PCN is little more than a counter; it has little to do with the disk contents. It reflects the physical position of the head carriage, as the FDC understands it (zero when the track 0 sensor is active; +1 for each step in and -1 for each step out). Whether or not the carriage is actually at the position, depends on the drive electromechanics.
So if a seek fails, you'll get a "sector not found" error (D2 in SR1). Some of the description in the Intel 8272A datasheet is just plain wrong; a seek is basically a simple mechanical operation--you don't even need the head loaded to do it--the same as if you tried to access a non-existent sector.
Things have to work that way, if, for example, you're trying to read a 360K disk in a 1.2M drive. You seek to twice the cylinder desired, which means that the only place that the NCN will match the ID headers on the disk is track 0.
You'll get a seek error if the track 0 signal doesn't come at the expected place. In other words, if you're at cylinder 5 and you seek to cylinder 0 and don't get an active track 0 signal from the drive.
You have to recall the vintage of the 765/8272--it's basically a 1978-vintage device originally intended for 8" drives. It's smarter than its WD 17xx counterparts, but not by much. For example, it will track the carriage position of up to 4 drives and will automatically produce the correct data (according to the map and gap data submitted) for formatting.
It's better than the immediate predecessors; the NEC 372 and the Intel 8271, but it's not a computer giant intellect. Like the x86 PC platform, after the 5150, the architecture was pretty much frozen--it was embellished, but the basic operating mode didn't really change.