To my understanding the merkle-tree can be used to separate the actual transactions (or other content) of a block and its validation process, by recursively applying hash functions to the blocks content.
This makes it possible to implement simplified payment validation (or the validation of any other data within a block).
The linked definition states that SVP is a
[...] method for verifying if particular transactions are included in a block without downloading the entire block.
If the merkle tree is a method to check if a specific transaction was part of a block, this implies that we do not "trust the block" and are thus checking if the hash of the searched transaction is part of the tree. If we do not trust the block (or the original validator of the block) how come we trust the merkle tree, which could also be faked (or could it not?).
Are we "just" relying on the fact that has values are pseudo random and that it would be hard to create a fake tree with the same root-value, but one different node value?
Thanks!