A host that refuses Apify's servers, or answers only from some countries, gets one more try through the Apify datacenter proxy (US-only government sites, hosts that block cloud IPs). Hosts that already worked never go through the proxy. Same price per PDF.
timeoutPerPdfSecs now limits the wait for data, not the whole download: a large file from a slow host that keeps arriving is no longer cut off (an 18 MB manual took 77 s). A host that sends nothing for that long still fails the row.
Clearer per-file errors when the host itself fails (HTTP 400, 405, 5xx) or does not answer: the row says to check the link in a browser or try later.
More public PDFs download. A host that still refuses after the browser-header retry, drops the connection (seen as UND_ERR_SOCKET), times out on connect or answers HTTP 520 is tried with a plain HTTP client and two neutral request profiles. A host with an expired, mismatched or incomplete HTTPS certificate is downloaded without checking the certificate. The file must still be a valid PDF. Hosts that already worked get the same single request as before.
Clearer per-file errors: a domain that does not exist, a file the host reports deleted (HTTP 422) and a file that needs a login (HTTP 401) now say so.
A pageRange wholly past the end of a file now says how many pages the file has, instead of suggesting the PDF is image-only. Such rows stay free.