I still don't get it... doesn't what you say contradict the bolded part above?
If everyone prunes a certain transaction (the content) from the block, the block is still valid and nothing has to be re-mined. However, checking validity of another transactios input that uses an output of the pruned transaction cannot be done any more?
Could someone please explain?
Sorry, I wasn't thinking about the double spending limitation. I meant that you can verify that a transaction belongs in a merkle tree and a merkle tree belongs to a block, but you still need to trust that another (full) node verified that that block doesn't spend any inputs that have already been spent.
Not to derail this thread, but I too would like an explanation as to how lightweight and pruned clients handle this. I'm guessing they rely on the fact that only the bitcoin network has enough power to produce new block hashes that match the target difficulty ?