Background is already set on the outer <figure>-element:

figure[typeof~='mw:File/Thumb'],
figure[typeof~='mw:File/Frame'] {
 border:var(--border-subtle,1px solid #c8ccd1);
 border-bottom:0;
 background-color:var(--background-color-interactive-subtle,#f8f9fa);
}

I don't see any reason why the same background is set again on the <img>:

figure[typeof~="mw:File/Thumb"] > :not(figcaption) .mw-file-element, 
figure[typeof~="mw:File/Frame"] > :not(figcaption) .mw-file-element {
	[...]
	background: var(--background-color-interactive-subtle,#f8f9fa);
}

Removing that line would solve this issue, as .skin-invert-image won't invert anything that is defined outside the <img>.

This issue may overlap with T365318 as the same CSS code is mentioned there.

In T366069#9837932, @Aklapper wrote:

Switch to dark mode

Hmm, how/where exactly to do that?

Thanks for your comments. Now I had a look at the source code myself. It does define the correct URL:

<link rel="canonical" href="https://www.transfermarkt.de/wechsel-im-winter-brentford-verpflichtet-dervi%C5%9Fo%C4%9Flu-watkins-verlangert-langfristig/view/news/342949" />

Open Graph has a different purpose and - in my opinion - shouldn't be used here. Or at least shouldn't be preferred over '<link rel="canonical"'.

I the fix live? Because there is a new case on dewp:

New case on dewp: https://quarry.wmflabs.org/query/42941
Must have happened in the last two days, because editing and reviewing the page still worked then: https://de.wikipedia.org/wiki/Spezial:Logbuch?type=&user=&page=Bell+Food&wpdate=&tagfilter=&wpfilters%5B%5D=review

In T237191#5910130, @Zache wrote:

AND (frp_revertedEdits>frp_totalContentEdits) < @revertedEditsRatio

Additionally there are 141649 records in the table 'flaggedpages' pointing to non-existing pages: https://quarry.wmflabs.org/query/42871

The cases on deWP are know fixed by temporary moving the pages to a non-review namespace (which deletes the wrong db entries) and moving them back afterwards.

In T246720#5940537, @Zache wrote:

In T246720#5938796, @Wickie37 wrote:

Looks like I was right with the faulty 'fp_stable' in the 'flaggedpages' db table:

Yes, and the fp_stable values should be unique least in the cases where both pages exists. However, I don't know how it handles the cases where page is deleted and then undeleted.

Looks like I was right with the faulty 'fp_stable' in the 'flaggedpages' db table:

Another page with same issue: Andreas M. Fleckner

I looked a bit through the code, the wrong page ID comes from this db query in FlaggedRevision::newFromStable().
When I got this right, it takes the page entry from the 'flaggedpages' table and joins the revision given by the 'fp_stable' field. And the wrong page ID is the one from the revision table.
So I would say, that it is most possibly a faulty 'fp_stable' entry, that could cause this behaviour.

And editing still doesn't work when "review last changes" is checked

Did someone change anything? Editing as IP was possible, and now I can even edit the page when logged in. But reviewing still doesn't work.

And this redirect page has the page ID 2784562 mentioned in the above Stack Trace...

One curios thing I forgot to mention: The "review" link on the "pending changes" page doesn't even link to the right page, but to Hnutie za demokraciu, which is a redirect to the page.

Sorry, this didn't come up on my search yesterday, so I opened a new task which most likely adresses the same issue: https://phabricator.wikimedia.org/T227033