Extensive Edit - I'm genuinely seeking a better understanding of this so I am putting up a new version of this question. I used an example design to make the matter easier to discuss, but it was pretty bad. I hope this one is good enough to place the topic within parameters that makes an analysis possible.
Design principles: A good design will restrict heat loss through conduction into the soil by placing thermal mass such that heat needs to traverse a large distance to reach the ground, and surface area of the thermal mass in direct contact with the ground is small. This implies building tall, thick walls. Let us consider a cylinder with height equal to diameter, being 40 m, with walls of 3 m thickness.
This is manageable through the use of earthbag techniques. Only the fabric bags or tubes to be filled with regolith would need to be shipped from Earth, and the equipment needed to fill and place them is also simple and wouldn't be too heavy. Placing the walls would be slow and require the movement of the loose top layer of regolith from a large area, since that layer is only 10 to 20 cm deep. On the other hand, if your equipment runs on solar power and can be operated remotely, time is cheap compared to the cost of shipping.
The radiation penetrating such walls would be inconsequential, leaving only that entering through the transparent covering at the top to consider. For the sake of argument let that covering be a shallow dome of plastic sections with protective coatings, and a high-strength membrane beneath holding in the atmosphere. 50% of light passes through, reflected in by a mirror of thin aluminum sheets on a light armature. That mirror has dimensions matching that of the transparent roof, is backed with MLI (Multi-Layer Insulation that reflects EM radiation), and is lowered over the opening at night to hold in heat. During the day it tracks the sun. Let's suppose the structure is 60 degrees N or S of the equator. The transparent area of the greenhouse would receive 1700 kW if it was in full sun. At 60 degrees, it would get approximately half that amount directed into the greenhouse, between the mirror and what directly enters, over the course of a day. Half of that would be blocked by the membrane and plastic, so 425 kW enters on average.
Limiting heat loss by radiation is simple through use of MLI, which is light and so costs little to ship from Earth. Sheets of it would cover the exterior walls. Below is an illustration of how heat would spread outwards from such a structure. This article says temperatures in the ground are 'fairly constant' at depths of 2 m, and are 'around -30 to -40 degrees C'. Let's call 3 m the depth at which temperatures are constant. If there is a 3 m skirt of MLI around the base of the cylinder, the beginning of the gradient between the surrounding soil and the greenhouse is a figure following that 3 m distance from the structure's thermal mass. Temperatures outside that area have no bearing on interior temperatures.
Now the math for how the heat in this system would move around over the course of many lunar days and nights before it found an equilibrium is very much beyond me, but with some rough calculations based on all the figures listed, it seems like that equilibrium could yield a fairly constant, comfortable temperature. Is that true?
I've assumed that all the contents of the greenhouse add up to 5000 m3 of stuff (10% of volume) with an average density of 1200 kg/m3 and average specific heat of 2 kJ/kgK. The sunlight into this system would be enough to heat everything in it by about 8 Celsius in one lunar day once the heat was evenly distributed. At first some of that would get sucked into the ground under the greenhouse but soon that would be saturated with enough heat to stop it from doing that any more. Some would be lost through the transparent dome. I don't know how much that could be, but I don't think it would be enough to stop the greenhouse slowly heating up. How would I calculate that?
Then some heat would be lost over the course of the lunar night. Fine-tuning for cooling is easy, by maybe keeping part of the walls in shade uncovered by MLI so they radiate heat slowly, or turning the mirror a little so it reflects in a bit less light, but I don't know how much difference that would make.
A temperature fluctuation over one complete lunar day within 10 or 15 Celsius would be fine. As long as temperature doesn't change rapidly plants adapt well.
Figures used for thermal properties and sunlight:
Thermal conductivity of regolith: 0.015 W/mK to 0.22 W/mK (both figures were quoted in the linked paper from two different sources, on page 16) Density of regolith: in the compacted layer below the powdery surface, > 1.5 g/cm3 (this was noted in the conclusions near the end of the paper, it seems good figures have been hard to establish because of the difficulty Apollo astronauts had getting drilling samples)
Heat capacity of regolith: about 0.8 kJ/kg*K (this is taken from dry soil and sand, couldn't find anything better)
Heat capacity of water: 4.18 kJ/kgK, of moist soil: 1.48 kJ/kgK Sunlight on lunar surface: average of 1360 W/m2