Skip to content

Append Only Evidence and Nostr Deletion

Status and recommendation

Design proposal, 2026-10-31. Neither the relay policy nor the protocol changes below are implemented by this note. No NIP amendment has been submitted or adopted.

Evaluate a dedicated, configurable Spurline evidence-relay mode in which author deletion requests cannot suppress accepted regular events. Use additional signed evidence to express corrections, retractions, revocations and reversals. OpenETR determines their consequential effects under explicit rules, independently of relay deletion and replacement.

Start with an advertised deployment policy, not a universal change to Nostr. Regular events also carry personal communications; applying evidence retention silently to a general-purpose relay would create privacy and compatibility risks. This policy is independent of signature algorithm and complements the optional PQ anchor proposal.

Separate three operations

Replacement introduces a new signed event for the same logical key or address. It does not edit an existing signed event. The NIP-01 replaceable/addressable conventions permit older versions to be discarded; they are not history-preserving archives. NIP-01

Deletion requests removal or suppression of referenced events. Current NIP-09 uses kind 5, references targets by e or a tags, and recommends same-author target suppression without a regular-event exemption. It recommends retaining deletion requests; deleting a deletion request does not undo it. NIP-09

Consequential state change is an application decision derived from evidence. Deleting an event or publishing a replacement is not, by itself, an authorized reversal of a transfer or discharge of an obligation.

OpenETR should distinguish:

  • Authenticity: whether the assertion was signed under the claimed key.
  • Availability and historical evidence: whether the exact assertion can be retrieved and independently verified, with explicit completeness limitations.
  • Consequential validity: what effect that assertion has under the applicable ruleset.

An authentic assertion can remain in history despite being mistaken, unauthorized, contradicted, expired or no longer effective. Retention does not endorse its truth.

Proposed relay policy

Event class Evidence-relay behavior
Regular Retain accepted events without honoring author-requested suppression
Replaceable Apply defined replacement and eligible deletion semantics; no promise of complete version history
Addressable Apply defined address-based replacement and eligible deletion semantics; no promise of complete version history
Ephemeral Do not persist; deletion is unnecessary
Unclassified or unsupported Apply an explicit admission rule; do not silently classify every remaining kind as regular

Use the existing NIP-01 classification rather than a per-event no_delete tag. Publish the classification/specification revision used by the implementation; future assignments must not silently change the retention promise for stored data. Special cases defined by individual NIPs still require review.

Recommended policy wording:

Within the configured evidence-relay policy, accepted regular events remain retrievable notwithstanding subsequent author deletion requests. Corrections, retractions, revocations and reversals are represented by additional signed evidence. Their effects are evaluated under application rules, not inferred from relay deletion or replacement.

“Append-only” here means a logical restriction on author-controlled protocol operations, not physically immutable storage or guaranteed eternal availability.

Admission and retention boundary

The operator decides which events to accept, subject to published size, resource, identity and content policies. An append-only policy does not imply unlimited storage or open publishing. For a dedicated evidence deployment, protect all accepted regular events consistently; narrow the admitted kinds if necessary rather than making undisclosed exceptions to retention.

Operators retain administrative removal powers for legal, privacy, security and operational reasons. Such removal changes availability, not OpenETR state. Where appropriate, record a minimal administrative audit entry without republishing the sensitive content or implying that a signed public removal notice is mandatory. Storage loss, outages and removal on other relays remain possible.

Advertise the mode before clients publish. Disclose any retention horizon, capacity-based eviction and migration policy; do not market a finite or evictable store as permanent. Changing modes must not silently apply historical tombstones and hide previously protected evidence. Require an explicit operator migration decision, backup and documented client impact.

Deletion requests are evidence, not automatic retractions

A kind-5 request may itself be accepted and retained as the author's statement about its targets. For protected regular targets, the request has no suppression effect. OpenETR must not infer that it constitutes an effective retraction unless the applicable ruleset explicitly assigns that effect and verifies authorization.

Evaluate each target separately in mixed-class requests. Derive its class and author from the verified target, not solely a request's claimed k tag. For an unknown target, retain a pending request if appropriate, but do not apply a blanket tombstone: when the target arrives, classify and authorize the action first.

Acknowledging storage of the request is not confirmation that every requested deletion occurred. Preserve normal EVENT acknowledgement meaning; consider a separate policy/status mechanism if clients need per-target outcomes. Do not invent a new standardized status code in this implementation.

OpenETR representation and reconstruction

Issuance, transfer, relinquishment, discharge and other consequential assertions should be represented by durable regular evidence events, not solely by mutable profiles or addressable records. Mutable records can act as discovery pointers or current-state summaries, but the evidence underlying a summary must be portable and independently verifiable by exact event ID.

A correction or reversal should identify its target evidence, statement type, applicable ruleset, authorized actor and scope. The ruleset determines whether it adds context, supersedes a claim, changes prospective state or has another defined effect. Authorization to issue the original statement does not automatically grant unilateral authority to undo all consequences relied on by other parties.

For example, after a transfer from A to B:

  1. A's subsequent deletion request does not remove the transfer from protected history.
  2. A's signed assertion that the transfer was mistaken becomes additional evidence.
  3. A verifier evaluates that assertion and any required counterparty/authority evidence under the transfer ruleset; it does not simply restore A's control.
  4. If essential evidence is missing or conflicts remain unresolved, the verifier reports an incomplete or contested result rather than inferring a clean history.

Portable evidence packages should carry exact signed events, required artifacts, digest algorithms and rule references. Package integrity does not prove that all relevant evidence has been included. Multi-relay retrieval improves availability, not completeness or consensus. Event absence never proves that an event did not exist. Self-reported timestamps do not supply trusted ordering by themselves.

Possible changes to NIP-09

Two routes should be evaluated separately:

  1. Specialized policy profile: describe how an evidence relay departs from current deletion recommendations, and how clients discover that policy. This is the recommended initial route and does not require claiming universal compliance with a new deletion rule.
  2. Protocol amendment: change the default eligibility rules in NIP-09 to restrict author-requested deletion to replaceable/addressable targets. This changes existing expectations and requires broader Nostr review and migration planning.

The following requirements are proposed amendment text, not current NIP rules:

A relay applying event-class-restricted deletion MUST NOT remove or suppress a regular event solely because its author published a deletion request. Author deletion eligibility is limited to replaceable and addressable targets, subject to target-author authorization. Ephemeral events require no retained deletion state. Undefined event classes require explicitly specified treatment.

If adopted universally, the conditional phrase would need removal and the change would become a new normative default. If scoped to a profile, its activation and discovery mechanism must be specified before interoperability is claimed.

An amendment would need to address these parts of NIP-09 explicitly:

Area Proposed change or clarification
General target rule Replace unrestricted same-author suppression guidance with class-based eligibility
e references Check the referenced event's actual class and author; pending requests cannot preemptively suppress regular targets
a references Preserve address/author validation and defined timestamp cutoff behavior for eligible mutable versions
k tags Treat as classification hints, not authority to override a target's actual kind
Mixed targets Apply eligibility independently; define request acceptance separately from per-target execution
Client rendering Separate local hide/mute preferences from evidence retention and consequential invalidation
Request retention Preserve the request as evidence and retain the existing no-undo rule for deletion of deletion requests
Operator powers Explicitly distinguish author protocol requests from administrative removal and retention expiry
Discovery Specify how clients learn the applied policy and limits without inventing a supported-NIP number
Transition Explain old tombstones, preexisting data, mode changes and communication with legacy clients

Keep deletion distinct from replacement: restricting NIP-09 would not make mutable version history durable under NIP-01. If historical versions of mutable records are also required, that is a separate archival policy or protocol proposal.

Existing clients may hide regular events as soon as they observe an authorized deletion request, even if the relay retains them. Evidence-aware clients must evaluate the relay policy and application rules and must not present a local hide decision as erasure of evidence. Users should still be free to hide unwanted content locally; the relay policy is not a command to display it.

No new wire field, kind, NIP number or relay-info schema is standardized here. Publication of policy metadata and clear user-facing retention notices should be prototyped before proposing a universal amendment. Retention disagreements across relays cannot be resolved merely by changing the text of a NIP.

Spurline implementation implications

The reviewed Spurline store records kind-5 e tombstones and hides targets with matching authors. It does not currently implement complete replaceable/addressable selection or address-based deletion. Therefore this proposal is more than a switch that disables one SQL predicate.

Before implementation:

  1. Define and test a shared event-class classifier and explicit undefined-kind behavior.
  2. Add opt-in settings and discoverable policy metadata; keep ordinary deployments unchanged.
  3. Apply eligibility to deletion recording, pending requests and query visibility, including the effect of tombstones already in the database.
  4. Implement and qualify replacement/addressable behavior before advertising the full table.
  5. Keep historical queries and live subscriptions consistent about visibility; distinguish new correction evidence from a request to erase cached copies.
  6. Keep administrative removal out of the public author-deletion path; define whether removed IDs are blocked from reimport and how the policy is audited.
  7. Apply the same validation and retention rules to synchronization, imports and alternate storage backends; none is a privileged bypass.

Required tests include regular-target retention; same-author and foreign-author requests; deletion before arrival; mixed classes; false k hints; addressable cutoff boundaries; replacement ordering/ties; ephemeral non-retention; request retention; mode migration with existing tombstones; administrative removal; backup/restore; classical/PQ parity; and reconstruction with missing or conflicting OpenETR evidence. Validate actual OpenETR rulesets separately before claiming their current implementations satisfy these requirements.

Guiding principle

Evidence may be superseded, contradicted or invalidated, but an author cannot retract its historical existence through relay protocol operations. Its consequential effect changes only under explicit rules; its continued availability remains a separate responsibility.

The architectural objective is independently verifiable consequential history, not dependence on mutable relay state or a guarantee that an operator can never remove stored bytes.