The article I read indicates China used a sim which means they have to guess at what the RCS of the b21 is. Further they shot down the simulated b21 with a hypersonic missile after they pretended to target something else then veered towards the b21, claiming the pilot wouldn't take evasive action based on it's initial trajectory.
So if China knows the RCS of the b21, the performance, and the pilot doesn't react to a missile targeting the b21, they can shoot it down.
In a simulation where the only radar reflector is one simulated jet, and there is no real hardware, it's pretty easy to hit since the Pfa is effectively zero. In the real world, there's a thousand different things which will give some similar energy return and there's no simple way to filter that out without actually seeing the thing in real life.
Stealth isn't about zeroing the energy, it's about pushing it into an ambiguous state space where it cannot be distinguished from birds and shit.
Doppler filtering will do alot to eliminate slow moving clutter sources but I have my doubts they can detect a 0.1 sq meter return much further than 20km.
Stealth uses many techniques to lower the return, shaping to stop direct returns to the source, mag ram (magnetic radar absorbing material) to absorb some of the energy and stop traveling waves at high frequency, and resistive cards or similar methods for low frequency which has sections of different resistivity that cause mini defractions instead of one large one.