[Perseids]

Why do you assume it is not base64 encoded? I have always come to believe it is some encrypted data. If I was Google I would probably give every server a UUID and a symmetric key. If an error occurs encrypt (and authenticate) the stacktrace and other debug information with that symmetric key and prepend the UUID. As a developer you can then find out the server that produced the error message, log in via SSH and decrypt the debug information.

But it would be interesting to collect a lot of this error messages to check if they appear to be completely random.

[nwh]

They change completely on every load. Encoding would likely have similar sections between page refreshes, as it doesn't we can assume your theory of encrypted data is correct. It makes sense anyway, it means they can have a user supply a stack trace or similar without exposing any information to them.

[aeon10]

Actually the first few characters remained the same with page loads.

[nwh]

CBC mode AES?

[drdaeman]

More likely a header. It's 3081 byte blob after base64 decoding. If we consider it's encrypted with some block cipher with 128 or 256-bit blocks (IV and/or MAC would be likely to be 128-bit, too), there are 9 bytes for some header and/or padding.

[nwh]

Good catch.