Skip to content

Relays as Accountable Witnesses

Policy brief | 31 October 2026

Policy proposition

A relay can be extended from a carrier of evidence into a producer of evidence. Instead of merely returning a signed record, it can issue its own signed statement that it observed and verified that particular signed record at a stated time.

For OpenETR, this would allow an Anchor Record to be accompanied by Witness Records from one or more relays. The testimony could be copied, archived, discovered, and independently verified without requiring the original witness to serve every subsequent inquiry.

The proposition is not that relays should decide what a record means. It is that they can make their observations attributable and portable, while the applicable ruleset and relying party's recognition policy determine what follows from those observations.

From availability to testimony

An ordinary relay answers a retrieval question: can it return this event now? A witnessing relay makes an additional assertion: it observed this exact signed event at a particular time and successfully checked its identifier and signature. That assertion is itself signed evidence.

This matters because the author's own timestamp does not provide independent evidence of when its signed assertion existed. A recognized witness can provide an additional time claim without requiring every participant to adopt a common application or entrust the relay with the underlying transaction's meaning.

For example, a publisher anchors a product document and later issues a caution notice. Witness Records can provide attributable observations of each signed event. They do not establish that the product is safe, that the publisher had authority, or that every possible notice has been found.

Separate responsibilities

Participant Responsibility
Author Makes and signs the underlying assertion
Witness relay Attests to observing and verifying that signed assertion
Commissioning authority Authorizes the service key to witness under a stated policy
Relying party Decides which authorities and time assertions it recognizes

The commissioning relationship supplies a basis for accountability. It does not make every commissioned relay universally trusted. Authorization to operate a relay and authorization to provide observation timestamps are separate roles.

A relying party may recognize an operator for witness commissioning while declining a particular clock policy or deployment. It can also distinguish cryptographic validity from trust: a receipt may be correctly signed yet provide no accepted time conclusion under that party's policy.

Portable testimony without a central adjudicator

Because the observation is a signed record, a different relay or archive can carry it without becoming its author. An evidence package can preserve the original event, its Witness Records, the commissioning and authorization chain, and the applicable policy.

This separates the source of testimony from its distribution. A verifier with sufficient supporting evidence can check the witness's signed claim without contacting that witness. Current knowledge about compromise or revocation may still require additional evidence; portability is not a guarantee of completeness.

Several relays can witness the same event. Their observations may strengthen confidence when the witnesses are recognized and meaningfully independent. Several hosting copies of one receipt, however, are not several witnesses. Even separate keys may share one operator or clock source.

This is a model of attributable observations, not an automatic consensus mechanism. Witness counts must not decide ownership, validate an assertion's contents, or choose among conflicting OpenETR transitions.

Security and time boundaries

A Witness Record must bind the actual subject signature as well as the event identifier. Otherwise, an earlier commitment to unsigned contents could be misrepresented as proof that a signature produced later already existed.

Even a properly bound receipt relies on the witness's clock, operating practices, and signing-key protection. A compromised witness key could produce backdated assertions. Explicit authorization, disclosed clock uncertainty, incident procedures, and independent supporting evidence are therefore central to the trust model, not optional implementation details.

The intended time conclusion is bounded: under an accepted witness policy, the signed event existed no later than the upper end of its attested observation interval. That is not its exact creation time or proof of first publication.

Witnessing also creates privacy obligations for deployment design. Public references can reveal relationships even when event content is not repeated. Operators should disclose the policy, limit eligible records, and specify retention and removal boundaries rather than silently witness all relay traffic.

Learning from NIP-03

NIP-03, the OpenTimestamps event-attestation proposal, is currently marked unrecommended with a vulnerability warning, not formally labeled deprecated (checked 31 October 2026). Its proof targets an event ID.

The important design lesson is that committing to contents and establishing the existence of a signature are different claims. Nostr event IDs exclude the signature. In this design's threat analysis, an attacker could precommit unsigned contents and sign them after a later key compromise; the earlier commitment would not prove that the signature existed then.

Our convention binds the exact verified signature as well as the ID. That closes this binding gap, but does not eliminate trust in the witness. A dishonest or compromised witness can still claim a false observation time. Relay receipts are not OpenTimestamps proofs and are not presented as an approved replacement for NIP-03. The design note separates these security and standards boundaries in detail.

Relationship to OpenETR

Core Record Ruleset 1.0 expressly leaves external time evidence outside its scope. Witness Records should initially be an optional supplementary assessment alongside the existing anchor and notice evaluation, not a silent amendment to that ruleset.

Applications could display an Anchor Record and a list of associated Witness Records, showing who witnessed it, who commissioned the witness, the claimed observation time and uncertainty, and the relying party's trust decision. No receipt found should mean only that no receipt was found in the queried sources.

This complements evidence retention. Retention helps keep assertions available; witnessing adds attributable observations about their signed existence. Neither alone establishes truth, legal effect, or global historical completeness.

An extension by convention, not yet a standard

This is an optional extension to both relay capabilities and OpenETR applications. It uses existing Nostr signed events and ordinary queries: no new transport, subscription mechanism or event envelope is required. The relay becomes an additional author of evidence, and participating applications understand its receipt schema. Other clients need not interpret it. Ordinary relay admission and retention limits still apply.

The new semantics are currently an experimental interoperability convention, not an approved NIP or normative OpenETR requirement. In particular, kind 1418 is not a reserved allocation. The convention does not amend NIP-09 or make append-only retention a prerequisite, and it does not change Core Record Ruleset 1.0. Protocol compatibility should not be confused with universal support.

The convention could inform future standards on both sides: a relay-facing NIP for observation receipts and an optional OpenETR specification for their discovery and assessment. That path requires independent implementations, security and privacy review, shared test vectors, agreement on allocation and versioning, and an explicit migration policy. Any future ruleset that gives witness evidence consequential effect must separately define that dependency. Deployment is evidence that an approach can work, not evidence of standards approval or universal trust.

Implementation path

Spurline's local implementation supports commissioning, witness authorization, signed renewal and revocation, and local suspension. Its frozen experimental receipt profile uses kind 1418, with wire-validation code and a cryptographic fixture; this is not a registered allocation. Automatic issuance is now implemented in Spurline. Oakgate discovers and displays receipts after checking signatures and binding to the exact signed anchor or notice, separately from the core ruleset result. Full application witness-trust evaluation remains outstanding.

The deployed version-3 profile uses the relay's system clock with explicitly unspecified accuracy. This supplies an independent, attributable observation claim, not a bounded-accuracy timestamp guarantee. A receipt's valid signature does not itself establish commissioning, current authorization or truthful time.

The next steps are to test the experimental receipt profile across applications, exercise durable observation and publication, assess compromise handling, and add witness-policy assessment to applications. Tests should cover multiple witnesses, duplicate copies, signature substitution, missing authorization, clock failures, and partial evidence retrieval.

The architectural opportunity is clear: relays can carry both assertions and portable testimony about those assertions, without becoming the authority over their meaning.

Supporting design

The Witness Record Specification (draft 0.1) proposes interoperable requirements based on the experimental convention. Kind 1418 remains provisional; publication does not imply standards adoption or certification of current implementations.

The Relay Witness Records design note describes discovery, verification, trust boundaries, and integration milestones. Spurline's operator runbook and application guide provide the companion operational and record-format details.