Decoding a JWT proves nothing. The payload is base64, not encryption, and anyone can rewrite it and re-encode it; the signature is the only part that establishes the token came from somewhere holding the key. Verification is therefore the step people actually want, and almost no free tool offers it, because offering it means being sent a signing secret. That is precisely why it belongs in a browser-only tool: the HMAC is computed here, the secret never leaves the machine, and there is deliberately no API endpoint that could receive one.
HMAC is implemented as RFC 2104 defines it, including the part that is easy to miss — a key longer than the hash’s block size — 64 bytes for SHA-256, 128 for SHA-384 and SHA-512 — is hashed down first, and skipping that makes every long secret silently fail to verify. The comparison is constant-time. Not because a timing attack is plausible against a page running on your own machine, but because this is the comparison people copy into their own verification code, and the fast version is the one that leaks.
Algorithms it cannot check are reported as unchecked rather than as invalid, by name. Telling somebody their signature is wrong when the truth is that you never verified it is the worst available answer. RS256, PS256 and ES256 are asymmetric and need a public key rather than a secret, so the report says which, and warns about the RS256-to-HS256 confusion attack — where an attacker re-signs a token using the public key as an HMAC secret, and a library that trusts the header’s declared algorithm accepts it.