Currently, all requests are queued as soon as they are received and then evaluated once they enter the execution phase. However, requests that are erroneous (like bad/missing titles) should be outright rejected and not even enter the queue (in this way freeing up the queuing slots for requests that can be successfully completed). This can be achieved either by asking the MW API information about the page via the mwApiGet() utility function or by asking RESTBase via the restApiGet() function from the same module (faster).
Going a bit further, the info returned can be used to correct a possible inconsistency in the rendering mechanism. Namely, if there is a new revision of an article between the time the user requests a PDF and the actual call to puppeteer to render it, the user will receive the new version, not the one they asked the PDF for. By requesting info about the page as soon as the request is received, Proton can retrieve the HTML for the specific revision the user is asking the PDF for.