The internal urls also behave properly so it seems that the issue is not on the Lift Wing side but has to do with how the API Gateway translates/encodes the URL.

curl "https://inference.svc.codfw.wmnet:30443/v1/models/article-descriptions%3Apredict" -X POST -d '{"lang": "en", "title": "Clandonald", "num_beams": 2}' -H  "Host: article-descriptions.article-descriptions.wikimedia.org" -H "Content-Type: application/json" 

{"prediction":["Hamlet in Alberta, Canada","human settlement in Alberta, Canada"],"blp":false,"lang":"en","title":"Clandonald","num_beams":2}

I suspect the fix for this is a relatively small change on the API gateway, but the change is a global one so I will need to take some time to test this, even if the impact is to make things standards-compliant. Hoping to get to it later this week

Change #1035481 had a related patch set uploaded (by Hnowlan; author: Hnowlan):

[operations/deployment-charts@master] api-gateway: add normalise_paths option, enable in api-gateway

https://gerrit.wikimedia.org/r/1035481

Change #1035481 merged by jenkins-bot:

[operations/deployment-charts@master] api-gateway: add normalise_paths option, enable in api-gateway

https://gerrit.wikimedia.org/r/1035481

The normalisation change has unfortunately not fixed this issue - docs indicate that it should have but I suspect this is something to do with the use of regex matching as opposed to static matching. I'll try to come up with a workaround for the short term

It seems Envoy only normalises a subset of urlencoded characters:

hnowlan@plunkett ~/Code/deployment-charts (hnowlan/T365439-apigw_normalise_path_urls *) $ curl -s localhost:8087/core/v1/wikisource/a/%3A| grep original-path
    "x-envoy-original-path": "/core/v1/wikisource/a/%3A"
hnowlan@plunkett ~/Code/deployment-charts (hnowlan/T365439-apigw_normalise_path_urls *) $ curl -s localhost:8087/core/v1/wikisource/a/%31| grep original-path
    "x-envoy-original-path": "/core/v1/wikisource/a/1"

I'll file an issue upstream, but short-term we'll need a workaround.

Change #1037449 had a related patch set uploaded (by Hnowlan; author: Hnowlan):

[operations/deployment-charts@master] api-gateway: add workaround for urlencoded characters in liftwing

https://gerrit.wikimedia.org/r/1037449

Change #1037449 merged by jenkins-bot:

[operations/deployment-charts@master] api-gateway: add workaround for urlencoded characters in liftwing

https://gerrit.wikimedia.org/r/1037449

I am now seeing results when using queries with urlencoded characters. Unfortunately we will need to add a manual hack if there are other non-alphanumeric chars in other parts of the URL in future, but for now I think this works:

hnowlan@plunkett ~ $ curl -s "https://api.wikimedia.org/service/lw/inference/v1/models/outlink-topic-model%3Apredict?hnowlan=123" -X POST -d '{ "rev_id": 123555, "lang": "en", "page_title": "foo" }'  -H "Authorization: Bearer $TOKEN"
{"prediction":{"article":"https://en.wikipedia.org/wiki/foo","results":[{"topic":"STEM.STEM*","score":0.9481645226478577},{"topic":"STEM.Computing","score":0.8080772161483765}]}}