The HC-SR04 is not calibrated, nor does it take any measurements.
All it does is allow software on a connected microcontroller or system like an Arduino to measure the sum of the round trip time delay and some small internal offset delay, and from this calculate distance.
As the timing and calculation are entirely in your program, calibration is entirely up to you.
Under known conditions you could measure both a short and a long distance. This would give you a slope and intercept - the slope from the speed of sound (which should not depend on the device at all, only on the conditions and your timing reference), and the intercept or offset, having to do with latencies in the system.
You could then cross-check at other distances, but under fixed conditions behavior should be linear.
As there is some initial ringing of the sender which can couple to the receiver and be mistaken for an echo, there will be a minimum distance below which you cannot reliably take a reading - as a guess, this is more than 5 cm, but you'll have to see, and it may even vary a little from unit to unit.
To a degree, it's a little bit amazing that these even work. They use an RS232 level shifter to generate transmit pulses, and it's been alleged that the receive circuit is horribly mistuned, not really matching the emitted frequency at all... but despite that they do seem to work fairly well.