm happy to provide information what happens within the math extension, but I will not look into the cite extension code to figure out at which point the strip markers get lost.

Here is another example where I was able to get rid of the UNIQ markers by changing {{Refimprove|date=September 2009}} to {{Sem-fontes|data=dezembro de 2016}}:
https://pt.wikipedia.org/w/index.php?diff=47582007

This is happening consistently on test2 right now.

Problem on en:Wikiversity gotten around by changing "cite web" to "cite book" or by changing "|month=July |year=2016" to "|date=July 2016", e.g.