Why would you assume that his experiences and yours are the same?
You might as well say "It's software engineering. There's no opinion involved..."
In the real world (non-defense), I did this once in my life. We went to and independent testing laboratory (actually the premier / highest-regarded one) with our device. They put it outside, pointed a big antenna at it, and it didn't pass. We made some random changes (adding shielding of some kind somewhere; the lab had it on-hand). Things ... changed. Sometimes they got better. Sometimes they got worse. We didn't know whether we were changing, or simply the radiation patterns to better match the test setup.
We kept making tweaks like that until it passed, and that was our final, independently lab-certified product design.
If we had tested at a slightly different angle, I'm pretty sure we would not have passed. Or in a different lab. It was deep voodoo. From what the lab guys said, EMC testing almost always looks like that.
I believed them. We can't really solve Maxwell's Equations in our head, and we know so little about antenna design that predicting the radiation from a complex device is not really possible. You just tweak, adding a gasket here or ferrite bead or whatever, and pray it works. Most of the small tweaks we did at the lab, but I do recall we did some larger design change (making a signal differential or something) which necessitated going to the lab a second time with a device with a new PCB.
We used stuff in defence sector I can’t talk about even today past saying they have specially designed facilities that rival the commercial sector and fairly capable high end 3D field mapping equipment that cost more than my house did. It was based on commercial kit from Agilent with their own software and hardware.
As for antennas, they’re not really that voodoo. I’ve built quite a few even up to 2.4GHz. I played with 10GHz as well but not successfully yet. Same for radiation. PCB traces are antennas and transmission lines. Impedance control is fun IMHO.
There are design patterns and crib sheets at most large companies that avoid such pitfalls though. Spinning another board is expensive so avoiding this sort of stuff is where your design engineers should be leveraged. Some stuff is indeed tweaking but that’s getting less these days with some of the CAD software around. But it costs real money.
I'll clarify my comment on the antenna end: If you do a standard pattern (a yagi, a quarter-length dipole, or whatnot), it's well-understood. What's deep voodoo are the radiation patterns from complex shapes. I've seen research projects where this is basically just done numerically -- a computer tries a bunch of shapes, computes their radiation patterns, and you get bizarre antennas.
EMC looks a lot more like the latter than the former. Our design had a PCB with several daughterboards, and a whole bunch of cables going in and out. That was voodoo. We'd add shielding. Emissions would go up. Or down. There was little rhyme or reason.
But in the end, we didn't break the bank. We went to the test facility twice, I think. We didn't have any sensible kit in-house, but we did have a scope with an FFT function, and we could make loops our of wire. We couldn't measure in real units, but we got a sense of when things went up and when they went down. And, I think, we had a healthy dose of luck.
I totally believe your equipment was superior than commercial. For defense, you want things like stealth and rad-hard. For us, we just want basic compliance.