Actually, sending the zero signal to a pid as in `kill -0 <pid>` checks whether that pid exists and whether the calling process has permissions to send signals to it.

... at the point that the system call completed. There are no guarantees from that point onwards.

Interestingly, it would only take the addition of a pdopen() to something like FreeBSD's pdkill() to get enough for a tool that opened a process descriptor, sent a signal to the process, and determined (with poll() and pdwait4()) whether the process then exited; with the open-descriptor-style interface guaranteeing that it was always referencing the same process.