A signature proves two things and stops

A signed message gets treated as a settled matter. It settles less than people think, and the boundary is sharp enough to state exactly.

The list of what it does not prove

• Who the key holder is. A key is a key, not a person.
• That the key belongs to who you were told it belongs to.
• That the statement in the message is true.
• When it was signed, unless the text says so, and the text can say anything.
• That the signer still controls the key today.
• That the signer meant it, wrote it freely, or is still alive.

Every one of those is a thing people have inferred from a valid signature at some point. None of them follow.

Why the boundary is exact

Signing is a mathematical operation over the message bytes using the private key. Verification checks the result against the public key and the same bytes. The maths knows nothing about names, dates or intent, because none of that was an input.

Everything beyond the two facts comes from context, and context is not signed. The name attached to a key was typed by a person. The trust in the key came from somewhere else.

The mistake that matters most

A signature verifying against a key you just downloaded proves that the key and the message go together. That is close to worthless on its own. Anyone can generate a keypair, sign a message about an awazon market address, and publish both. It verifies perfectly and means nothing.

The whole weight rests on where the public key came from and whether you have seen it before in a place you had reason to trust. That question is not cryptographic and no tool answers it.

The replay problem

A message signed a year ago verifies exactly as well today. If it says one address is current, it says that about a moment you cannot locate. Unless the signed text contains a date, and unless you have a reason to believe that date, a valid signature is timeless in a way that is unhelpful.

Which is why an old signed announcement is not evidence about today. See the live status page for the same problem in a different costume.

What it is genuinely good for

Continuity. If you have a key from earlier and a new message verifies against it, you have learned that the same key holder wrote both. That is a narrow fact and a real one, and it is the whole basis for treating an announcement as coming from the same source as the last.

It also makes tampering visible. A signed text that has been edited by one character fails verification loudly. Anyone passing along a modified version has to either strip the signature or accept that it breaks, and both are noticeable if you check.

What would move this

Nothing, without changing what signing means. Extra properties can be layered on top, such as timestamps from a third party or transparency logs that make a key's history public. Those add facts. They do not widen what the signature itself proves, they sit beside it.

• Signatures, plainly
• Whether you need PGP for a given step
• A screenshot as proof