What a Signed Article Actually Proves

Content provenance is full of terms that sound forbidding. Hash. Manifest. Hard binding. Claim signer. Verifiable credential. The ideas underneath are not difficult, and anyone deciding whether to buy, build, or believe a provenance system should be able to follow them.

So here is the whole stack in plain language. Each piece gets three things: what the piece is, why the piece exists, and what the piece does not tell you. That last part matters most, because most confusion in this field comes from a component being credited with work it does not do.

The problem underneath

A physical document collects history. Paper ages. Ink smudges. A letter arrives in an envelope with a postmark. None of that is proof, but all of it is friction, and friction is what forgery has to overcome.

A digital file collects nothing. A copy is identical to the original. A file edited ten seconds ago looks exactly like a file untouched for ten years. Nothing about a photograph on a screen tells you who made the photograph, what was changed, or whether the person publishing it is who they claim to be.

Provenance is the attempt to give digital content the history it does not naturally have. Every piece below exists to answer one part of that.

A hash: the fingerprint

What it is. A hash is a short code calculated from a file’s contents. Change one letter of an article, or one pixel of a photograph, and the code changes completely.

An analogy. A hash is a fingerprint of the content. Two identical files have identical fingerprints. Two files differing anywhere at all have entirely different fingerprints, with no resemblance between them.

Why it exists. Comparing two large files is slow. Comparing two fingerprints is instant. A hash gives a cheap, reliable way to ask one question: is this the same content as before?

What a hash does not tell you. A hash says nothing about who made the file or whether the contents are true. A hash only detects change. And a hash is absolute — one comma altered produces a completely different fingerprint. Absoluteness is useful for detecting tampering, and awkward for ordinary life, because quoting two paragraphs from an article produces a different fingerprint too. A hash cannot distinguish an honest excerpt from an edit.

A digital signature: the seal

What it is. A signature is made with a secret key that only one party holds. Anyone can check the signature using a matching public key that is published openly.

An analogy. A signature is a wax seal pressed with a signet ring. Only the ring makes that impression, and only one person has the ring. Anybody can recognize the seal without borrowing the ring.

Why it exists. A hash detects change but cannot be trusted on its own, because anyone can recalculate a hash after editing. A signature over the hash fixes the gap: an attacker can change the content, but cannot produce a matching seal without the key.

What a signature does not tell you. A valid signature proves that whoever holds the key signed the content. Nothing more. The signature does not establish who that key holder is, whether they are honest, or whether they are entitled to speak for the organization named in the file. Every remaining piece in this article exists to answer questions a signature cannot.

A certificate: the passport

What it is. A certificate connects a public key to a named entity. A certificate authority issues the certificate after checking that the entity is real.

An analogy. A certificate is a passport. A passport is not convincing because of the booklet. A passport is convincing because a government checked documents before issuing the booklet, and other countries have agreed to accept that government’s checks.

Why it exists. A key by itself is an anonymous number. Somebody has to connect the key to a company or a person, and somebody has to be accountable for having checked.

What a certificate does not tell you. A certificate moves the question rather than answering it. Trust now rests on the issuer. So the questions worth asking are: who issued this certificate, what did the issuer actually verify, and is the issuer independent of the company selling the provenance service? A certificate issued by a vendor to its own customers, checked by that vendor’s own staff, is a different kind of evidence from one issued by an audited certificate authority working to published rules.

C2PA: the standard vocabulary

What it is. C2PA is the industry standard for attaching provenance to media. A C2PA manifest travels inside the file itself. A manifest holds assertions, which are individual statements — a creation date, an editing action, a camera model, a copyright line — plus a signature covering them.

An analogy. A C2PA manifest is a provenance card for an artwork, sealed inside the frame rather than taped to the back where it can drift away.

Why it exists. Before a shared standard, every company invented its own metadata, and none of it survived a trip between systems. A standard means a photograph signed by one company’s tool can be read by another company’s viewer, years later, with no prior arrangement.

Why it works the way it does. C2PA is deliberately strict. A manifest contains a hard binding — a hash tying the manifest to the exact bytes of that file. Alter the file and validation fails. Strictness is the point: a provenance system that tolerated small edits would tolerate small forgeries.

What C2PA does not tell you. C2PA answers “is this file unchanged, and which key signed it.” C2PA was never designed to answer “is the signer honorable.” A perfectly valid manifest can be attached to a file by anyone who obtained a signing key, including someone whose intentions are bad. Validity and truthfulness are different properties, and a green checkmark in a verification tool means the first.

There is one more consequence of strictness worth naming. Because the binding covers exact bytes, a quotation, a crop, or a re-encode breaks it. Provenance is strongest for whole files travelling intact, and weakest for fragments, which is exactly how text tends to travel.

Ingredients: the recipe

What they are. Published work is assembled. A photograph goes into an article. A clip goes into a documentary. C2PA records those inputs as ingredients, each with its own hash and often its own manifest.

An analogy. Ingredients form a bill of materials, like the parts list for a manufactured product, or a chain of custody for evidence.

Why they exist. Almost nothing published is a single untouched capture. Without a way to record composition, provenance would only work for photographs straight out of a camera.

What ingredients do not tell you. An ingredient list records what went in. The list does not judge whether the assembly is fair or accurate. A misleading article can be built from perfectly authentic photographs.

CAWG: the people

What it is. The Creator Assertions Working Group defines how to name the people involved. A CAWG identity assertion binds a named actor — a reporter, an editor, a photographer — to the signed content, with a role attached.

An analogy. C2PA is the manufacturer’s seal on the box. CAWG is the craftsperson’s signature on the underside of the piece. Different questions, both worth answering.

Why it exists. The organization operating the signing software is often not the author. A newsroom signs, but a named reporter wrote. Attribution matters to readers, and matters more to the people being attributed.

How the design holds up. CAWG’s approach is careful. An identity assertion refers to the content by its hashes, so the identity claim and the content are bound together and cannot be separated cleanly. Each actor gets a separate assertion, which means a story with a reporter, an editor, and a fact-checker carries three distinct identity statements rather than one blurred credit.

What CAWG does not tell you. CAWG defines how to state an identity and how to check the statement’s signature. CAWG does not decide whose identity statements are worth believing. The specification says as much, plainly: the working group has not yet provided guidance on which identity issuers to consider trusted. So the mechanism is solved and the judgement is left to whoever runs the verifier.

A DID: the address you own

What it is. A decentralized identifier is an identifier that resolves to a document listing an organization’s current keys and related information. One common form, did:web, resolves at the organization’s own domain.

An analogy. A DID is a phone number you own and can update, rather than an entry in one company’s private directory. Change providers and the number still reaches you. Anyone can look it up without asking permission.

Why it exists. A certificate is issued and fixed. Organizations change keys, rotate staff, and outlive vendors. A DID gives a stable public place where the current answer lives, and gives the organization control of the entry.

What a DID does not tell you. A DID is an address, not a reputation. Anyone can create one. A DID makes an organization findable and consistent. Being findable is not the same as being reputable.

A verifiable credential: the diploma

What it is. A verifiable credential is a signed statement made by one party about another. The issuer signs it. The subject holds it. Anyone can check the issuer’s signature.

An analogy. A verifiable credential is a university diploma. The graduate holds the certificate, but the value comes from the university that issued it. An employer checks the university, not the paper.

Why it exists. Standing cannot be self-declared. A company writing its own name in a metadata field has made a claim. A certification body granting an assessed credential has produced evidence.

Where the idea lands for publishers. Journalism already has bodies doing the assessment. The Journalism Trust Initiative evaluates news organizations against a published standard covering ownership transparency, editorial independence, corrections practice, and sourcing. Carrying that assessment as a credential bound to the newsroom’s DID turns institutional standing into something a machine can check and a person can look up.

What a credential does not tell you. A credential is worth exactly what the issuer is worth. Reading the issuer’s criteria is the real work, and a credential’s value is properly a matter of judgement rather than mathematics.

Putting the pieces together

Stack all of it and a signed article can support five separate statements. Each one comes from a different part of the machinery.

Each mechanism has a different failure mode. Anyone evaluating a provenance system should be asking which of the five that system actually delivers, because several products deliver the first two and describe the result as though all five were present.

The honest limit

One thing none of it does, and the limit deserves to be stated clearly rather than buried.

Provenance says nothing about whether an article is true. A perfectly signed, fully attributed, accredited-publisher article can be wrong. Provenance answers questions about origin and integrity: where did this come from, has it changed, who stands behind it. Accuracy is a separate matter, decided by editors and evidence.

That limit is not a weakness. Knowing reliably who is speaking is the precondition for holding anyone accountable for what was said. Provenance makes accountability possible. Provenance does not deliver accountability by itself.

Three questions worth asking

For anyone evaluating a provenance product without wanting to become an expert, three questions separate the architectures quickly.

  1. Who holds the signing key — your organization, or the vendor?
  2. Who issued the certificate, and are they independent of the vendor selling you the service?
  3. Who vouches for your organization’s standing, and can a reader read that party’s criteria?

Answers to those three describe where trust actually rests. Everything else is detail.