If MD5 is broken, why does it still exist?
Because collision resistance and error detection are different jobs. MD5 is still fast and still perfectly good at spotting an accidentally flipped bit, a truncated download, or a duplicate file, which is why it survives in ETags, cache keys, and file manifests. What it cannot do is stop a determined attacker, who can craft two different inputs with the same digest, so it must never guard a password, a signature, or a certificate.
Why does my digest not match the one published on a download page?
Almost always an encoding or whitespace difference. This tool hashes the exact characters in the box as UTF-8, so a trailing newline that your editor added, a CRLF instead of an LF, or a leading space produces a completely different hash. It also hashes text, not files, so hashing the text of a file's name or a pasted excerpt will never match the file's own checksum.
Can I get the original text back from the hash?
No, and no tool can. MD5 is a one-way compression: it maps any input, of any length, to 128 bits, so infinitely many inputs share each digest. What the so-called reverse-MD5 sites actually do is look your hash up in a precomputed table of common strings. That works for "password" and fails for anything with real entropy.
Does uppercase or lowercase output matter?
No, it is purely cosmetic. The digest is the same 128 bits either way; the case toggle only changes how those bytes are written out as hex. Some systems print hashes uppercase (Windows CertUtil) and some lowercase (md5sum), so the option exists to match whatever you are comparing against.
What should I use instead for anything security-related?
For integrity and signatures, SHA-256. For passwords, never a plain hash of any kind: use bcrypt, scrypt, or Argon2, which are deliberately slow and salted so that a stolen database cannot be brute-forced at billions of guesses per second. A fast hash like MD5 is exactly the wrong property for password storage.