On most unmanned programmes the link budget is closed late, by one person, against assumptions nobody else reads. It then quietly determines the operating radius, the launch site options and the crew size, none of which are radio decisions.
The order things happen in
The airframe is sized first. The payload is selected. The operations concept is sketched around a range figure that came from the endurance calculation. Somewhere after that, someone is asked to confirm the communications link will support it.
By then the antenna locations are constrained by structure that exists, the available power is whatever is left, and the mass budget has no slack. The link budget is being asked to ratify a decision rather than inform one. It usually can, on paper, by leaning on assumptions that are individually defensible and jointly optimistic.
Where the optimism accumulates
Free space path loss is the honest part of the calculation and the part everyone gets right. The margin gets eaten elsewhere.
Antenna gain is quoted from the datasheet, measured in an anechoic chamber on a ground plane that is not your airframe. Mounted on a real vehicle, next to real structure, the pattern has nulls in it. Those nulls are at specific aspect angles, which means the link degrades as a function of vehicle attitude relative to the ground station: exactly when the aircraft turns, which is exactly when the operator is most likely to be doing something.
Terrain is modelled as line of sight, and line of sight is computed to the aircraft at altitude, over terrain sampled at whatever resolution the tool defaulted to. Fresnel zone clearance is frequently skipped. Ground reflection multipath is skipped more often than that, and it is what actually causes the dropouts at low elevation angles that get written up as unexplained.
Interference is assumed absent, because it is difficult to predict. In the 2.4 GHz band, near anything populated, it is not absent.
Each assumption is reasonable in isolation. Stacked, they routinely account for six to ten dB, which at the margins is the difference between a link that works to the edge of the operating radius and one that works to two thirds of it.
The consequence is operational
This is why it belongs on the programme risk register rather than in the radio section. A link that closes at two thirds of the planned radius does not usually get fixed by a better radio at that stage. It gets absorbed by the operations concept.
The mission is flown from a closer launch point, which means more sites, more transit, and more crew days per unit of work. Or a relay is added, which is another aircraft, another operator and another certification argument. Or the aircraft is flown with degraded link tolerance and more autonomy, which is a flight software change late in the programme. Each of those is more expensive than the antenna study that would have caught it, and all of them arrive after the airframe is fixed.
What sizing early actually costs
Very little, which is the frustrating part. A first-pass budget with honest assumptions can be done in a day, long before the structure exists, and its output is not a number but a constraint: this radius requires this much gain at these aspect angles, which requires an antenna of roughly this size in roughly this location with this much clearance.
Given that at concept stage, the airframe can accommodate it. Given it at integration stage, the airframe cannot, and the operations concept absorbs the difference.
Measure the pattern on the vehicle
The single highest-value test on the ground segment side is a measured radiation pattern on the actual airframe, with the actual payload fitted, rather than the antenna datasheet. It is not a difficult measurement and it does not need a chamber to be useful: a ground range and a rotating fixture will find the deep nulls, which is what you need.
What comes back is almost never what the datasheet promised, and the difference is actionable while there is still time to move something. Programmes that do this find their nulls in a car park. Programmes that do not find them in the air, at range, with a crew standing by.