Help getting custom error pages from S3 into Cloudfront Distribution

I've been reading this page about custom error pages in CloudFront, and going a bit crazy trying to get custom error pages set up using CloudFront and S3.

I have an S3 bucket that's set up as a static site. You can access it here:

https://cl-error-docs-static-website.s3.amazonaws.com/

And I put a custom 404 page in it that you can access here:

https://cl-error-docs-static-website.s3.amazonaws.com/errors_5xx/404.html

I think that should be enough on the S3 side of things, so now it's just about having CloudFront access that page when it gets a 404 from the normal origin (an ALB).

In CloudFront, I:

• created a new origin and pointed it at the S3 bucket.
• created a behavior in CloudFront with a Path Pattern of /errors_5xx/*, which should serve up the S3 bucket nicely.
• created custom error pages that return /errors_5xx/404.html whenever there's a 404 response from the normal origin.

I think that should do it! But when I load a page that returns a 404, like this one:

https://www.courtlistener.com/some-random-url-asdf

I get the following S3 result:

<Error>
<Code>NotFound</Code>
<Message>The resource you requested does not exist</Message>
<RequestId>SCN2DCJBR4JRHCJR</RequestId>
<HostId>
OEG57JmYL5dclH5biQ8/QrlEsEmITH+6fwmu/01feEjeT0TCARqdklfKryvcdzg1EM9f3FpEATI=
</HostId>
</Error>

Seems like CloudFront is getting an error from the normal origin (the ALB), then going to the bucket and looking for the same path instead of looking for page that's at /errors_5xx/404.html.

It feels like I'm missing something, but I've tried everything, I think, read documentation many times, and even pulled up old pages from the Internet Archive.

Help!