You can’t have a network with more than one cell where no one outside of the cell knows that it exists, and without those people using a communication method to people in the cell that at least one person in the cell knows exists. And when you minimize down to the minimum contact, the network is very much non-resilient: its hard to discover and roll-up, but (especially if the cells are doing high-risk functions, which is probably why you want to do it) its also easy to disintegrate accidentally, even without being discovered and rolled-up.
But practically there's a whole bunch of problems as you say - you've no idea how many - if any - cells exist, they can't coordinate on which one picks up a particular target, the communicator is obviously at risk from being tracked down, how do you tell people where the radio broadcast is, authorities can obviously listen in once they've found the broadcast, etc.
Couldn't be done by sending messages encrypted with private key and the cells decode it with public key?
Then the problem is: How to trust public keys to have been provided by the "right" correspondent, and not someone from another gang or law enforcement?
Cell systems have order, but only where the most risk is imposed. Lesser risk, less order. Infact, too much order imposes far more risk.