Screenshots:

This seems like yet another case of T288897

ping @MSantos

This corresponds to previous revision of respective OSM object. It was corrupt briefly around September 27th, see https://osm.virtuelle-loipe.de/history/?type=relation&ref=192800.

In osm.org bad tiles are usually swiftly renewed in such case. Wikimedia apparently doesn't have as reliable mechanism for this and so bad tiles may be cached for months.

Hmm, so we have cached entries tiles from over a month old indeed.

https://maps.wikimedia.org/osm-intl/7/77/61@2x.png

Caching relevant headers:
Last-Modified: Thu, 01 Jan 1970 00:00:00 GMT
Cache-Control: public, max-age=900, s-maxage=86400
Server-Timing: cache;desc="hit-front", host;desc="cp3051"
Date: Mon, 14 Nov 2022 15:18:04 GMT
Age: 51
ETag: "d0ea5f8a5bb0a4187cf0bae95060a1e9"
x-cache-status: hit-front
x-cache: cp3055 hit, cp3051 hit/4

So the tile has a cache time of 5 minutes, with a server cache time of a day... but that's not what this is I think.

I'm assuming it is just using one of the broken (out of sync) db nodes ? T321885: Resync stale maps postgres replicas

In T322104#8392818, @TheDJ wrote:

I'm assuming it is just using one of the broken (out of sync) db nodes ? T321885

I don't really know, but if this (also) has an effect here then I'd suspect that it rather affects tiles that didn't go blue as the data in these stale replicas is generally much older than September 27th.