@dkg, I changed the title and adjusted the description to more accurately describe the situation.

Those changes make the script work for me, specifically passing the input as an argument and not through standard input. Digging more, it looks like the underlying issue is related to using pinentry-tty (my case) or pinentry-curses when passing the OpenPGP input via standard input. This causes pinentry to give up before prompting. For pinentry-tty it fails with "ERR 83886340 Invalid IPC response" and pinentty-curses fails with "ERR 83918950 Inappropriate ioctl for device".

Considering that early interop testing, you're probably right that this is a bug in the spec, not GnuPG. Otherwise this would have been pretty obvious long ago. The wording in RFC4880bis hasn't been corrected to match practice, so I should probably report this issue there.

Re-examining this now, I'm noticing the problem is not at all that it's being treated as signed, but that GnuPG is internally using a 32-bit unsigned integer for the time even though the key expiration scheme allows for expiration dates beyond 2106. Seeing dates in the past threw me off, and when I had originally tried using a zero creation time to test a broader range I ran into T4670.

I'm using Debian 10 "Buster" on x86-64, but for this ticket I used my own build of GnuPG so that I could demonstrate with the latest version. The system's GnuPG 2.2.12 has the same behaviors I showed here.