Why is the encoded data signed, not the raw data itself?

[nschloe]

In license files and offline licenses, the base64-encoded data is signed. Why is that so? Couldn't we sign the raw data (JSON strings) directly?

[ezekg]

For license files, having to decode the enc dataset before verifying the data's signature adds a needless step when enc data is invalid (e.g. tampered with), and it adds one more step where somebody could unintentionally use data from inside of the encoded dataset before verifying its authenticity, which is dangerous. The enc dataset should always be cryptographically verified before it's decoded, and our order of operations asserts that. And keep in mind that the enc property isn't always a raw JSON string — it could be an encrypted JSON string as well (i.e. a random series of bytes until successfully decrypted). In addition, we utilize a prefix during verification, and having the data be base64 encoded allows the prefixed string to be easily created — trying to concatenate a license/ prefix to a random byte array is more challenging and error-prone, relatively, than simply concatenating 2 strings (see below for further explanation).

For signed license keys, the embedded dataset is signed after being encoded and prefixed with key/. This allows us to differentiate between different types of signed datum. For example, if we didn't have a prefix, then using e.g. another signature, forged into a similar structure of a license key would pass verification (e.g. $DATA.$SIGNATURE), even though it's not a valid license key. Encoding the data and combining it with the prefix allows us to sign the embedded data along with the prefix, protecting the scheme from forgery.

tl;dr: Both of these order of operations protect against forgery and tampering, while trying to prevent unsafe usage.