you can (and should) put monitors on the outside of your infrastructure as well as inside. I've multiples hitting the login page and robots file where I work, and never got an unaccounted 503
Like I wrote, for some integrations it is not possible to gather all the logs. Also how will you know that a client accessing your website in a browser gets 503 instead of your web page?
You don't, but your monitors will show the 503 happening, how often, on which endpoint operation and in which regions; that will give you a pretty good picture of whether is actually your CDN layer or something else triggering the 503