Base64url encoded representation puzzle - authentication

Question

I'm writing a cookie authentication library that replicates that of an existing system. I'm able to create authentication tokens that work. However testing with a token with known value, created by the existing system, I encountered the following puzzle.
The original encoded string purports to be base64url encoded. And, in fact, using any of several base64url code modules and online tools, the decoded value is the expected result.
However base64url encoding the decoded value (again using any of several tools) doesn't reproduce the original string. Both encoded strings decode to the expected results, so apparently both representations are valid.
How? What's the difference?
How can I replicate the original encoded results?
original encoded string: YWRtaW46NTVGRDZDRUE6vtRbQoEXD9O6R4MYd8ro2o6Rzrc
my base64url decode: admin:55FD6CEA:[encrypted hash]
Encoding doesn't match original but the decoded strings match.
my base64url encode: YWRtaW46NTVGRDZDRUE677-977-9W0Lvv70XD9O6R--_vRh377-977-92o7vv73Otw
my base64url decode: admin:55FD6CEA:[encrypted hash]
(Sorry, SSE won't let me show the unicode representation of the hash. I assure you, they do match.)

Answer 1

This string:
YWRtaW46NTVGRDZDRUE6vtRbQoEXD9O6R4MYd8ro2o6Rzrc
is not exactly valid Base64. Valid Base64 consists in a sequence of characters among uppercase letters, lowercase letters, digits, '/' and '+'; it must also have a length which is a multiple of 4; 1 or 2 final '=' signs may appear as padding so that the length is indeed a multiple of 4. This string contains only Base64-valid characters, but only 47 of them, and 47 is not a multiple of 4. With an extra '=' sign at the end, this becomes valid Base64.
That string:
YWRtaW46NTVGRDZDRUE677-977-9W0Lvv70XD9O6R--_vRh377-977-92o7vv73Otw
is not valid Base64. It contains several '-' and one '_' sign, neither of which should appear in a Base64 string. If some tool is decoding that string into the "same" result as the previous string, then the tool is not implementing Base64 at all, but something else (and weird).
I suppose that your strings got garbled at some point through some copy&paste mishap, maybe related to a bad interpretation of bytes as characters. This is the important point: bytes are NOT characters.
It so happens that, traditionally, in older times, computers got on the habit of using so-called "code pages" which were direct mappings of characters onto bytes, with each character being encoded as exactly one byte. Thus came into existence some tools (such as Windows' notepad.exe) that purport to do the inverse, i.e. show the contents of a file (nominally, some bytes) as they character counterparts. This, however, fails when the bytes are not "printable characters" (while a code page such as "Windows-1252" maps each character to a byte value, there can be byte values that are not the mapping of a printable character). This also began to fail even more when people finally realized that there were only 256 possible byte values, and a lot more possible characters, especially when considering Chinese.
Unicode is an evolving standard that maps characters to code units (i.e. numbers), with a bit more than 100000 currently defined. Then some encoding rules (there are several of them, the most frequent being UTF-8) encode the characters into bytes. Crucially, one character can be encoded over several bytes.
In any case, a hash value (or whatever you call an "encrypted hash", which is probably a confusion, because hashing and encrypting are two distinct things) is a sequence of bytes, not characters, and thus is never guaranteed to be the encoding of a sequence of characters in any code page.
Armed with this knowledge, you may try to put some order into your strings and your question.
Edit: thanks to #marfarma for pointing out the URL-safe Base64 encoding where the '+' and '/' characters are replaced by '-' and '_'. This makes the situation clearer. When adding the needed '=' signs, the first string then decodes to:
00000000 61 64 6d 69 6e 3a 35 35 46 44 36 43 45 41 3a be |admin:55FD6CEA:.|
00000010 d4 5b 42 81 17 0f d3 ba 47 83 18 77 ca e8 da 8e |.[B.....G..w....|
00000020 91 ce b7 |...|
while the second becomes:
00000000 61 64 6d 69 6e 3a 35 35 46 44 36 43 45 41 3a ef |admin:55FD6CEA:.|
00000010 bf bd ef bf bd 5b 42 ef bf bd 17 0f d3 ba 47 ef |.....[B.......G.|
00000020 bf bd 18 77 ef bf bd ef bf bd da 8e ef bf bd ce |...w............|
00000030 b7 |.|
We now see what happened: the first string was decoded to bytes but someone fed these bytes to some display system or editors that really expected UTF-8. Some of these bytes were not valid UTF-8 encoding of anything, so they were replaced with the Unicode code point U+FEFF ZERO WIDTH NO-BREAK SPACE, i.e. a space character with no width (thus, nothingness on the screen). The characters where then reencoded as UTF-8, each U+FEFF yielding the EF BF BD sequence of three bytes.
Therefore, the hash value was badly mangled, but the bytes that were altered show up as nothing when interpreted (wrongly) as characters, and what was put in their place also shows up as nothing. Hence no visible difference on the screen.