| Internet-Draft | Witnessed Integrity Attestations | November 2026 |
| Kornoelje | Expires 22 May 2027 | [Page] |
This document specifies a Witness Attestation Record (WAR), a signed and hash-linked record created by a witness after it accepts an observation about a subject system. Each record identifies the subject, observation stream, checkpoint, observation time, evidence commitment, witness-local sequence number, and reference to the preceding record. A verifier can validate the record signature, reconstruct the supplied chain, evaluate the applicable policy, and assess freshness against a trusted checkpoint or independently retained chain head.¶
The component is designed to preserve a verifiable chronology of observations and actions across vulnerability exposure, incident investigation, remediation, and post-remediation verification. It complements evidence-generation and transparency systems. It does not itself establish that a measurement source is truthful, that a host was uncompromised, or that every real-world event was observed.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 22 May 2027.¶
Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
This document defines a proposed, general-purpose component, referred to as Witnessed Integrity Attestations (WIA), for creating and verifying cryptographically linked witness attestations about observed system state. It is intended for technical scrutiny, implementation experiments, and standards-community discussion.¶
This is an independent working proposal. It is not an RFC, an adopted IETF work item, an IETF-endorsed specification, or evidence that the design has been accepted by a standards body. This initial version is a consolidated proposal baseline; it does not mean the protocol has completed interoperability testing, formal security analysis, or standards review.¶
The proposal is vulnerability-independent. A PHP2Shell-style exploitation and remediation sequence is included only as a motivating example. No claim is made here that a particular system was compromised or that the illustrative incident timeline represents observed facts.¶
The specification distinguishes cryptographic authenticity, chronological continuity, policy conformance, evidence quality, and freshness. These are separate verification questions and MUST NOT be collapsed into a single unqualified 'verified' result.¶
Incident response commonly produces many separate artifacts: vulnerability findings, host measurements, package inventories, forensic snapshots, remediation records, restart events, and post-remediation scans. These artifacts may be individually signed or hashed while still leaving uncertainty about when a witness accepted them, whether the presented history is complete, or whether records were later reordered or substituted.¶
A content digest can show that presented bytes match a previously committed digest. A digital signature can show that a key signed a particular statement. Neither property, by itself, establishes continuity of a witness's accepted observation sequence. This proposal adds an explicit witness-local sequence and a cryptographic predecessor reference to each attestation.¶
The intended outcome is a portable evidence structure that allows a relying party to ask: who witnessed what, for which subject and stream, under which policy, at what recorded observation point, and how does this record connect to the witness's earlier accepted state?¶
The component addresses the risk that a party receiving incident evidence through an untrusted relay cannot independently determine whether the witness's previously recorded observations have been omitted, reordered, or replaced.¶
Design goals:¶
Bind each attestation to a specific witness, subject, observation stream, checkpoint type, evidence commitment, and policy context.¶
Provide a sequence that increases by exactly one within a defined witness/subject/stream scope.¶
Cryptographically link each attestation to its immediate predecessor.¶
Permit independent verification of signatures, chain continuity, policy results, and freshness as distinct dimensions.¶
Support late-arriving historical evidence without retroactively inserting it into the witness's accepted sequence.¶
Permit multiple witnesses and transparency services without pretending their independent sequence spaces are globally ordered.¶
Define a conservative baseline encoding and signature profile suitable for interoperable prototyping.¶
Non-goals:¶
Proving that the subject system was never compromised.¶
Guaranteeing that a compromised endpoint, sensor, attester, or witness supplied truthful measurements.¶
Replacing incident response, digital forensics, secure boot, remote attestation, software transparency, or a transparency service.¶
Establishing absolute time solely from a witness-provided clock.¶
Guaranteeing availability, preventing withholding, or proving that no unobserved event occurred.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
| Term | Definition |
|---|---|
| Attester | Entity or measurement source that produces evidence about a subject. |
| Witness | Entity that evaluates received evidence under a stated acceptance policy and signs a WAR. |
| Subject | System, device, workload, software instance, service, or other object being observed. |
| Observation stream | A named sequence of observations for one witness and subject; sequence numbers are scoped to this stream. |
| Checkpoint | A typed observation or lifecycle event represented by a WAR. |
| Evidence manifest | A structured description of evidence objects and their digests, identifiers, and acquisition metadata. |
| Chain head | The latest WAR or authenticated checkpoint known for a particular stream. |
| Relying party | Verifier or consumer that evaluates WARs and their associated evidence. |
| Transparency service | An optional service that records signed statements and provides inclusion or consistency evidence. |
The architecture separates evidence production, witness acceptance, durable sequencing, and independent verification.¶
The attester produces evidence or a reference to evidence about the subject.¶
The witness validates the evidence against an identified policy. It records the result and the evidence commitment it accepted.¶
The witness allocates the next sequence number in the relevant stream, links the new record to the prior record, and signs the resulting WAR.¶
The witness durably stores the signed record and advances its protected chain state.¶
The WAR may be submitted to one or more transparency services or distributed to relying parties.¶
A relying party verifies the signature, chain, policy, evidence references, and freshness using independently obtained trust anchors and chain-head information.¶
A witness is not automatically an independent measurement authority merely because it is operated by a different organization. Independence depends on the provenance of measurements, administrative control, hardware and software trust, key custody, and operational separation.¶
A transparency-service receipt proves only the properties defined by that service's protocol. It does not, by itself, prove that the witness observed the event at the stated time or that the witness maintained a unique sequence.¶
A WAR is a signed statement. The signed payload contains all fields that affect identity, scope, ordering, interpretation, or verification. Implementations MUST NOT place security-critical WAR semantics exclusively in unprotected COSE headers.¶
The payload MAY include additional extension fields. An extension that changes security interpretation MUST be defined by a profile and MUST be covered by the signature.¶
The witness's evaluation result MAY be carried as a signed policy-result field in a future profile. A verifier's own conclusions MUST remain verifier-generated and MUST NOT be trusted merely because a producer labels a record 'verified'.¶
| Field | Type / form | Requirement and meaning |
|---|---|---|
| war_version | unsigned integer | REQUIRED. Version of this WAR profile; value 1 for this specification. |
| witness_id | text string | REQUIRED. Stable identifier for the witness within the deployment's trust model. |
| subject_id | text string | REQUIRED. Stable identifier for the observed subject, preferably privacy-preserving or pseudonymous where appropriate. |
| stream_id | text string | REQUIRED. Identifier for the observation stream; sequence numbers are scoped to witness_id + subject_id + stream_id. |
| witness_seq | unsigned integer | REQUIRED. Positive sequence number that increases by exactly one within the stream. |
| observed_at | text string | REQUIRED. UTC timestamp in RFC 3339 format, with 'Z' suffix and stated precision. |
| event_time | text string | OPTIONAL. Claimed time of the underlying event when distinct from observed_at, in the same RFC 3339 UTC syntax as observed_at; it is not the sequence-ordering time. |
| checkpoint_type | text string | REQUIRED. Registered or deployment-defined checkpoint type. |
| checkpoint_digest | byte string | REQUIRED. SHA-256 digest of the deterministic-CBOR evidence-manifest bytes. |
| evidence_manifest_ref | text or URI | OPTIONAL. Retrieval reference; the digest remains authoritative for integrity. |
| previous_attestation | byte string or null | REQUIRED. SHA-256 digest of the complete serialized predecessor COSE_Sign1 object; null only for sequence 1. |
| policy_identifier | text string | REQUIRED. Identifier for the policy used by the witness. |
| policy_version | text string | REQUIRED. Version or immutable digest-based identifier of that policy. |
| witness_key_id | byte string (bstr) | REQUIRED. Key identifier corresponding to the signing key. |
This initial version defines a conservative prototype profile to make independent implementations possible. It does not preclude later algorithm agility through a separately specified profile.¶
Payload encoding: deterministic CBOR as defined by [RFC8949] . Implementations MUST use deterministic encoding for the WAR payload and evidence manifest.¶
Signature algorithm: Ed25519 (COSE algorithm identifier -19) as specified in [RFC9864] , which defines the fully specified COSE Ed25519 identifier and updates [RFC9053], subject to the key-management and implementation requirements of the applicable COSE profile.¶
Digest algorithm: SHA-256 for checkpoint_digest and previous_attestation.¶
Key identifier: the COSE protected 'kid' header parameter MUST identify the signing key. The payload witness_key_id MUST be consistent with the protected key identifier or a defined key-registry mapping.¶
The COSE protected header MUST contain the algorithm identifier and key identifier. Any additional header parameters that affect interpretation MUST be protected. The payload MUST be the deterministic-CBOR encoding of the complete WAR map.¶
The COSE protected-header map MUST use the deterministic encoding requirements of [RFC8949]. The resulting protected-header byte string is part of the COSE_Sign1 signature structure and MUST be preserved and used unchanged during signature verification. A verifier MUST NOT decode and re-encode the received protected-header map before verifying the signature. The baseline profile's protected map contains the Ed25519 algorithm identifier and the key identifier; any future profile that adds protected parameters must define their deterministic encoding and criticality requirements.¶
The predecessor digest is computed over the exact serialized bytes of the predecessor COSE_Sign1 object, including its COSE structure and signature. Implementations MUST NOT hash a decoded-and-re-encoded approximation of the predecessor.¶
The checkpoint digest is computed over the exact deterministic-CBOR bytes of the evidence manifest. The manifest MUST define the digest algorithm and each evidence object's identity, media type where relevant, byte length where available, and digest. The manifest itself MUST be retained or retrievable for meaningful evidence validation.¶
This profile is a baseline for experimentation. A production deployment MUST document key generation, secure storage, rotation, revocation, algorithm deprecation, and recovery procedures.¶
The baseline profile uses [RFC8949] core deterministic encoding, COSE_Sign1 per [RFC9052] , Ed25519 through COSE algorithm identifier -19, and SHA-256. The application profile requires an untagged COSE_Sign1 array and an empty unprotected-header map. Implementations MUST reject a tagged object under this profile rather than silently accepting two equivalent encodings. This restriction is an application-profile choice, not a general restriction on COSE.¶
WAR payload map keys are the UTF-8 text strings defined in Appendix C. Required fields and types are defined in the accompanying CDDL.
checkpoint_digest
is exactly 32 bytes and commits to the deterministic-CBOR encoding of the evidence manifest.
previous_attestation
is null only at witness_seq 1; otherwise it is exactly 32 bytes and equals SHA-256 over the complete serialized predecessor COSE_Sign1 byte string. The
witness_key_id
value
MUST
equal the protected COSE kid byte string in this profile.¶
A producer MUST validate its own output before durable commit. It MUST allocate witness_seq atomically per (witness_id, subject_id, stream_id), persist the exact serialized COSE_Sign1 bytes and updated head before acknowledging issuance, and prevent concurrent writers from allocating the same sequence. A verifier MUST validate the COSE structure, protected headers, signature, payload deterministic encoding, field schema, stream scope, sequence and predecessor links, and any available evidence-manifest digest. Trust in the public key, policy, time source, and chain-head anchor is an external verifier decision.¶
The reference implementation uses
cbor2
with canonical encoding and
cryptography
Ed25519 primitives. It is an implementation aid, not a substitute for the normative RFCs. Deployments
MUST
pin and review dependency versions and
MUST NOT
use the included deterministic test key in production.¶
A chain is maintained independently for each tuple (witness_id, subject_id, stream_id). A sequence number has no global meaning outside that tuple.¶
For the first record in a newly initialized stream: witness_seq = 1 previous_attestation = null For each subsequent record n: witness_seq(n) = witness_seq(n-1) + 1 previous_attestation(n) = SHA-256(exact_serialized_COSE_Sign1(n-1))¶
A witness MUST allocate sequence numbers atomically and MUST persist the signed record and updated chain state before acknowledging successful issuance. A witness MUST NOT reuse a sequence number within a stream. It MUST NOT silently replace an already issued record.¶
A verifier supplied with a contiguous chain MUST check that sequence numbers increase by exactly one and that every predecessor digest matches the exact preceding serialized COSE_Sign1 object. A missing record produces an incomplete chain unless an authenticated checkpoint or separately defined compact-proof mechanism establishes the missing prefix.¶
A chain beginning at a later sequence number MUST be treated as anchored or partial, not as a complete genesis-to-head history, unless the verifier has an authenticated initialization record or checkpoint that binds the starting state.¶
A valid predecessor chain demonstrates continuity among the records presented and anchored. It does not, by itself, prove that the witness never signed a conflicting fork or that no records were withheld.¶
A witness can potentially sign two different WARs with the same stream scope and witness_seq. Hash chaining alone does not prevent this: both forks may be internally valid. This specification therefore treats equivocation detection as a distinct requirement.¶
A deployment SHOULD publish or distribute signed chain-head statements at defined intervals to one or more independently operated transparency services, auditors, or relying parties. A chain-head statement SHOULD bind witness_id, subject_id, stream_id, witness_seq, digest of the head WAR, publication time, and the witness key identity.¶
A relying party SHOULD compare a presented chain head with a previously retained or independently obtained head. Two different head digests for the same stream and sequence are evidence of equivocation. A later head that does not descend from a previously trusted head is evidence of a fork or discontinuity.¶
The protocol does not claim that a transparency service alone prevents equivocation. Stronger detection depends on independent observation, gossip, cross-logging, or other mechanisms that make conflicting signed statements visible.¶
If a verifier has no prior head, no independent publication, and only one presented chain, it can validate the chain's internal consistency but cannot conclude that the witness never issued another fork.¶
The witness sequence orders the witness's acceptance and recording of observations. It does not automatically establish the true chronological order of underlying real-world events.¶
When historical evidence arrives late, the witness MUST create a new WAR at the next sequence number and MAY include event_time to describe the evidence's claimed event time, using the syntax defined for observed_at. The witness MUST NOT retroactively insert the late-arriving event into an earlier sequence position.¶
observed_at is the witness's recorded observation time. In this baseline profile it MUST use the RFC 3339 date-time form with a UTC designator (uppercase "Z"), a complete date and time including seconds, and an optional fractional-second component of one through nine decimal digits. Numeric UTC offsets and leap-second values are not accepted by this profile. Implementations MUST reject values outside this syntax. The timestamp records the witness's assertion; it is not a trusted timestamp unless the deployment provides a separately specified trusted time source and binds its evidence to the WAR. Clock synchronization, clock rollback, timestamp precision, and time-source provenance SHOULD be documented.¶
If a time source is suspected of compromise or manipulation, the verifier SHOULD report time assurance as indeterminate or degraded even if signature and chain checks succeed.¶
The following checkpoint labels are recommended baseline values. Deployments MAY define additional values, but MUST document their semantics and versioning.¶
A checkpoint label describes the type of observation, not its success. For example, remediation-applied does not prove that remediation was correct or complete. The evidence manifest and policy must define what was actually checked.¶
| Checkpoint type | Meaning |
|---|---|
| baseline-observed | A baseline state or inventory was observed and committed. |
| exposure-observed | A relevant vulnerability or exposure condition was observed. |
| investigation-started | A formal investigation checkpoint was recorded. |
| evidence-collected | A defined evidence set was acquired and committed. |
| integrity-anomaly-observed | An integrity deviation or unexplained state was detected. |
| remediation-started | A remediation activity was initiated. |
| remediation-applied | A remediation change was reported as applied. |
| runtime-restarted | A restart or redeployment checkpoint was observed. |
| post-remediation-observed | The subject's state was measured after remediation. |
| verification-completed | A defined verification procedure completed with a recorded outcome. |
A verifier SHOULD report independent result dimensions rather than one Boolean status. At minimum, this specification defines the following dimensions:¶
| Dimension | Possible results | Interpretation |
|---|---|---|
| signature_status | verified / failed / indeterminate | Whether the COSE signature validates under an accepted key and algorithm. |
| chain_status | verified / failed / incomplete / indeterminate | Whether predecessor links and sequence continuity validate relative to the supplied anchor. |
| policy_status | verified / failed / indeterminate | Whether the evidence and checkpoint satisfy the identified policy as evaluated by this verifier. |
| freshness_status | current / stale / unknown | Whether the presented head is sufficiently recent or matches a trusted/retained head under local policy. |
| evidence_status | digest-matched / mismatch / unavailable / not-evaluated | Whether referenced evidence matches the committed manifest and whether the evidence was actually evaluated. |
A deployment MAY add dimensions such as timestamp_assurance, witness_trust, or transparency_status. It MUST define their semantics and MUST NOT silently map indeterminate or incomplete to verified.¶
A signature can be verified while the chain is incomplete. A chain can be internally consistent while stale. A policy can be satisfied while the underlying measurement source remains untrusted. User interfaces and machine-readable outputs SHOULD preserve these distinctions.¶
The verifier MUST independently compute its result. It MUST NOT accept a producer-supplied status field as proof of successful verification.¶
The proposal is intended to complement, not replace, existing standards and architectures.¶
COSE provides a standardized signed-message container. This proposal defines the semantics of a witness record and its chain relationships.¶
CBOR provides a compact data model and deterministic encoding foundation.¶
RATS provides an architecture and terminology for attesters, verifiers, and relying parties as described in [RFC9334] . A WAR can carry a commitment to evidence produced or evaluated within a RATS-oriented deployment.¶
SCITT provides mechanisms for registering signed statements and obtaining transparency receipts. A WAR may be submitted as a signed statement. A SCITT receipt can help establish registration in a transparency service but does not replace the witness-local predecessor chain.¶
Certificate Transparency ( [RFC9162] ) demonstrates append-only logging and consistency concepts. The witness chain and external transparency mechanisms have related but distinct roles.¶
Any claim that this component provides a novel property relative to SCITT, RATS, Certificate Transparency, or other transparency and attestation systems requires a careful feature-by-feature comparison. This specification does not claim that existing standards lack append-only or non-equivocation mechanisms.¶
All security-relevant metadata MUST be inside the signed payload or otherwise cryptographically bound to it. Implementations MUST NOT rely on COSE unprotected headers for integrity-critical values.¶
The SCITT architecture is specified in [RFC9943] . This proposal does not modify that architecture or claim to be an adopted SCITT extension; any integration requires separate technical review.¶
A subject MAY be observed by multiple witnesses. Each witness maintains its own sequence space and chain. Sequence number 42 from one witness is not inherently earlier or later than sequence number 43 from another.¶
Cross-witness corroboration requires an explicit correlation mechanism, such as a shared incident identifier, common evidence digest, common challenge/nonce, or a trusted external timestamping or transparency mechanism. A relying party SHOULD evaluate witness independence rather than count witnesses as interchangeable votes.¶
Where witnesses observe the same evidence manifest, each witness SHOULD bind the same manifest digest and identify its own acquisition and evaluation context. Differences between witness observations SHOULD be preserved rather than normalized away.¶
The primary security claim is deliberately narrow: given a trusted witness key, an accepted stream initialization or chain anchor, the exact serialized records, and suitable freshness information, a verifier can validate the authenticity and internal continuity of the witness's recorded sequence. Detection of equivocation requires independent comparison or publication. Truthfulness of measurements remains dependent on the attester, acquisition process, witness policy, and operational trust assumptions.¶
| Threat | What the component can provide | Residual limitation |
|---|---|---|
| Untrusted relay modifies a WAR | Signature failure or digest/chain mismatch should expose modification. | A relay can withhold records or deny service. |
| Historical insertion or reordering | Predecessor references and sequence continuity expose inconsistency against a trusted chain head. | A wholly fabricated first presentation without a prior anchor may remain internally consistent. |
| Witness equivocation | Conflicting signed records can be detected when independently compared or logged. | A single isolated verifier may not know a fork exists. |
| Witness key compromise | Key identifiers, revocation data, and external checkpoints help bound trust. | A compromised key may produce apparently valid new signatures until detected or revoked. |
| Attester or endpoint compromise | Evidence provenance and independent measurements can reduce risk. | A truthful signature over false measurements is still false evidence. |
| Clock manipulation | Separate time-assurance evaluation and external time sources can identify degraded confidence. | A witness timestamp alone is not proof of absolute time. |
| Record deletion or withholding | External chain-head publication and retained copies can reveal some gaps or stale views. | The component cannot guarantee record availability. |
| Policy substitution | Policy identifier and version are signed into the WAR. | A weak or incorrect policy can be faithfully applied. |
| Replay | Subject/stream identity, sequence, predecessor, and freshness checks constrain replay. | Verifier policy must define acceptable age and context binding. |
| Privacy leakage | Pseudonymous subject identifiers and selective disclosure can reduce exposure. | Metadata, timing, and repeated identifiers can still enable correlation. |
A deployment MUST define how witness keys are generated, protected, identified, rotated, revoked, and retired. Key rotation MUST NOT silently reset a stream's sequence or sever its chain.¶
A key transition SHOULD be represented by a signed transition statement that identifies the old and new key identifiers, the affected witness and streams, the effective sequence boundary, and the authorization basis. The transition statement SHOULD be cross-signed by the old and new keys where both remain trustworthy, or anchored through an independently trusted administrative or transparency mechanism.¶
After loss or restoration of witness state, the witness MUST NOT resume issuance from an unverified backup or invent a sequence continuation. Recovery MUST use an authenticated last-known chain head. If continuity cannot be established, the witness MUST create a new stream identifier and explicitly record the discontinuity and recovery reason.¶
Revocation does not erase the historical fact that a signature was produced. Verifiers need policy for evaluating records signed before and after the effective compromise or revocation time, recognizing that the exact compromise time may be unknown.¶
WAR metadata can reveal system identities, software deployment patterns, incident timing, vulnerability exposure, and organizational relationships. Deployments SHOULD minimize identifying metadata, use scoped or pseudonymous subject identifiers where feasible, restrict evidence-manifest access, and define retention and disclosure policies.¶
A digest is not necessarily privacy-preserving: low-entropy or guessable evidence can be recovered through dictionary testing. Sensitive evidence SHOULD be committed using an appropriately designed manifest and access-controlled storage model. Selective disclosure and salted commitments require a separately specified profile.¶
A conforming producer MUST:¶
Emit deterministic-CBOR WAR payloads using the required fields.¶
Wrap each payload in COSE_Sign1 using the specified baseline signature profile.¶
Protect the algorithm and key identifier in the COSE protected header.¶
Maintain an atomic, durable sequence for each witness/subject/stream tuple.¶
Hash the exact serialized predecessor COSE_Sign1 bytes when populating previous_attestation.¶
Use null predecessor only for sequence 1 in a newly initialized stream.¶
Commit the deterministic-CBOR evidence-manifest bytes using SHA-256.¶
Document its policy identifier, policy version, key lifecycle, and recovery behavior.¶
A conforming verifier MUST:¶
Validate the COSE structure and signature against an accepted key and algorithm policy.¶
Validate required payload fields, types, scope, and version.¶
Recompute and compare checkpoint digests when the evidence manifest is available.¶
Validate sequence continuity and predecessor digests for the supplied chain.¶
Report incomplete chains, missing evidence, stale heads, and indeterminate results distinctly.¶
Evaluate policy independently and preserve the separate verification dimensions.¶
Check available revocation, key-transition, and trusted chain-head information according to local policy.¶
Interoperability testing SHOULD include at least two independently developed producers and one verifier, with shared test vectors covering valid chains, modified payloads, missing predecessors, duplicate sequence numbers, forks, late evidence, key rotation, and recovery.¶
The package includes a reference implementation and deterministic test scenarios. A second implementation MUST independently produce and consume the same byte-level vectors before interoperability can be claimed. At minimum, the second implementation SHOULD use a different COSE/CBOR library or language. The test report MUST record implementation name, version, operating environment, vector identifier, expected result, actual result, and any byte-level differences.¶
Required vector classes: valid genesis and continuation; modified payload; invalid signature; missing predecessor; wrong predecessor digest; sequence rollback; same-sequence fork; late-arriving historical evidence; mismatched evidence manifest; unsupported algorithm; non-empty unprotected headers; tagged COSE_Sign1; and key-rotation transition. A pass by the included producer and verifier alone demonstrates internal consistency only.¶
This example demonstrates how the component could structure evidence across a vulnerability response. It is not a claim about a specific affected host, a confirmed compromise, or a verified incident timeline.¶
The sequence provides a verifiable record of what the witness accepted and in what order. It does not establish that the investigation found every malicious artifact or that the system was clean before the baseline. The quality of the result depends on evidence coverage, acquisition integrity, and policy design.¶
| Sequence example | Checkpoint | Evidence commitment may cover |
|---|---|---|
| 1 | baseline-observed | Host identity, package inventory, runtime version, configuration snapshot. |
| 2 | exposure-observed | Vulnerability scanner result, affected component version, exposure window assessment. |
| 3 | investigation-started | Incident case identifier, acquisition plan, initial log preservation record. |
| 4 | evidence-collected | Relevant logs, process inventory, persistence checks, file integrity results, outbound connection records. |
| 5 | remediation-started | Approved change record and remediation plan. |
| 6 | remediation-applied | Package update or mitigation evidence, configuration diff, deployment record. |
| 7 | runtime-restarted | Service restart or redeployment evidence. |
| 8 | post-remediation-observed | New package inventory, runtime version, configuration and integrity measurements. |
| 9 | verification-completed | Policy evaluation, residual findings, exceptions, and verifier outcome. |
A minimal prototype SHOULD include:¶
A deterministic-CBOR encoder and decoder with strict duplicate-key and type validation.¶
A COSE_Sign1 producer and verifier using the baseline profile.¶
A durable per-stream sequence store with atomic append semantics.¶
A chain validator that accepts an explicit trusted anchor and reports gaps or forks.¶
An evidence-manifest generator and digest verifier.¶
A chain-head export mechanism suitable for independent retention or transparency-service submission.¶
A test harness with positive and adversarial test vectors.¶
Implementations SHOULD fail closed on malformed encodings, duplicate or ambiguous fields, unsupported extensions that change security interpretation, invalid signatures, sequence rollback, predecessor mismatch, and policy violations. Diagnostic output SHOULD preserve the reason for failure without exposing secrets or sensitive evidence.¶
This version makes no IANA requests. A future standards-track or community-reviewed revision may request registries for checkpoint types, extension identifiers, policy-result codes, or profile identifiers after the data model and interoperability requirements stabilize.¶
Is a witness-local hash chain plus externally retained chain heads a sufficiently clear and useful incremental component alongside SCITT and existing transparency mechanisms?¶
Should this version retain the single Ed25519/SHA-256 baseline or define an explicit algorithm-agility registry/profile mechanism now?¶
Should chain-head statements be specified in this document or delegated to an existing transparency protocol?¶
What exact initialization and re-anchoring procedure provides adequate protection against rollback after witness restoration?¶
Which evidence-manifest structure best supports large evidence sets, selective disclosure, and privacy without ambiguity in digest computation?¶
How should a verifier express uncertainty about the time of key compromise and the validity of historical signatures?¶
What minimum test vectors and independent implementations would demonstrate interoperability rather than merely self-consistency?¶
| Property | Provided by this component? | Qualification |
|---|---|---|
| Authenticity of a WAR | Yes, conditionally | Requires valid signature and trusted witness key. |
| Integrity of presented WAR bytes | Yes | Signature and exact-byte digest checks detect alteration. |
| Continuity of presented chain | Yes, conditionally | Requires complete chain or trusted anchor and valid predecessor links. |
| Detection of witness equivocation | Not by chain alone | Requires independent head comparison, gossip, or transparency publication. |
| Proof of absolute event time | No | observed_at is a witness assertion unless separately time-attested. |
| Truth of endpoint measurements | No | Depends on attester, measurement path, and witness policy. |
| Proof that no compromise occurred | No | Absence of evidence is not evidence of absence. |
| Evidence of remediation sequence | Yes, conditionally | Shows accepted checkpoints, not necessarily remediation completeness or efficacy. |
| Availability of historical records | No | Requires external retention and distribution mechanisms. |
Witnessed Integrity Attestations define a narrow, composable mechanism for binding a witness's accepted observations into a signed, sequential, hash-linked history. Its intended value is to make the provenance and continuity of post-incident system-state records easier for independent parties to inspect and verify.¶
The component's claims are intentionally bounded. It can support authenticity and continuity checks for recorded observations, while relying on separate mechanisms for trustworthy measurement, independent time, non-equivocation publication, evidence retention, and policy assurance. The next validation step is an adversarial standards-overlap review and independent prototype implementation, not an assertion of novelty or adoption.¶
This initial version defines a conceptual model and an implementation boundary. It includes an explicit implementation profile, a CDDL data model, an executable Python reference implementation, and a test-vector package. No claim of independent interoperability, formal verification, or security certification is made.¶
The normative wire format is deterministic CBOR payloads carried in COSE_Sign1. This version specifies an untagged four-element COSE_Sign1 array for the baseline profile: [protected, unprotected, payload, signature]. The protected header contains alg = -19 (Ed25519) and kid. The unprotected header map MUST be empty. The payload is the complete deterministic-CBOR WAR map. Detached payloads are not supported by this baseline.¶
The following JSON is explanatory only. It is not the canonical wire format. Binary values that are CBOR byte strings (bstr) are shown as Base64URL text placeholders for readability. The normative payload encoding for this profile is deterministic CBOR.¶
{
"war_version": 1,
"witness_id": "witness.example.org",
"subject_id": "asset:opaque-7f3a",
"stream_id": "incident:case-2026-0042",
"witness_seq": 9,
"observed_at": "2026-11-03T12:30:00Z",
"checkpoint_type": "post-remediation-observed",
"checkpoint_digest": "BASE64URL(SHA-256(manifest-cbor))",
"previous_attestation": "BASE64URL(SHA-256(prior-COSE_Sign1))",
"policy_identifier": "urn:example:system-integrity-policy",
"policy_version": "sha256:...",
"witness_key_id": "witness-key-2026-01"
}¶
The actual COSE_Sign1 protected headers and signature are outside this JSON illustration. Production serialization MUST follow the encoding and cryptographic profile in Section 6.¶
| Test | Expected behavior |
|---|---|
| Valid genesis and continuation | Accept sequence 1 with null predecessor, followed by sequence 2 linked to exact sequence-1 bytes. |
| Payload tampering | Reject a modified signed payload. |
| Signature substitution | Reject a record whose signature does not validate under the identified trusted key. |
| Missing predecessor | Report incomplete when sequence n references a predecessor not supplied and no trusted anchor is available. |
| Wrong predecessor digest | Reject chain continuity when the predecessor digest does not match exact serialized bytes. |
| Sequence rollback | Reject or flag a sequence that is not the next expected value for the stream. |
| Same-sequence fork | Detect conflicting signed records for the same stream and sequence when both are presented. |
| Stale chain head | Report stale when presented head does not match or descend from a more recent trusted head. |
| Late-arriving evidence | Accept as a new sequence with event_time metadata; do not insert retroactively. |
| Key rotation | Validate a documented transition without resetting the stream sequence. |
| Witness recovery | Require authenticated recovery from a retained head or a new stream with explicit discontinuity. |
| Evidence mismatch | Report mismatch when manifest bytes do not match checkpoint_digest. |
; WAR payload. CBOR map with text-string keys.
war = {
"war_version": 1,
"witness_id": tstr,
"subject_id": tstr,
"stream_id": tstr,
"witness_seq": uint .gt 0,
"observed_at": tstr, ; RFC 3339 UTC timestamp ending in Z
? "event_time": tstr, ; same syntax as observed_at
"checkpoint_type": tstr,
"checkpoint_digest": bstr .size 32,
? "evidence_manifest_ref": tstr,
"previous_attestation": bstr .size 32 / null,
"policy_identifier": tstr,
"policy_version": tstr,
"witness_key_id": bstr,
* tstr => any ; extensions are profile-governed
}
; Application profile COSE_Sign1 structure (untagged).
cose-sign1 = [
protected: bstr,
unprotected: {},
payload: bstr,
signature: bstr
]
; Protected header CBOR map: { 1: -19, 4: kid-bstr }.
; No detached payload. No unprotected parameters.¶
The data model is expressed in CDDL as defined in [RFC8610] .¶
The accompanying ZIP contains:
wia_v11.py
(reference producer, verifier, and chain validator),
wia_v11.cddl
(machine-readable data model),
test_wia_v11.py
(automated tests),
requirements.txt
,
vectors/
(scenario definitions and expected outcomes), and
README.md
. Run the suite in a clean Python 3.10+ environment after installing the pinned dependencies. The reference implementation deliberately fails closed on unsupported COSE shapes, algorithms, protected-header mismatches, payload schema errors, invalid signatures, and chain discontinuity.¶
The original synthetic SCITT/WIA validation annex remains supplemental evidence. Its six synthetic outcomes are not promoted to interoperability results. In particular, the late-arrival case distinguishes agent checkpoint ancestry from witness observation order; the fork case demonstrates conflicting authenticated commitments, not malicious intent or which branch is authoritative.¶
Author: Nicholas Michael Kornoelje. Affiliation: Independent Researcher. No co-author or independent reviewer is represented in this version. Independent technical review is requested, not claimed. The next acceptance gate is execution of the shared byte-level vectors by at least one independently developed implementation, which shows only that the vectors are implementable. An interoperability claim requires the condition in Section 17: at least two independently developed producers and one verifier. Both are followed by review of key lifecycle, duplicate-map-key handling, deterministic-CBOR edge cases, recovery, privacy, and threat-model assumptions.¶
This document is a proposal for technical review. It is not an RFC, IETF-endorsed specification, SCITT extension adopted by the IETF, security certification, or proof of production readiness. No IANA action is requested.¶