Just to confirm we are both working on the same system, I have -

Number of radials = 36
Length of radials = 10 m
Diameter of radials = 2 mm
Frequency = 7 MHz
Antenna height = 9 m
Antenna diameter = 1.64 mm = 14 AWG
Ground resistivity = 150 ohm-metres
Ground permittivity = 16

IMPORTANT:

If NEC4 gives you the input impedance of the radial system I should be
very pleased to know what it is.

Otherwise we shall have no idea where the discrepancy arises - in the
radial system or in the antenna efficiency calculation.

Radiating efficiency is estimated by my program by the well-known
formula -

Efficiency = Rrad / ( Rrad + Rradials )

provided antenna and radials reactance are tuned out.

Whereas NEC4 calculates efficiency by integrating power flow over a
hemisphere WITHOUT tuning out antenna and radials reactance.
Altogether different.

Correct Reg, Those are the parameters I used, with the
exception that the radials were also # 14 AWG (1.64 mm).

You raise some interesting points -- How do I measure
the radial impedance? I have to think; given a vector
network analyzer, how would I measure a radial system
under laboratory conditions? this is what I need
to replicate with NEC. Since I have never made
such a measurement, I am not sure where to begin.
Would it be valid to consider one radial wire as
an "End fed zepp", and feed one end with an
ideal transmission line? As long as I know the current,
and voltage at the measurement point, I can determine
the input impedance -- problem is; voltage input
with reference to what?

As for the reactive input; this is of little concern to
NEC since it drives the load from a complex
conjugate source.

So far as I have been able to determine NEC does
not provide the total radiated power, only the
normalized far field in peak "Volts" -- i.e. V/m at
1 meter, at every angular increment. Usually
every degree. I take these data to determine the
power density at each increment, and sum
over a hemispherical region; where I take the
elemental area to be:
(r^2)*sin(theta)*d(theta)*d(phi). Since the
pattern is symmetrical I only need 91 points.

Frank