|Your JW part will accept new calibration memory setting, and therefore won't
|run with the correct timing unless you SET the calibration memory for each
|individual chip.  If you didn't write this down before you erased the chip
|first time, (how many times have I forgotten to do this?) then it is gone.

|Trust that the OTP calibratrion memory is programmed properly.  You can
|whistle confidently past the graveyard, no ghosts will come out to haunt you
|this time.

On a "nice" programmer, if your code includes the proper opcode for a
calibration instruction but that location in the chip contains a different
value, the programmer will just accept the one in the chip (not try to
burn the bits that happen to be different).  If you are using such a prog-
rammer, you can just include a movlw (50x's) or retlw (67x's) at that loc-
ation and be fine whether you're using OTP or /JW parts.

Otherwise, for the '50x parts you may (if you don't care about oscillator
speed) leave the location blank and things will still work.  On the '671,
this won't work since a call to the last location will cause code execu-
tion to wrap to the first location creating and endless-loop condition.

To avoid this, you may wish to code something like [at address zero]:

        btfsc   PCLATH,0
         goto   START
        bsf     PCLATH,0
        movlw   default_cal_value_plus_one
        call    $7FF
START:
        movwf   Calib

If location $7FF is unprogrammed, this will use the default calibration
value; otherwise, it will use the one at $7FF.  Similar tricks may be used
on the '50x parts to allow for a useful default when the calibration addr-
ess isn't programmed, but care must be used with the bank-switching jumps.