“W3C-VC or SD-JWT?” is usually asked as if it were a choice between two flavours of the same object, like picking a file format. It is not. Each name designates a complete stack: a data model and a proof mechanism that answer two independent design questions, and they answer both of them in opposite ways. This page sets out those two questions, the map they produce, and why the shape of that map matters for institutions issuing credentials meant to last for decades. It follows the reasoning of the VC guide v1.1 (EUNIS SIG, July 2026), Verifiable Credentials in Higher Education: W3C-VC and SD-JWT VC, published by the EUNIS Special Interest Group – Mobility & Digital Credentials (EUNIS SIG).
Data model and proof mechanism are separate things
Every verifiable credential combines two ingredients. The data model defines how the content is structured and, above all, where its meaning lives: is “EQF level 7” a bare string a human has to interpret, or a reference to a governed European concept that any machine can resolve? The proof mechanism is how it is secured: the scope of the issuer’s signature, the way the holder shows the credential belongs to them, the selective disclosure method, and whether two presentations can be linked.
In principle the two are independent, and that independence is not a technicality. It determines whether, in 2040, a university can move to new cryptography without having to reissue its entire archive of awarded credentials. When a specification welds data model and proof into one object, that flexibility is gone by construction. Two product sheets that both say “supports selective disclosure” can hide exactly this difference.
The two questions
Question one: where does meaning live?
Inside the credential. Every field points to a shared, governed definition, and that pointer travels in the signed payload itself, through JSON-LD and its @context. Data and meaning are signed together, so a verifier elsewhere in Europe, in another sector or thirty years later lands on the same interpretation as the issuer intended.
Outside the credential. The payload holds plain key-value pairs such as "eqf_level": 7. Which framework that number belongs to, and what the key signifies, is written down somewhere else: specifications, verifier software, agreements between parties who coordinated beforehand. What the issuer signs is the value as text.
Both answers are reasonable. Flat claims are simpler and perfectly adequate inside an ecosystem whose members have already agreed on semantics. Embedded meaning costs more at the start and earns its keep when issuer and verifier are strangers, which is exactly the situation in cross-border recognition.
Question two: is the proof a layer or the format itself?
A detachable layer. The credential is a structured object, and the proof is calculated over a canonical serialisation of it and then attached. The same structure can be secured by different proof suites, from ordinary signatures to BBS, with which the holder produces a new, unlinkable proof at each presentation. Since the proof sits on top, it can be upgraded, replaced or moved to post-quantum algorithms while the credential’s design stays as it is.
Welded to the format. SD-JWT is one particular construction: a JWT in which the claims that can be disclosed selectively are hidden behind salted hashes, and the issuer’s signature binds everything together. It does its job well, but disclosure mechanism, signature and data structure are one object, with no seam where the cryptography could be exchanged and the credential kept.
One map, four cells
Placing the two questions on two axes gives a simple map of the design space.
| Proof as a detachable layer | Proof welded to the format | |
|---|---|---|
Meaning inside the credential (JSON-LD graph, @context) | The W3C-VC stack. VCDM 2.0 secured with Data Integrity proofs, including BBS (bbs-2023) where unlinkability is required. | The hybrid cell. VCDM 2.0 enveloped and secured with SD-JWT via VC-JOSE-COSE. Real, standardised and useful as a transition path, though whether national wallets will accept it is not settled by the current acts. |
| Meaning outside the credential (flat claims) | Essentially unpopulated. Little incentive: anyone who accepts the cost of a detachable proof layer tends to take the semantic layer too. | The SD-JWT stack. IETF SD-JWT VC: flat claims, salted-hash disclosure, semantics out of band. |
Technical note: the specification names
- Semantic axis: JSON-LD 1.1 with
@context(W3C-VCDM, VCDM 2.0) versus flat JSON claims typed byvctwith Type Metadata (IETF SD-JWT VC draft). - Assurance axis: W3C Data Integrity with cryptosuites such as ecdsa-rdfc-2019, eddsa-rdfc-2022 or bbs-2023, or enveloping via VC-JOSE-COSE, versus SD-JWT salted-hash disclosure (RFC 9901) as the sole mechanism.
- Hybrid cell: specified in the W3C Recommendation “Securing Verifiable Credentials using JOSE and COSE” (VC-JOSE-COSE), which includes an SD-JWT option.
A frequent confusion: an SD-JWT credential is not a W3C-VC
An SD-JWT credential is not a W3C-VC that happens to use another signature. Its data model, SD-JWT VC (IETF), is its own: flat JSON claims typed by vct, with no JSON-LD and no @context inside what is signed. The specification calls the result a “Verifiable Digital Credential” precisely to distinguish it from the W3C Verifiable Credentials Data Model.
The confusion comes from the name. SD-JWT is also a mechanism (RFC 9901), and the W3C-VCDM can be secured with it: that is the hybrid cell above, specified in VC-JOSE-COSE §3.2.1. The two objects even carry different media types: application/vc+sd-jwt for a W3C-VC secured with SD-JWT, and application/dc+sd-jwt for SD-JWT VC. The IETF draft adopted the second in November 2024 to avoid the clash with the W3C registration.
ETSI TS 119 472-1 (V1.2.1, February 2026), which profiles electronic attestations of attributes for eIDAS 2.0, keeps them apart. Clause 5 covers EAAs based on SD-JWT VC. Clause 7 covers EAAs based on JSON-LD W3C-VC, which must carry @context, and within it clause 7.6.4.3 covers JSON-LD W3C-VC secured with SD-JWT. The implementing acts currently apply only clauses 2 to 6 of that specification, so clause 7 sits outside their binding annexes for now. The EUDI Wallet ARF (v3.0.0) also lists SD-JWT VC and W3C-VCDM 2.0 as separate formats. Same mechanism, two different objects.
The hybrid cell runs in one direction only
The map is lopsided, and that is its most important structural fact. The W3C data model can travel inside the SD-JWT mechanism. An institution can issue VCDM 2.0 credentials now, enveloped in SD-JWT under VC-JOSE-COSE, and switch the same data model to a BBS proof once unlinkability becomes an operational requirement. That envelope is an option the W3C specifies; it is not the IETF SD-JWT VC format named in the EUDI Wallet implementing acts, and whether national wallets will accept it is not settled by the current acts. The meaning, the structure and the work invested in semantics carry over the cryptographic change. Re-securing still requires getting the new proof to each holder; what survives is the credential’s identity, content and meaning, not the need for a delivery channel.
Nothing equivalent exists in the other direction. An SD-JWT credential has no separate proof layer to replace, because in that stack the format and the proof mechanism are one and the same. If the cryptography must change, whether because an algorithm is broken, post-quantum migration arrives or unlinkability is demanded, the credential cannot be carried across: every holder receives a new one, in a new format. One stack has a seam; the other does not. (Key rotation alone, without an algorithm being deprecated, forces reissuance on neither stack as long as historic public keys remain resolvable.)
Why the false symmetry misleads
Treating the two as interchangeable flavours invites flavour criteria: this year’s wallet support, the easier pilot, the smoother vendor demo. Such criteria are fair game for a two-year pilot. They are close to irrelevant for a credential like a degree, which must stay verifiable for thirty years or more.
Thinking in stacks replaces those with better questions. The question is not which format wallets currently prefer. It is where meaning has to live when the people checking our credentials are institutions we have never dealt with, and whether we want a seam between credentials and cryptography when both will outlast all the systems we operate today. Honest institutions may reach different answers. None should choose without asking.
Internalise or externalise
Across every dimension, the differences reduce to one fact: W3C-VC carries meaning, validation, privacy, suspension and audit inside the credential, while SD-JWT leaves them to procedures and agreements, which is cheaper and faster in a coordinated ecosystem and costlier whenever coordination fails. W3C-VC vs SD-JWT sets out the full comparison row by row.
For the detail of each stack, see The W3C-VC stack and The SD-JWT stack. For how to decide which credentials go where, see Choosing a stack.
Key points
- W3C-VC and SD-JWT are two stacks, not two flavours of one format.
- Every architecture answers two questions: where meaning lives, and whether the proof is a layer or the format itself. The two stacks answer both in opposite ways.
- The hybrid cell (VC-JOSE-COSE) lets the W3C data model borrow SD-JWT’s mechanism as a transition; SD-JWT cannot borrow anyone else’s.
- Pilot criteria (wallet support this year) are not thirty-year criteria (where meaning lives, whether a seam exists).
- Internalising costs more today; externalising costs more whenever coordination fails, and cross-border education is the textbook uncoordinated ecosystem.
