Choosing an ACDC section: attribute (a) vs aggregate (A)¶

A distilled, reusable design principle for authoring ACDC schemas in this repo. It generalizes a decision made for the SEDI credential family; the repo-specific record is in this.i (@sdav5t), and the discussion that produced it is WebOfTrust/keripy#1505.

Derived doc. Per the repo methodology, this.i is the source of truth for decisions; this page distills the reusable principle so future authors (human or AI) don't re-derive it.

The two graduated-disclosure mechanisms¶

Both let a holder reveal some fields and withhold others, each withheld field blinded by a per-block nonce so its value can't be brute-forced from its SAID. They differ in what the top-level blocks look like:

• Attribute section (a) — a labeled field map. Each attribute is a nested block under a meaningful key (a.dob, a.residenceState). Partial disclosure: reveal a nested block, others collapse to their SAID. Field labels are public (they're in the schema).
• Aggregate section (A) — an ordered array of unlabeled blinded blocks, committed by a zeroth-element AGID. Selective disclosure: reveal chosen array elements, the rest travel as bare SAIDs. The block labels live inside each block, so they're blinded; array position is arbitrary.

The deciding question¶

Does hiding the field labels or positions buy you anything?

The only thing the aggregate provides over the attribute section is that its blocks carry no outward labels, so their number and position can be arranged so as not to leak information. That is valuable in exactly two situations:

1. The presence or identity of a field is itself sensitive — e.g. "does this credential even carry attribute X?" is something you must not leak.
2. The field set is variable or homogeneous — the fields are interchangeable members of a set (an indexed series, a bag of flags), so a meaningful label per position would be noise, and position could otherwise leak which members are present.

If neither holds, the aggregate is pure cost, because you lose label-based pathing. Graduated- disclosure negotiation (apply/offer path lists, e.g. via keri's Pather) wants to name a field as a/residenceState. In an aggregate you can't — you'd need a label→index reverse-lookup table, and if you put the index in the path you defeat the unlinkability the array was for. For the most-used credentials, that complexity is not worth paying.

The rule of thumb¶

The worked case (SEDI)¶

• sedi-id — the core identity credential: name, birth date, image, residence components. A fixed, heterogeneous, labeled set → attribute a, each attribute a nested partially-disclosable block. Bonus: as a plain v1 ACDC it version-stamps and SAIDs cleanly under this repo's pinned oracle.
• sedi-age — age-threshold flags ageOver13/16/18/21/55/65. A homogeneous boolean vector you disclose a subset of → aggregate A, matching the ISO mDL age_over_NN element.

Caveats worth remembering¶

• For a homogeneous vector where all members are always present, an attribute form wouldn't leak much either; the aggregate's real wins are parity with mDL, natural fit, and forward-compatibility (position-randomization/decoys and the sparse-Merkle upgrade). Don't over-claim a correlation delta.
• Attribute-form partial disclosure forces the issuee (a.i) and the a-section metadata to be revealed on any disclosure; an aggregate can disclose one element with no issuee. Usually the holder binding is wanted anyway, but note it when unbound disclosure matters.
• Under this repo's pinned keri 1.2.13 oracle: a v2 aggregate credential can't be version- stamped (examples are unversioned; the AGID is computed by the v2 Aggor, not the linter), and nested blocks aren't compacted (a partial-disclosure form's a.d/d differ from the full form's). See this.i @sd2qfw. This is an oracle limitation, not a spec one.