Copyright © 2026 SumSet Tech LLC and the Universal Manifest contributors. This specification is available under the Apache License, Version 2.0. It is free and remains free. Earlier published revisions retain their original licence notices.
v0.5.0. Published as tag
v0.5.0ofUniversalManifest/universal-manifest, with the same wire files as release candidate 1 (v0.5.0-rc.1). Wire version0.5; namespacehttps://universalmanifest.net/ns/v0.5; reading routes/spec/v0.5/. It is a mechanism release, not a completed implementation or conformance qualification. Feature-specific PREVIEW and conditional requirements remain in force.
Copyright (c) 2026 SumSet Tech LLC and the Universal Manifest contributors. Grig Bilham — inventor, architect, and author of the Universal Manifest. This specification is available under the Apache License, Version 2.0. Earlier published revisions retain their original licence notices. Earlier versions were reviewed in the OMA3 Technical Working Group, and publication asserts no OMA3 ratification. This specification does not settle standards IPR or patent adoption.
Universal Manifest is an open specification for portable state manifests. A manifest is not static metadata: v0.5 pairs the manifest with a six-stage evaluation that checks signature, freshness, projection, and consent, then emits a structured receipt of what was actually verified. Current permission additionally requires the authority and target checks described in the Base; it is not implied by an envelope signature. The specification is independent of programming language, framework, or runtime: an abstract data model is written to bytes through a production rule. JSON-LD is the reference encoding; CBOR-LD is a separate profile. The public specification package supplies the contract and fixtures; separately maintained reference implementations provide bounded evidence, not a required implementation dependency.
v0.5.0 is a new-version contract derived from the accepted v0.4
editorial set, published as tag v0.5.0 with the same wire
files as release candidate 1 (v0.5.0-rc.1). The immutable
revision of UniversalManifest/universal-manifest
identified by a specification release is normative; main is the working
copy. This complete first-class source set lives in
spec/v0.5/. The /spec/v0.5/ website is its
generated reading view. /spec/v0.5/ is always this
version's reading view; the maintainers set where
/spec/latest/ points.
The v0.5 document set is organized as a conformance-tier stack. The question that decides what you read and build is "what level of assurance am I certifying to?" A breadth-first Primer orients you; the Base specifies everything required for a Tier-0 / Tier-1 evaluator and holder; four companion profiles add depth by tier or by optional feature; and the Cookbook collects worked examples.
| Document | Role | Status |
|---|---|---|
| Primer | Breadth-first overview: what UM is, the shape of a manifest, the lifecycle, the core guarantees, a one-glance map of tiers and extensions. No normative requirements. | Informative |
| Base (Tier-0 / Tier-1 Core) | Everything required to build a conformant Tier-0 / Tier-1 evaluator and holder. Names advanced and optional features and points to the extensions for their mechanics. | Production-candidate (some members PREVIEW) |
| EXT-T1 (Tier-1 Binding Profile) | Tier-1 binding mechanics: holder binding, presentation proof, liveness attestation, the Stage-2 sub-steps, the end-to-end claim-proof path, v0.3 backwards compatibility. | Normative (Sections T1.3.1–T1.3.3 PREVIEW) |
| EXT-T2 (Tier-2 Cryptographic-Binding Profile) | Tier-2 ZKP proof profiles, the BBS+ selective-disclosure track, the post-quantum migration path. | PREVIEW |
| EXT-T3 (Tier-3 Ceremony Profile) | Tier-3 multi-party ceremony: model, signer-role taxonomy, threshold-protocol guidance. | PREVIEW |
| EXT-OPT (Optional-Feature Profiles) | Tier-orthogonal optional features: the abstract data model and its CBOR-LD encoding, the statusRef resolution protocol, receipts-as-a-class with hash-chaining, the bilateral session model, federation, profile registration, the agent-delegation scope registry, trustWeight and category trust, extended schemas, the cryptographic-requirements summary, and the design-stage candidate profiles (current policy-state resolution, represented-person consent, referral-route attribution, GeoPose primary pose values). | Mixed (PREVIEW per part) |
| Cookbook | Worked examples, organized by scenario and tagged by tier. | Informative |
The following features are carried from the accepted v0.4 editorial candidate. The migration and release-candidate notes state the actual changes from the published v0.4 snapshot; this list does not claim every feature was published or implemented.
https://universalmanifest.net/ns/v0.5; the
exact current source body is schema.jsonld in this
directory, reproduced verbatim with its content hash in Base Appendix B.
The current public and local-candidate snapshots are distinguished under
Namespace URIs below. Evaluators MUST pin the content-hashed context and
verify its multihash (the expected value is under Namespace URIs below
and in Base Appendix B) before trusting term definitions, rather than
fetching an arbitrary remote context at evaluation time. The
um: term prefix is version-independent and expands to
https://universalmanifest.net/ns/um#. See Base Section 6.10
and Appendix B.sd-jwt-kb,
bbs-holder-commitment, reciprocal-control).
Presentation proof (presentationProof)
provides proof-of-possession that binds a manifest to a specific
verifier and moment, preventing replay. Liveness
attestation (livenessAttestation) proves human
presence at its attestedAt time but not at verification
time. See Base Section 6.6 and EXT-T1.requiredLiveness floor
(minFreshness plus optional userVerified) and,
for locked-tier portable-unlock profiles, a
requiredAssuranceClass floor (software <
hardware-uv < hardware-bound). These floors
compose with requiredTrustTier, can only raise the bar, and
fail closed with receipt evidence such as
assuranceInsufficient / assuranceStatus.
PREVIEW. See Base Sections 2.1, 3.3.1, and EXT-T1 Section T1.3.ZkLinkedSecretProof,
ZkHdDerivationProof) for proving control of multiple DIDs
in zero knowledge, plus the BBS+ selective-disclosure track and a
post-quantum migration path. PREVIEW. See EXT-T2 and Base Section
6.4.7.ceremonyProof
(ThresholdAttestationProof) carrying a
threshold, attesters, ceremonyId,
and an aggregateProof for co-signed, highest-stakes
deployments. PREVIEW. See EXT-T3 and Base Section 6.4.8.keyRefResolution, per-consent consentStatuses,
per-claim claimStatuses (with claimRef,
status, and optional tier),
unprocessedEntries, the credential-binding status fields
(holderBindingStatus, presentationProofStatus,
livenessStatus, crossDidBindingStatus), and
effectiveTrustTier. The facetStatuses status
set adds assuranceInsufficient and written
(both PREVIEW-driven) to v0.3's values. See Base Section 3.3.facet-key-rotated,
facet-key-shredded), and unlock-window audit fields for
time-boxed access. PREVIEW where marked in the source profiles. See Base
Sections 2.3.5, 3.1.4, 7.2 and EXT-OPT Sections O3 and O4.5.manifestId, status in active /
revoked / suspended, updatedAt,
optional reason / cursor /
nextCheck) and the status-endpoint conformance criteria.
See Base Section 3.4 and EXT-OPT Section O2.exchangeId
and evaluatorId on receipts) and the bilateral-exchange
security model. EXT-OPT specifies the full session object
(um:BilateralSession with sessionId,
exchangeId, two participants,
initiatedAt, expiresAt, and strictly-forward
lifecycle state), paired-receipt correlation, receipt
hash-chaining (seq / prevHash), federation,
and manifest forwarding across resolver operators. See Base Section 3.5
and Section 6.4.6; EXT-OPT Sections O3 and O4.trustWeight, an actorState member, a
two-component devices schema, and a derived-variant
sensor-consent vocabulary, each independently optional. See EXT-OPT
Sections O6 through O9.algorithm,
canonicalization) pair, the algorithm
enumeration in schema.json is the current materialization
of the profile registry (EXT-OPT Section O6, signature
category), and any number of further suites are recognized by that same
registration step rather than by re-architecture. The first additive
suite lands with it: Signature Profile B
(ES256 — ECDSA over NIST P-256 with SHA-256;
JCS-RFC8785 canonicalization and a signing input both
identical to Profile A; signature.value is the base64url of
the raw 64-byte R || S, with DER rejected; and deliberately
no low-S rule and no RFC 6979 requirement, so a Secure
Enclave, StrongBox, or TPM key is conformant as it signs). Profile B
exists for hardware key custody: those chips hold P-256 keys and only
P-256 keys, so without it the hardware-bound assurance rung
is unreachable on the hardware people carry. Signature Profile A
(Ed25519) remains the default and the mandatory-to-implement baseline
for every evaluator; Profile B is optional-additive for evaluators; an
unrecognized (algorithm, canonicalization)
pair is rejected fail-closed and never reinterpreted. Profile B itself
adds no context term. The v0.5 context reuses the accepted local
editorial definitions, while the published v0.4 context has different
bytes and remains immutable; see Migration below. See Base Sections
1.5.1 and 1.6; EXT-OPT Sections O6, O10, O11, and O15.um:reason:policy:* codes — is a consent/authorization
record still current before a governed or agentic action proceeds?),
asset-bound represented-person consent (the consent
state of the person depicted, scanned, performed, or replicated
inside a portable asset, as distinct from asset/token
ownership, with carrier binding and currency resolution), signed
referral-route attribution (verifiable route provenance,
route-scoped disclosure via consent conditions, and minimal attribution
receipts — with ad-network, marketplace, payment-rail, and
buyer-data-sharing behavior explicitly out of UM scope), and
GeoPose primary pose values (an OGC GeoPose payload or
pointer as the primary position-plus-orientation value for an asset,
object, dataset record, or scene element, with UM governing signature,
provenance, consent/current-state policy, projection, and receipts
around that value). These are design drafts: their members are not in
the published schema/context or fixture suite and carry no conformance
tests until finalized. See EXT-OPT Sections O2.6, O12, O13, and
O14.Signature Profile A (JCS + Ed25519) remains the default and mandatory
evaluator baseline, but its acceptance predicate is explicitly tightened
for wire 0.5 by Base §1.6.1.1. Its
discriminator pair, SPKI carriage and signing-input procedure remain;
Profile B is additive and its raw R || S/high-S behavior is
unchanged. The six-stage evaluation sequence (Arrive, Verify, Project,
Consent, Compose, Receipt) remains the conceptual spine, with
binding-verification sub-steps in EXT-T1.
Use the specification and fixtures as your source of truth, regardless of language, runtime, or encoding:
/spec/v0.5/ as its generated reading view (start with the
Base). The main branch remains the working copy.spec/v0.5/CONFORMANCE.mdspec/v0.5/schema.jsonspec/v0.5/schema.jsonldconformance/README.md with v0.5 fixtures at
conformance/v0.5/README.md - this fileCONFORMANCE.md - conformance requirements (evaluator,
holder, bilateral participant) plus the fixture indexschema.json - JSON Schema for structural validation of
v0.5 manifestsschema.jsonld - JSON-LD context defining v0.5
terms/spec/v0.2/SIGNATURE-PROFILE.md; it does not define
the tightened new-version predicate. Signature Profile B remains in
EXT-OPT Section O15.@context to include the v0.5
versioned context document
https://universalmanifest.net/ns/v0.5 (or include this URI
when an array form is used).schema.jsonld in this directory; the
verbatim body and its content hash are reproduced in Base Appendix B.
Evaluators verify the sha2-256 multihash
zQmR1W9vsPrdbXmhH38jdNhdEhBB1C7JUL1Z9kZMi6u16i7 over the
exact document bytes rather than trusting any retrieval channel (Base
Section 6.10; computation rule in EXT-OPT Section O1.7).um: term prefix is version-independent and expands
to the base IRI https://universalmanifest.net/ns/um#.https://universalmanifest.net/ns/v0.5,
https://universalmanifest.net/ns/universal-manifest/v0.5/schema.jsonld,
https://universalmanifest.net/ns/universal-manifest/v0.5/schema.json
and https://universalmanifest.net/spec/v0.5/. The
schema.jsonld and schema.json in this
directory are the release artifacts, and the served bytes must equal
them.
https://universalmanifest.net/ns/universal-manifest/v0.5/schema.jsonld
— context endpoint for the exact bytes of schema.jsonld in
this directoryhttps://universalmanifest.net/ns/universal-manifest/v0.5/schema.json
(schema.json in this directory)https://universalmanifest.net/spec/v0.5/ (rendered from
the document set in this repository)Public v0.4 and new-candidate context identity. The
published v0.4 state is the canonical repository snapshot
c13fb2c3de157050a5862695d1e6ae8d8943deb5, which carries no
release tag. Its repository and hosted v0.4 context bytes have SHA-256
91167fd510c39f93a023fc15a7474b77bc4414188ddb2d73806dc08e0f69345c,
multihash zQmY71ixd6ez6UkXpzGJ7fiMpwL26jzQwWp9CV6kwX4Kb4f.
Those bytes remain valid and must be preserved at the old identity. This
v0.5 candidate intentionally reuses the exact local editorial context
bytes, SHA-256
27ab5a3cc491cee7a5947b73f550c07ecbff285a81a6ba801aa98f60046dcea4,
multihash zQmR1W9vsPrdbXmhH38jdNhdEhBB1C7JUL1Z9kZMi6u16i7;
its URI differs because the contract version differs, although no
version-specific IRI occurs inside the context body. Base Appendix B
reproduces that same body. A content hash identifies bytes, not the
complete versioned processing contract. Do not replace the old public
context or reinterpret old signatures.
Every conformant evaluator MUST execute these stages in order for every v0.5 manifest it processes. Each stage produces a defined output that feeds the next, and the evaluator MAY stop early at any stage by emitting a rejection.
@type
includes um:Manifest; preserve (but do not act on) unknown
members so they survive signature verification.consents[].facetRef
against each projected facet's @id; check scope, purpose,
and validity window. Sealed (encrypted) facets are acknowledged as
present, never guessed at.accepted, accepted-with-warnings,
accepted-partial, or rejected. The result MUST
be machine-readable and MUST include per-facet status.See Base Section 3.1 for normative details and CONFORMANCE for the portable contract and evidence limits.
Four tiers, strictly additive. A claim verified at a higher tier automatically satisfies every lower tier. Relying parties choose their tier from their own threat model; the specification mandates no minimum.
trustTierUnsupported and
MUST NOT downgrade.See Base Section 6.4.2 for normative tier definitions, Section 6.4.5
for requiredTrustTier enforcement, and Section 6.4.6 for
bilateral exchange semantics.
v0.5 separates what a manifest means from how it is written down.
See Base Section 1.7 and Section 4.5; EXT-OPT Section O1.
v0.5 carries forward all v0.2 and v0.3 security guidance and adds:
attestedAt; evaluators MUST NOT
treat a stale liveness attestation as current human presence.presentationProof binding the
manifest to the verifier's challenge and
audience.sessionId and exchangeId MUST each be
generated with at least 128 bits of entropy and MUST NOT encode party
identifiers.requiredTrustTier enforcement remains
normative and raise-only - claim and facet floors MUST NOT lower the
manifest-level floor.UM is an envelope, evaluation, and receipt layer that composes with neighboring standards. It does not replace them, and several capability families are deliberately deferred rather than silently missing.
Carrier and provenance standards (C2PA, COG, and similar). A manifest can carry — or point to — an external-standard artifact unmodified, while UM supplies the governance record around it: signature, consent and current-state policy, projection, and receipt evidence. The carried artifact remains independently verifiable under its own standard: a receiver performs two independent checks (the UM envelope evaluation, and the carried standard's own verification), and passing one never substitutes for the other. UM defines no C2PA/COG trust-list, certificate-chain, asset-binding, or media-provenance verification semantics. See EXT-OPT Section O1.6. The same posture holds for OGC GeoPose (O14 PREVIEW carries or points to GeoPose payloads; validating GeoPose data and reference frames remains OGC's job) and for W3C VC/DID material (carried and verified as claims/claimProof; UM issues nothing and mandates no DID method).
Explicitly deferred capabilities. The following capability families remain outside the current Base contract or require a future registered profile. PREVIEW work remains subject to the maturity and evidence limits in CONFORMANCE; this list does not certify implementation coverage:
ageOver) and Tier-2 privacy hooks; holder-side predicate
synthesis is a future registered profile, and an issuer-signed boolean
is not equivalent to a holder-generated proof.scope/conditions/purpose-binding,
encrypted facets, sealed entries, and pointer-first storage; a
declarative storagePolicy object, retention-duration tiers,
and a storage-conflict reason code are future registered-profile work,
and unenforceable storage requests fail closed today.Domain integration guidance (non-normative).
RP1/spatial-fabric, smart-glasses/bystander consent, healthcare handoff,
education/competence credentials, social/profile projection, agent
ecosystems (ERC-8004, MCP, A2A, OASF), OMATrust proof alignment, and
public discovery/publication policy are supported today by composing
Base + EXT mechanics; they are documented as integration guidance
(integrations/), not Base normative scope. Domain-specific
profiles may be registered later through EXT-OPT Section O6.
Not UM scope (composes, never owned). Ad-network/marketplace operation, payment rails, custody, escrow execution, buyer-data sharing; C2PA/COG verification semantics; GeoPose validation; VC issuance and DID-method governance; authentication and federation protocols (OIDC, ActivityPub/AT); legal or regulatory compliance certification; and device/TLS/network side-channel control.
v0.5.0 is published for review and implementation. It keeps the wire
files of release candidate 1 unchanged and corrects text only. Its
supported runtimes are the two reference evaluators: TypeScript
(universal-manifest 0.1.0 on Node.js 22) and Rust
(um-core, built for macOS on Apple silicon). Each matches
all 198 cases of the v0.5 index and all 79 of the v0.4 index. No
browser, game-engine or device runtime is supported in v0.5.
Qualification of the protected exchange, with the release obligations
that depend on it, is planned for v0.6. Individual advanced/optional
surfaces retain their existing PREVIEW grades; a Grade-2 candidate
member is not promoted into the schema/context by this version change.
The public v0.4 snapshot and older versions remain available at their
existing identities. No full capability, hardware, protected-carrier or
institutional-approval claim is made.
Selective minimum disclosure is in v0.5 as a bounded Base profile: holder-controlled projection, consent gating, sealed-entry handling, and receipt evidence. The deeper high-privacy per-target work remains a separate implementation chain: v0.5 does not claim unlinkability, full or partial, cover-slot indistinguishability, resolver-operator privacy, or complete high-privacy conformance unless a later profile explicitly does so.
Future changes are proposed as issues and pull requests on the
canonical UniversalManifest/universal-manifest repository,
and its maintainers accept them after review there. Change control for
the media types and the profile registry rests with those
maintainers.
v0.5 releases as a mechanism with reference suppliers. It claims no real-world identity, issuer or revocation result. All 43 obligations and the 40 permission outcomes stay in force and stay counted. This boundary limits what a result can show, and it removes no requirement. Real suppliers are later work, and this release does not drop them.
unchecked. Short-lived re-issuance limits staleness only up
to the expiry. It is not immediate revocation, and it gives the issuer a
log of who refreshed and when. No public status pointer is
required.High-privacy mode is an unresolved draft. This candidate selects no high-privacy mechanism, and this release does not offer the mode as available. In the draft, a presentation shows a new holder key for each check in place of a stable holder.
The mode does not make a holder unlinkable. This release claims no unlinkability, full or partial, and gives no measured result that the mode hides a holder or keeps two checks apart.
The draft is for one-off checks only. It requires a new holder key for each check. It must refuse delegated and long-lived use, and this release states that those uses are linkable. It is not promised for checks where lending defeats the purpose, such as age checks. This release offers no age check for real use. Its age tests run on reference credentials, reliance on them is refused, and they state no real fact about a person. Base §7.1 states these limits.
The open parts are:
This candidate selects manifestVersion: "0.5",
@context including
https://universalmanifest.net/ns/v0.5, and the
spec/v0.5/ artifact family. Every new v0.5 manifest must be
newly signed over its exact production signing input; changing the
version/context of a v0.4 document invalidates its old signature. Keep
original signed v0.4 bytes, context and interpretation unchanged.
Consumers supporting old versions must route by the declared version to
that version's rules, not apply the new Profile A predicate
retroactively. A consumer without a supported declared-version path
rejects the document; it does not silently downgrade or relabel it.
Relative to the saved published v0.4 snapshot, the candidate context
omits exactly facetSensitivityOrdering,
labels, order, profile and
trustWeight; shared term definitions are unchanged.
Sensitivity ordering remains profile-supplied and the derived-write join
requirement remains normative. trustWeight remains
specification-only until its value shape and schema/fixture/context
coverage are settled; this candidate adds no replacement private
carrier. The context pins under Namespace URIs distinguish public old
bytes from candidate bytes, even though the accepted local v0.4
editorial context and the v0.5 context are intentionally identical.
The candidate schema materializes Profile B (ES256, JCS,
P-256) alongside Profile A. Profile B uses raw 64-byte
R || S, rejects DER signatures, accepts high-S signatures
and imposes no RFC 6979 requirement; these rules are unchanged from the
accepted editorial contract. The inherited fixture plan adds three valid
ES256 controls (minimal, facets, high-S) and five invalid controls (bad
value, DER encoding, key substitution, SPKI algorithm mismatch,
post-signing tampering). The old sensitivity-ordering fixture remains
preserved historically but is absent from the accepted candidate index.
The unsupported-profile negative uses ES384 instead of
now-recognized ES256. These inherited controls are now
freshly signed and exercised for wire0.5 through the TypeScript command
adapter; the two explicit v0.3 compatibility fixtures retain their
original bytes. Finite fixture success does not qualify all normative
behaviors.
For v0.5 Profile A, canonical non-identity prime-order A and R, canonical S and the full-challenge uncofactored equation are required by Base §1.6.1.1. Weak identity, small-order and mixed-order points, noncanonical point encodings and noncanonical scalars cannot gain acceptance through library-specific normalization or a cofactored shortcut. Some inputs accepted by older library routes will be refused by the new rule. An ordinary valid signature is not an identity-authority proof: the existing key-resolution/authorization and tier rules still apply. Do not infer this rule from a function called strict verification; qualify the actual adapter against all six mathematical requirements.
The new source set has thirteen documents/artifacts: the eight-document tier stack, README, CONFORMANCE, draft profile registry, structural schema and context. It retains all 43 evaluator/holder/bilateral/suite obligations and conditional advanced/protected requirements, along with existing honest-result, evidence-ceiling and proximity limitations. Version selection does not select an unresolved high-privacy mechanism, adopt a private request/permission/receipt carrier, grant private initial trust or turn PREVIEW into implemented cryptography.
Dated finite qualification, 2026-10-21. The
candidate index contains 188 cases, all matched by the
actual TypeScript command adapter. Its SHA-256 is
6334fe4f61198d087ddb8260ba467ba42507dfb852d2643d5de84619dc804e3b.
The strict schema/ingress sweep passed 192 checks: 188
indexed inputs, three examples, and one targeted malformed-JWE control;
the current API qualification passed 38 named groups.
All 164 earlier fixture bodies and every original verdict, receipt
expectation and caller context remain unchanged, as does all
supplemental JWE index data. The earlier 114-to-164 expansion added
fifty signed consent/time/operation/gate-order/trust-floor controls and
clarified two caller read contexts; this context repair
does not change those again. The 24 additional cases have independently
valid outer signatures: 11 accepted controls and 13 context refusals.
Twelve refusals are structurally valid; the empty-root-array refusal is
structurally invalid. Schema validity alone does not establish trusted
context interpretation. Caller intendedScope and
intendedPurpose remain explicit context, never inferred
from a grant; omitted, empty, or mismatching purposes cannot authorize a
purpose-bearing consent. The separate 46/46 I-JSON checks are
dated 2026-10-16 and are not a rerun of the changed evaluator.
The finite report remains at conformance level none;
positive Tier-1+ verification, authenticated services, other holder
implementations and protected/private mechanisms remain unproved.
Refusal evidence is not positive conformance.
Receipt check assertions (2026-10-30). The candidate
index now asserts signatureCheck and
freshnessCheck on all 119 fixtures in evaluation mode, with
values derived from each fixture's bytes and evaluation clock. Its
SHA-256 is
940e23629c246596ef91c7ed7dc8f6fef6242fa6af3b287a2ba85555bc872a19.
The case count stays 188, and entries dated earlier describe the earlier
index. The TypeScript and Rust adapters each match 188 of 188. Rewriting
either field in a receipt now fails every fixture that asserts the
rewritten value.
Resource limits, consent entries and created
(2026-10-31). The candidate index adds eight signed cases and
now holds 196 cases, SHA-256
e8c72773514ddd0822c5b2b81f890ae9b557bad775309def1be70ed58b053434.
Six test the Base section 6.3 limits at each boundary, and two test more
than one consent entry on a facet. The TypeScript and Rust command
adapters each match 196 of 196, and 79 of 79 on the v0.4 index. Entries
dated earlier describe the earlier index. This is finite local evidence,
with conformance level none.
I-JSON ingress (2026-11-06). Base section 1.6.3 now
requires the whole manifest as received to be I-JSON, including members
outside the signing input. A fresh check had found that the TypeScript
evaluator accepted a 1.2 MB signed manifest with a lone surrogate inside
signature, which the Rust evaluator rejected. Both now
reject such input. The candidate index adds two signed cases and now
holds 198 cases, SHA-256
f3c812984f47f284a3254dc8d51b1cba8140a06ba41238f6bce3b7f8e4cdb97c.
The TypeScript and Rust command adapters each match 198 of 198, and 79
of 79 on the v0.4 index. Entries dated earlier describe the earlier
index. CONFORMANCE.md section 5 lists the known differences
between the two evaluators outside the suite.
Fixed offline reference context (implementation
limit). Before interpreting UM terms, the reference
implementation verifies the actual packaged normative UTF-8 context
bytes against the existing content hash. Its trusted reference forms are
the exact https://universalmanifest.net/ns/v0.5 scalar or a
nonempty dense list containing only that reference; repeated trusted
references are supported. In interpreted UM scopes, including the root,
facets, claims and consents, substitutions, inline remaps, resets and
arbitrary remote contexts refuse before signature execution or resolver
use. Refusal receipts report
signatureCheck: "not-evaluated",
freshnessCheck: "not-evaluated" and
keyRefResolution: "not-attempted". This is fixed offline
context support, not arbitrary JSON-LD expansion or remote loading, and
not a new restriction on the normative wire model. Entity data, embedded
credential/VP proof payloads and unknown-member payload contexts remain
byte-preserved, uninterpreted signed data; they confer no UM
interpretation, proof verification, identity or authority. A separate
selected profile or proof processor must establish its own semantics and
verification.
Detached TypeScript and Rust production Profile A predicates, in the separately dated 2026-10-16 experiment, agreed on 937 identical inputs: 3 accept, 934 reject, with every result matching its retained expectation. This is signature-predicate agreement, not full evaluator, SDK, application, or runtime parity. The current corpus retains the two encrypted-Entity manifests signed for that qualification, preserving four accepted two-recipient ciphertexts. TypeScript verifies their outer signatures and records all four facets as opaque; it does not decrypt them. Separate Rust/panva-jose calls read those persisted signed files and recovered the exact canonical Entity bytes in 16 opens: 8 cross-implementation and 8 self-controls. Earlier Entity evidence of 112 ordered wrong-key refusals reaching unwrap and 24 tamper refusals is reused through exact unchanged ciphertext and dependency identities, not rerun or counted as additional fixtures. The older primitive experiment (8 cross-opens, 36 wrong-key refusals, 156 malformed/tampered refusals) remains separate and is not combined into these totals. See the conformance qualification limits and implementation-specific JWE limits.
Bounded Rust and local two-party evidence (2026-10-21). An explicit v0.5 Rust core/command adapter separately matched all 188 unchanged cases, including 119 with strict receipt assertions, and passed 25 HTTP controls on its phase-one binaries. Two persistent Swift callers subsequently exercised actual Rust evaluations, signed actions, receipt verification and reciprocal acknowledgments over injected local channel evidence. The full 52-control online suite per version, including eight completed encounters each, belongs to the preserved bilateral predecessor. The final compatibility successor was qualified by all eight positive v0.5 encounters, one v0.4 encounter, eight HTTP responses, two byte-equal legacy comparisons, four byte-equal witness comparisons and focused independent controls; the predecessor's full matrix is not reassigned to that successor. Exact binary identities, API boundaries and test lineage are recorded in the reference-implementation evidence in CONFORMANCE §8.
Across four retained caller runs and a fresh final-binary run,
independent inspection traced 270 action/delivery/persistence
links and 54 verified signed receipts across 27 completed
encounters. These are sums across distinct scoped runs, not a
single matrix or certification total. Receipt bodies match both actual
evaluator digests and signatures verify against independently installed
role keys. The fresh run contains two mixed-suite positives and a
signed-body tamper refusal. The reused FFI library supplies unchanged
signing/prerequisite machinery, not a newly qualified v0.5 ABI
evaluator. This local evidence does not establish holder generation,
peer identity, real radio, source permission or withdrawal, protected
adoption, physical/device behavior, whole-SDK parity or complete
conformance; level remains none. No fixture, wire member,
profile or normative obligation changed.
The contract clarifications preserve distinct boundaries: a status
locator/cursor inside the excluded signature object is not
envelope-authenticated; current permission needs independently trusted
authority and target binding; opaque cursors and HTTP ETags supply no
implicit ordering or permission grant. Capability claims identify
implemented, partial, and unimplemented obligations with actual checks
and unmet baselines. Circuit verification uses trusted pinned
key/public-input bindings, and model-integrity claims require a
profile-defined computation and evidence binding. A request omitting its
purpose cannot match purpose-bearing consent. These clarifications add
no protected current-state mechanism, capability carrier, signed-receipt
mandate, atomic-write mechanism, or new proof-system choice.
All 43 mandatory/conditional obligations and advanced protected/privacy requirements remain. The finite corpus and source evidence above are reconciled, and each release records its own reader, public byte, route and citation checks. Historical v0.4 reports remain old-version evidence. Publication as a tagged release claims no native publication, full implementation qualification, hardware provenance, privacy certification, or institutional ratification.