Interesting, some I've some googling to do!
Modern receivers use quadrature sampling detectors rather than traditional superheterodyne. In that setup any leakage would be on the same frequency and harder to detect.
[0] http://www.kr1st.com/swlloop.htm
[1] https://www.fmuser.org/fm-receiver/receiver-antenna/DE31MS-l...
More importantly, there's a lot of background noise on HF bands that will mask that weak signal.
If you've never built a radio and tried to shield this unintended export than I can totally see how you might think this is just a matter of careful design and more shielding but it really is a lot harder than that and you will simply never reach zero to the point where even an ordinary spectrum analyzer hooked up to the input of your radio will not show the oscillator frequency as a nice fat peak.
I learned a lot during that project, especially how hard it is to make an oscillator that does not radiate. So, it got to the point where I could reliably detect the receiver from about 100 meters away, fortunately the counterparty never started out from the assumption that it would be in that particular location to begin with. Trawling for a signal is a lot harder than verifying that is is there. But if you know the modulation and the frequency the receiver uses for its mixing stage this is a very hard problem to solve in such a way that there is absolutely no power radiated out of the reception antenna. Any kind of magnetic or capacitive coupler is bi-directional. Maybe with today's hardware capabilities it would be possible to pull the whole thing into the digital domain at a very early stage and that way I can see a few options to make it 100% clean but in the analog domain I do not see a bullet proof way of achieving this.
Number stations on short waves all use AM, so you know the modulation. But you don't need to know it, superhet works the same way with any modulation. You need to know the number station frequency, receiver's intermediate frequency, and guess whether its above or below.
> in such a way that there is absolutely no power radiated out of the reception antenna.
I'm not saying there is absolutely no power radiated out of the reception antenna, only that there is not enough power to reliably detect and localize, given the noise and interference from other sources.
If you want absolutely no power radiated out of the reception antenna, you can still do it. Feed some local oscillator frequency, inverted, into the antenna to cancel the remaining leak. But as far as I know, nobody bothers since some leakage is not a problem.
> Any kind of magnetic or capacitive coupler is bi-directional.
True, but in many designs there's also at least one transistor stage in the preamp, and that is not bi-di. There is some stray capacitance between collector and base, but not much.
> Maybe with today's hardware capabilities it would be possible to pull the whole thing into the digital domain at a very early stage
It is possible, but unnecessary. The last radio I built has quadrature sampling detector with FST3253 and handful of op-amps. Most SDRs also do I/Q sampling with two slow ADCs, much simpler than a single high-speed ADC.