This is a Huffman code compression tool. Any text in the input will be profiled for generating an optimum variable-length prefix-free code, then compressed into bytes using that code.
The compression statistics include the description of the Huffman tree itself.
The tree here is represented as a preorder traversal, with the internal node
symbol defaulting to
~a~bc stands for encoding a, b, c as 0, 10,
and 11 respectively). This encoding is
2N-1 characters for an alphabet of size N,
as there are always
N-1 internal nodes in the tree.
The first four bytes of the output are a length header, which ensures that overhang bits are discarded while decoding.