It seems like some non-nomalized (ie non NFC - non "normalized form C") is injected into the html and some validators mark this as a failure. It can be questioned if this is really a failure or just validators being picky, but it should be a decision if we want this to be like this or not.
Note that parsing the text or calling htmlescapespecialchar will not fix this issue.
Version: unspecified
Severity: minor
Go to Helium on test (http://wikidata-test-repo.wikimedia.de/wiki/Data:Q2) with Firefox, right click and select all, right click and view selection source, make sure all is selected, right click and copy, go to W3 Validator (http://validator.w3.org/), chose tab validate by direct input, paste content from selected source.
To get the doctype right copy this from the normal source page.
Also mark for verbose output.
There are several errors, some that should be easy to fix, but note errors like
Warning Line 87, Column 2666: Text run is not in Unicode Normalization Form C.
…idhlig","gl":"galego","glk":"گیلکی","gn":"Avañe\'ẽ","got":"