Przejdź do treści

Solution 011: Whisper — Social Signal Exchange — Implementation Guidelines

(Solution 011-whisper; based on Proposal 013-whisper-social-signal-exchange.)

Proposal: doc/project/40-proposals/013-whisper-social-signal-exchange.md

Cross-federation policy reference: doc/project/40-proposals/079-cross-federation-alliance.md

Solution: doc/project/60-solutions/011-whisper/011-whisper.md

Capability catalog: doc/project/60-solutions/011-whisper/011-whisper-caps.edn

Related impl notes:

  • doc/project/60-solutions/008-agora/008-agora-topic-addressed-relay-impl.md (carrier for public whispers)
  • doc/project/60-solutions/017-inter-node-artifact-channel/017-inter-node-artifact-channel-impl.md (carrier for private whispers)

Purpose of this document

This note is the implementation entry point for Orbiplex Whisper. It does not duplicate solution-level responsibilities (that lives in whisper.md), the per-capability status catalog (that lives in whisper-caps.edn), or the fine-grained backlog (that lives in the sibling node repository alongside the other Whisper implementation notes).

It exists to:

  • name the layers that have to be built,
  • fix the invariants that hold across layers (envelope-vs-content split, signing domain, pseudonymous authorship semantics, disclosure/scope distribution rule),
  • track current status per layer so the next implementer knows where to start,
  • enumerate the open decisions that must be resolved before implementation can meaningfully proceed,
  • give a stable commit order when shaping the whisper stack,
  • own the post-M4 productization tracker formerly kept as a workspace-root draft note.

Current implementation status

This implementation note started as a layer-by-layer build plan. The layer tables below are useful for historical sequencing, but the authoritative current status is:

  • whisper-signal.v1, whisper-threshold-reached.v1, and association-room-proposal.v1 are implemented for the Agora M4 public signal path.
  • whisper-trace.v1 is implemented as a strict content contract with positive and negative fixtures, runtime validation, sender-host consent resolution, authoring and inspection, ingress registration, metadata-only read models, and public Agora plus private AD/INAC carrier acceptance.
  • Public/federated Whisper records are carried as agora-record.v1 envelopes with content/schema = "whisper-signal.v1".
  • Public Whisper authorship is nym-authored at the envelope boundary: author/participant-id = nym:did:key:... requires verifier-usable author/nym-proof; private-correlation and direct-only signals are rejected on public Agora topics.
  • whisper-intake exists as a bounded supervised Rust HTTP middleware with its own operator UI surface, local-private SQLite intake store, host-authenticated loopback API, sealed personal Memarium sync for raw/draft/quarantine/candidate stages, and manual retry for private sync.
  • whisper.redaction.prepare is the stable redaction/paraphrase capability boundary. The first provider path is deterministic and local: JSON-e Flow calls sensorium.directive.invoke, and Sensorium OS runs a bounded whisper.redaction.prepare action. The provider returns a draft for human review; it never publishes to Agora and never marks a candidate approved.
  • Node C's Agora projection can derive deterministic threshold/proposal state from two eligible public signals, sign derived records through the host signer, ingest them through the ordinary Agora path, and prevent derived-record loops.
  • Story-005 now has a three-node laptop operator pack and English/Polish runbooks that exercise A/B Whisper Intake plus C Agora projection. The hard-MVP readiness snapshot treats this Story-005/M4 path as complete.

Post-M4 productization is tracked below in this document. It covers richer external Inquirium/model-runtime redaction policy, outbound privacy realization, production correlation, association-room lifecycle, production Monus/Sensorium source integration, and explicit public-gossip promotion behind the same Whisper/Agora envelope boundary.

Migration: signal/polarity enum rename

Date: 2026-06-28.

whisper-signal.v1 now uses signal/polarity = "idea" instead of the earlier draft value signal/polarity = "inspiration". This is a pre-compatibility wire-contract rename, not a runtime migration with a grace period. Producers MUST emit idea; schema gates MAY reject new inspiration records.

Existing historical records are not rewritten by this solution. Any replay or import path that intentionally consumes old draft records must normalize inspiration -> idea before validation, or handle those records as legacy data outside the current public Whisper contract.

Post-M4 productization invariants

These invariants apply to every later productization slice. They are stricter than the current M4 smoke because the M4 smoke proves only the minimal public correlation path.

  1. Host-owned authority. Whisper may request redaction, routing, correlation, or publication, but host policy decides provider selection, privacy resolution, projection authority, and final publication authority. Model/runtime selection goes through Inquirium/model-runtime routing, not a caller-supplied raw model name. Outbound privacy is resolved by host policy, not by a caller naming a concrete relay.
  2. Facts before views. Meaningful transitions should leave append-only facts before any read-model or UI view is updated: raw intake accepted, redaction requested, draft produced, candidate approved/rejected/quarantined, privacy resolution attempted, signal published/refused, threshold observed, room proposal created, participant invited/accepted/rejected/left, and public-gossip promotion proposed/approved/published/refused.
  3. Privacy is a policy boundary. routing/profile, routing/failure-mode, disclosure/scope, and classification are not decorative metadata. Private-correlation and direct-only signals must not reach public Agora topics. hard-fail + onion-relayed must refuse when no onion-capable path exists. Providers may draft or diagnose, but may not approve, publish, or widen disclosure scope.
  4. Preflight decision binds to the exact candidate bytes (no TOCTOU). whisper-intake computes a canonical candidate_digest for the candidate being preflighted, checks the existing private preflight-fact count before appending another fact, stores that digest in the private preflight fact, and requires any operator downgrade acknowledgement to carry the same digest. Publish re-checks that the signed Agora record content still matches the preflighted digest; a mismatch voids the ack and refuses publication. Relay evidence is point-in-time: publish re-resolves agora.relay evidence and re-computes the preflight from the current candidate rather than trusting a stale Route decision. Expiring or one-time downgrade ACK tokens are post-MVP hardening for multi-console replay resistance; hard-MVP binds ACK scope to candidate bytes and preserves retryability for the same candidate digest.
  5. Empty relay evidence is a valid fail-closed input. When no passport-scoped agora.relay evidence is discoverable, relayed/onion-relayed with hard-fail MUST refuse and with soft-fail MUST downgrade visibly — the preflight ships without any relay infrastructure, it does not silently allow.
  6. The refuse path also leaves a fact. Per Facts before views, a refused publish records a privacy resolution attempted / signal refused fact with the reason, not just a blocked call.
  7. Domain capability before provider mechanism. The caller-facing redaction contract remains whisper.redaction.prepare. JSON-e Flow, Sensorium OS, Inquirium/model-runtime, simulator adapters, and remote model adapters are provider mechanisms behind that boundary. Neutral adapter output must be translated back into whisper-redaction-prepare-response.v1 before Whisper consumes it.
  8. Deterministic smoke, richer production. M4 deterministic matching remains the replayable baseline. Production correlation may add semantic features, trust-tier diversity, source-class weighting, and richer threshold policy, but every threshold decision must remain reconstructable from source facts and an explicit policy version.

Post-M4 productization dependency map

The slice names are topical, not a strict build order.

  • Already closed for M4: deterministic public/federated correlation over topic/class + signal/similarity-key, projection-authority threshold/proposal emission, the secretless Inquirium simulator adapter, and the Story-005 local profile acceptance bridge.
  • Hard prerequisite for model-backed redaction providers: daemon inquirium.generate must invoke HTTP adapter instances through host-owned runtime/ref and model.binding/ref. This is implemented for the simulator and remote-provider adapter seam.
  • Independent contract tracks: redaction provider policy, outbound privacy resolver data, semantic correlation policy data, source-class validation, and public-gossip promotion facts may evolve independently as long as they preserve the invariants above.
  • Strict runtime ordering: association-room lifecycle runtime must exist before public-gossip promotion from room/case state can be productized. Monus/Sensorium source-class carry-through must exist before correlation can weight or exclude those sources.
  • Continuous hygiene: any schema-visible source/evidence field must be mirrored into node/protocol/contracts/schemas and schema-gated before code starts enforcing it. Tracker status changes must update this document, the Node implementation ledger, and the MVP readiness snapshot when readiness interpretation changes.

Matrix transport continuation (P035-004)

P035-004 owns the opaque Agora carrier; Whisper consumes it. The reviewed 2026-10-12 matrix-opaque-envelope profile is qualified at clean Node 20c4c16c and Orbidocs 862586f: four SQLite and 21 Matrix cases. It qualifies two independent inline-proof whisper-signal.v1 nym signals with confidence 0.5, whisper-trace.v1 regression, and host-signed whisper-threshold-reached.v1, followed by restart and deterministic first-source carrier replay without forwarding. Exact signed author bytes remain unchanged; admission and proof/disclosure refusals retain their authority. effective_weight is a gossip projection result, not a threshold field.

Every Node-created full PDU fits the protocol bound (maximum 50,043 bytes). Full-PDU enforcement is a separate environment result for the disclosed pinned Conduit/Ruma fixture; direct boundary events reach 65,534 bytes, with actual refusal codes retained. Initial v0.7.0 oversize evidence and the qualifier refusal remain unchanged. Retained qualified evidence is node:docs/evidence/matrix-opaque-envelope/2026-10-12-review/qualification.json. The frozen run selected v2 explicitly under the former default-v1 policy. The accepted write policy now defaults to v2 with explicit legacy v1. P035-006 owns the separately implemented and qualified Node-default change below; the policy decision alone did not add Whisper runtime evidence.

These qualified families do not close all Whisper, nym issuance, production key custody, external redaction, Anon egress or inter-homeserver federation. P035-005 independently qualified official unmodified Synapse v1.162.0/PostgreSQL 16.13 room v10 on 2026-10-13 at clean Node dcc8133c / Orbidocs c84df99: four SQLite and 22 Matrix cases. The same permitted signal confidence 0.5, trace and host-signed threshold families preserve exact bytes through restart and first-source replay. Both gates pass: 42 Node full PDUs, maximum 49,927 within 65,536; separate direct protocol maximum 65,535 and actual HTTP 413 M_TOO_LARGE below the request-body ceiling. Full stock serialization verifies hashes, server signatures and event IDs. Homeserver/database restart, real limited-sync backfill and late room binding are covered. Original failures and Conduit evidence stay frozen. Qualification: node:docs/evidence/matrix-opaque-envelope/2026-10-13-synapse/qualification.json. This remains explicit v2; P035-006 and the other Whisper exclusions above remain independent.

P035-006 separately qualifies the omitted-config default-v2 path on 2026-10-14 at Node 42fde666 / Orbidocs 4ecec76, with five SQLite and 27 Matrix cases on stock Synapse room v10. The same permitted signal confidence 0.5, trace and host-signed threshold families preserve author bytes through restart and source replay. All 692 Node full PDUs fit 65,536 (maximum 49,927); the independent homeserver gate passes. First-intent, filtered history/bootstrap and archive equivalence controls also pass. Evidence: node:docs/evidence/matrix-opaque-envelope/2026-10-14-default-v2/README.md. This transport default does not extend the qualified Whisper families, nym issuance, key custody, Anon, federation or production scope described above.

Post-M4 productization tracker

Status values:

  • done — implemented and covered by tests, schema validation, or documented operator evidence;
  • partial — contract or primitive exists, but runtime/UI/product behavior is incomplete;
  • not-started — not yet implemented beyond conceptual documentation;
  • deferred — intentionally postponed beyond the current productization pass.

Redaction provider policy and simulator bridge

Work item Status Evidence / remaining work
Keep whisper.redaction.prepare as the domain boundary above provider mechanisms done Solution text and Story-005 docs preserve the boundary; adapters do not become publication authorities.
Host-owned provider policy for local deterministic, local model, supervised HTTP adapter, and external model adapter classes partial Policy shape is documented and the simulator/remote adapter substrate exists; full operator-facing provider policy UX and remote deployment profiles remain later work.
Inquirium/model-runtime provider selection by runtime/ref or profile, not raw caller model name done Daemon inquirium.generate routes through host-owned runtime/model-binding context and rejects model override.
Translate neutral adapter or GenerateResponse output into whisper-redaction-prepare-response.v1 partial Boundary requirement is documented; deterministic provider path is implemented. Full model-backed redaction bridge for production providers remains a later slice.
Metadata-only diagnostics for provider calls partial Inquirium trace records are metadata-only; Whisper-specific provider-call diagnostic views remain product work.
JSON-e Flow inference grants for inquirium.generate provider bridges done Host capability path fails closed without explicit inference grants.
Secretless supervised Inquirium simulator adapter done Story-005 simulator is opt-in, egress-free, exposes simulated bindings, rejects unknown bindings, and is tested through daemon Host API.
Story-005 simulator local profile and acceptance checks done Local profile enables the simulator explicitly, verifies routability, generation, fail-closed unknown binding, stop/start non-routability, and restoration.

Anon / outbound privacy realization

Work item Status Evidence / remaining work
Host-owned outbound privacy resolver contract done whisper-core parses routing failure mode, relay constraints, forwarding hops, and forwarding budget; whisper-intake now consumes the resolver before public/private publish.
Represent direct, relayed, and onion-relayed routability states done whisper-core carries typed posture data.
Fail closed for hard-fail + onion-relayed without an onion-capable path done Core contract and tests cover refusal semantics.
Relay capability discovery based on capability evidence partial whisper-intake now reads host-owned agora.relay capability provider evidence for publish preflight and requires passport-scoped relay class data rather than deriving classes from module labels. Federation-scale relay discovery and Anon relay routing remain Node/Anon runtime work.
Derived forwarding nym scope, TTL, replay, and idempotency rules partial Scope contract exists in the Anon/Whisper implementation ledger row; concrete relay runtime consumption remains partial.
First derived-nym relay transport path for Whisper egress not-started Required post-MVP Anon runtime work.
Operator/user visibility for privacy downgrade before publication partial whisper-intake returns operator-visible preflight reasons, queries existing preflight facts through host memarium.read, enforces a per-intake preflight fact limit, writes a private preflight fact through host memarium.write, binds downgrade acknowledgement to the canonical candidate digest, validates host signing/write responses, and requires explicit downgrade acknowledgement before publish. Richer UI plus expiring/one-time ACK tokens remain product hardening.
Acceptance proving soft-fail and hard-fail egress behavior end to end partial Core behavior and whisper-intake preflight behavior are covered; full Anon relay runtime acceptance remains pending.

Production correlation policy

Work item Status Evidence / remaining work
Keep M4 deterministic rule as correlation_policy = deterministic-m4 done M4 projection uses deterministic topic/class + signal/similarity-key.
Policy fact schema for richer threshold configuration partial Production-shaped policy primitives exist; durable policy-fact lifecycle remains product work.
Trust-tier diversity option partial Schema fields expose trust summaries/tiers; runtime threshold weighting remains pending.
Semantic feature extraction contract without raw rumor-text exposure not-started Needs explicit feature/digest contract and provider path.
Treat embeddings as derived sensitive content with bounded persistence not-started Must be defined with Inquirium embedding/data-plane policy.
Semantic correlation explainability surface not-started Needs threshold explanation read model and operator diagnostics.
Source-class weighting and exclusion rules partial source/class is carried and parsed; weighting/exclusion policy is not productized.
Version policy used by every threshold decision partial Deterministic M4 is replayable; richer policy-version facts remain pending.
Replay-equivalence tests across policy versions not-started Required once richer policy facts exist.

Association-room lifecycle

Association-room transport and replication decision

Status: accepted for post-MVP implementation.

An association room MUST NOT be modeled as a room that lives on one Agora server, Matrix room, or any other single transport host. The room's source of truth is a verified, append-only room event log. Local room state is a projection rebuilt from signed room events that may be fetched from multiple replicas.

Implications:

  • Association-room servers, Agora relays, Matrix mailbox rooms, and object-store fetch surfaces are carriers or caches, not semantic authorities.
  • Future room artifacts should be shaped as signed events such as association-room-event.v1, association-room-message.v1, association-room-membership.v1, and association-room-moderation-event.v1.
  • Room messages and lifecycle events should be transported through Artifact Delivery so the same idempotency, retry, admission, route policy, and operator trace machinery applies to room traffic.
  • Multi-Agora availability should be achieved by fanout to several Agora relay targets and by read-side merge of verified events from several replicas. A relay outage must not redefine room state.
  • A replica may advertise bounded retention, for example "last N days" or "last N events". It MUST expose a compact retention/status view that tells clients where the retained log starts and how many room events are currently available, so operators can distinguish a complete room history from a bounded window.
  • Privacy remains private-correlation: Agora-visible room data should be encrypted envelopes, opaque refs, or minimal metadata. Plain room transcripts do not belong on public Agora topics.
  • Membership and moderation consequences must be derived from signed room events plus current membership/capability policy, not from the server that happened to deliver an event.
  • Events that continue an exchange should carry both the predecessor event id and the predecessor digest. This is a per-thread or per-lineage integrity link, not a global linear blockchain for the whole room. It lets a receiver prove that a reply chain is complete within the retained window or up to a known checkpoint, and detect missing, substituted, or reordered predecessor events. If a predecessor falls before a replica's retention boundary, the replica should report the lineage as pruned rather than complete.

The initial implementation may start with best-effort fanout and local retry. Stronger k-of-n write visibility, per-epoch room keys, and richer replica selection are later hardening layers.

Work item Status Evidence / remaining work
Decide module ownership: Whisper proposal semantics vs dedicated association runtime partial Whisper owns proposal semantics; agora-projections now owns the first local association-room read model and lifecycle fact seed. Dedicated room/case-management runtime remains later work.
Define room lifecycle facts partial association-room-proposal.v1 exists; Node now persists bounded local association_room_lifecycle_facts for room state transitions with FK-backed proposal references. Richer room/case facts remain pending.
Invite/accept/reject/expire transitions partial Node now exposes minimal operator-driven transitions using the whisper-core FSM, with authenticated actor binding and path/request bounds on the operator control route. Enrollment UX and expiration scheduler remain pending.
Define association-room signed event log and replica model not-started Decision accepted above: room state is a projection over verified signed events from multiple carriers, not a server-hosted room. Event schemas and runtime are pending.
Transport room events through Artifact Delivery not-started Required so association-room traffic reuses AD idempotency, retry, route policy, admission, and trace machinery.
Multi-Agora fanout and read-side merge not-started Required to avoid a single Agora relay as room availability/source-of-truth bottleneck.
Replica retention and log-range status not-started Replicas may keep bounded windows such as last N days or last N events, but must expose retained-log start and event-count status so clients can reason about completeness.
Per-thread predecessor id + digest links not-started Message/continuation events should link to predecessor event id and digest for lineage integrity without imposing one global room chain. Completeness is scoped to retained windows or explicit checkpoints; older missing predecessors must be reported as pruned.
Moderation/witness policy data not-started Required room/runtime policy work.
Room transcript/storage classification policy not-started Required before real case-management rooms.
Operator/user UI for opt-in enrollment partial Operator HTTP API can transition proposals into rooms; full user enrollment UX is not built.
Close/leave/retention semantics partial The FSM supports close/leave and Node persists transitions; retention policy and room transcript storage remain pending.
Acceptance proving no automatic human enrollment from threshold/proposal done M4 threshold/proposal smoke stops at proposal state.

Monus/Sensorium production sources

Work item Status Evidence / remaining work
Source-class validation in whisper-core done direct-user, pod-user, operator-observed, derived-local, monus-derived, and monus-sensorium-derived are parsed.
monus-derived draft submission contract not-started Monus remains planned/partial; no production handoff runtime.
monus-sensorium-derived evidence requirement for Sensorium host capability at author time partial whisper-intake now requires a source evidence ref before publishing monus-sensorium-derived candidates. Live Sensorium evidence verification remains pending.
Protocol-visible evidence refs, schema sync, and schema-gate fixtures if needed not-started Add only when evidence refs become wire-visible.
Stricter review defaults for machine-derived signals not-started Needs Whisper intake policy/UI work.
Help-mode/emergency diversion policy partial whisper-intake now fails closed for Monus-derived candidates that indicate help-mode diversion. Full emergency/help workflow remains pending.
Preserve source class through candidate and public signal metadata partial Content schema/core can carry it; complete source integrations remain pending.
UI indicators for machine-derived and sensor-informed drafts not-started Product/UI work.

Public-gossip promotion

Work item Status Evidence / remaining work
Promotion proposal fact from room/case state to public-gossip.v1 draft partial Node now stores explicit public-gossip promotion drafts from accepted association rooms with bounded opaque lineage refs. Publication remains a separate act.
Approval policy: participant quorum, moderator approval, or local governance policy not-started Required before promotion runtime.
Final redaction review at promotion time not-started Required before public-gossip.v1 publication from rooms.
Publish public-gossip.v1 only as a separate explicit act done Story/Solution semantics forbid automatic threshold/proposal promotion.
Lineage refs to threshold/proposal/room decision without leaking private rumor text partial Promotion drafts now persist opaque lineage refs. Schema-level public promotion artifacts and tombstone semantics remain pending.
Refusal, withdrawal, and tombstone semantics not-started Required public-gossip lifecycle work.
Acceptance proving threshold/proposal does not automatically emit public gossip done M4 smoke asserts threshold/proposal stops below public gossip.

Artifact trace statements

Work item Status Evidence / remaining work
Separate whisper-trace.v1 from signal/polarity done The trace is a separate content contract and record kind; problem and idea remain social-signal polarities.
Digest-only and consent-bound inline shapes done Canonical schema and positive/negative examples cover dispatch, held-artifact, forbidden inline-without-consent, forbidden content in digest-only mode, and routing hop bounds; inline content is capped at 32 KiB, the extensions namespace at 8 KiB, and the full trace JSON at 128 KiB.
Exact inline digest and byte-length verification done whisper-core transiently decodes bounded inline bytes, enforces the 32 KiB cap, and rejects digest or size mismatches before signing or admission; read models retain metadata only.
Digest privacy and classification preflight done Pure validation emits stable digest-linkable and digest-may-be-guessable findings. Authoring and inspection surfaces expose those findings without treating digest-only disclosure as anonymity.
Agora ingress schema/disclosure gate done schema-gate registers the trace family; Agora accepts only the conventional public topic and public/federated posture, rejects private traces, and projects metadata without feeding signal threshold/association state.
Sender-host consent authority done Inline authoring resolves an active one-time operator consent bound to the exact canonical intent digest, subject, capability, requester, scope, and expiry; the opaque consent/ref alone grants no authority.
Operator/user trace authoring and inspection UI done Whisper Intake exposes consent and publish forms with digest-only defaults plus bounded metadata list/detail views. Raw inline content is transient and absent from the read model.
Idempotent authoring and bounded local retention done SQLite atomically reserves each (idempotency/key, request digest), conflicts while a publish is active, replays completed results, permits retry only after failure, and prevents one-time consent reuse. Metadata-only trace rows are pruned under a 30-day default retention window at startup and on writes.
AD/INAC private trace acceptor smoke done Story-005 sends a private/direct trace through the existing AD/INAC carrier, verifies the signed Agora envelope and exact topic/scope, asserts inac-direct rather than local fallback, and proves the private record is absent from the Agora projection. The public trace analogously asserts agora-publish; the same smoke guards the pre-existing private signal path.

Implementation order is contract-first and refusal-first:

  1. add pure base64 decode, decoded-size, and SHA-256 verification without I/O;
  2. resolve sender-host consent and classification policy for the exact candidate;
  3. register record/kind = "whisper-trace" and content/schema = "whisper-trace.v1" at Agora authoring/ingress boundaries;
  4. add the metadata-only read model while keeping inline bytes transient;
  5. bind private/direct acceptance to the existing AD/INAC carrier;
  6. expose bounded authoring/inspection UI and run public plus private carrier smoke tests.

Architectural posture

Whisper is not a transport; it is a content family plus an authoring/reception pipeline that rides on two distribution surfaces:

  1. Agora (proposal 035) — for disclosure/scope values other than private-correlation; topic convention: ai.orbiplex.whispers/<topic-class> (proposal 046).
  2. INAC (proposal 042) — for private-correlation, for routing/profile = direct, and for any scope the participant prefers to keep off a public substrate.

The wire format on both surfaces is the same signed agora-record.v1 envelope; only the transport and the addressable scope differ. The whisper-signal.v1 schema validates only the content body — identity, authorship, timing, and routing key live in the envelope.

Invariants that cross layers

These must be identical wherever they appear. If they drift between layers, pseudonymous authorship, cross-surface verification, or threshold detection will silently diverge.

  1. Envelope-vs-content split.
  2. whisper-signal.v1 validates the content body only. required starts with the const discriminator schema: "whisper-signal.v1".
  3. All identity, authorship signature, nym certificate reference, authoring timestamp, record id, and topic routing key are envelope concerns (proposal 035 §3).
  4. Any implementation that re-introduces signal/id, created-at, sender/*, rumor/nym, auth/nym-certificate, or auth/nym-signature into the content body is regressing this split and MUST be rejected at review.

  5. Signing domain. AGORA_RECORD_DOMAIN_V1 = "agora.record.v1" is the sole signing domain for whisper envelopes. The signing key is the author's when author/participant-id begins with participant:did:key:…, and the nym's when it begins with nym:did:key:…. No separate whisper-specific signing domain.

  6. Pseudonymity is an envelope property. Pseudonymous whispers are expressed in the envelope through:

  7. author/participant-id = "nym:did:key:…",

  8. author/nym-certificate-ref pointing at a nym-certificate.v1,
  9. envelope signature made by the nym private key.

The attestation-gate (proposal 041 §5) is the single authority that validates nym certificates and enforces council/scope policy. Whisper code does not re-implement this check.

  1. disclosure/scope as distribution-surface selector.
  2. private-correlation → INAC direct exchange.
  3. federation-scoped, cross-federation, public-aggregate-only → Agora.
  4. Public Agora deployments SHOULD refuse private-correlation at ingest (proposal 013 §Distribution note on intentional SHOULD); closed federations MAY relax. The rule lives in each relay's ingest policy, not in whisper code.

  5. Transport-identifier separation. sender/node-id and sender/federation-id are peer-session concerns; they MUST NOT leak into the signed content body or the envelope. Transport identity and authored identity are distinct axes.

  6. Byte-identical envelopes across surfaces. A whisper authored for INAC delivery and later promoted to Agora (threshold crossing, §Layer 7) MUST carry the same envelope bytes. No re-signing on surface migration; record/id is stable.

Historical build plan (Layers 0–8) — NOT authoritative for current status

Read this before trusting any status below. The per-layer tables in Layers 0–8, the "Open decisions blocking implementation", and the "Recommended commit order" are the original greenfield build plan kept for historical sequencing. Their ❌ not started / 🟡 / "new crate" markers are stale and contradict the current status and the post-M4 tracker above, which are authoritative. Examples of layer statuses that are now wrong: whisper-core exists (it is not a "new crate" suggestion and is consumed by the privacy preflight), whisper-signal.v1 is done, the threshold/proposal path and projection-authority emission are implemented, whisper-intake is a built supervised middleware, and the outbound-privacy resolver is wired. Do not re-plan or re-decide from this section; use the Current implementation status and Post-M4 productization tracker sections as the source of truth. This section should be pruned or folded into the tracker on the next pass; until then, treat every status marker here as historical only.

Layer 0 — JSON Schemas

Canonical schemas (historical shape — inline statuses removed as stale; this lists schema shape and purpose, not build state, which lives in the Post-M4 tracker above):

  • whisper-signal.v1 — content body (content-body-only after proposal 013 Next Action #7).
  • whisper-interest.v1 — interest registration artifact for threshold detection (fields and lifecycle are an open decision, see §Open decisions).
  • whisper-threshold-reached.v1 — aggregate notice after threshold crossing.
  • association-room-proposal.v1 — bootstrap artifact for opt-in association rooms.
  • nym-certificate.v1 — reused from proposal 015.
  • agora-record.v1 — reused from proposal 035.

Publication: schemas are served through the daemon's schema surface (same style as key-delegation). Schema names are stable; surface (Agora vs INAC) is not encoded in the schema id.

Status

Per-component status table removed as stale — see the Post-M4 productization tracker above for authoritative status.

Layer 1 — Kind registration

Whisper must be registered as a first-class kind in every substrate that carries or stores it.

(Status columns removed as stale — see the Post-M4 tracker above.)

Where What to register
Agora kinds table (proposal 035 §3) whisper-signal.v1 + topic convention for ai.orbiplex.whispers/<topic-class>
Agora kinds table whisper-durable.v1 for threshold-promoted whispers
Artifact Delivery inbound acceptor for INAC-carried private/direct artifacts whisper-signal.v1 acceptor + authorization modes allowed
Memarium observe rules capture envelope byte-identically + action-trace for whisper lifecycle

Layer 2 — Envelope builder and canonicalization

One authoring primitive that produces a signed agora-record.v1 envelope from validated whisper content:

build_whisper_envelope(
  content: WhisperSignalV1,            // schema-validated
  author: ParticipantOrNym,
  nym_cert_ref: Option<NymCertRef>,    // required when author is Nym
  topic_key: Option<String>,           // Some("ai.orbiplex.whispers/<topic-class>") for Agora, Some("private/<name>") for INAC
  authored_at: DateTime<Utc>,
  signer: SigningService,              // proposal 037
) -> Result<AgoraRecord, WhisperEnvelopeError>

Responsibilities:

  • JCS canonicalization of the unsigned envelope,
  • record/id = sha256:<base64url-no-pad(sha256(canonical_bytes))>,
  • signature in agora.record.v1 domain using the author's or nym's private key via the signing service,
  • verification that the signer's returned key_public binds to author/participant-id (same hard-check rule as agora-core sign adapter, proposal 035 impl §Layer 1),
  • population of record/id, signature.value, and author/nym-certificate-ref when applicable.

Crate placement suggestion: node/whisper-core (new crate), modeled on agora-core's sign adapter but specialized for whisper content validation and the nym-or-participant authorship choice.

Status

Per-component status table removed as stale — see the Post-M4 productization tracker above for authoritative status. (Note: the canonicalization, record/id, and signing responsibilities described here are realized in the whisper-core crate; the "golden vectors" testing note still stands as future work.)

Layer 3 — Authoring flow (whisper-core)

The authoring pipeline turns raw rumor input into a publish-ready envelope. Stages:

  1. Intake — from a direct user, pod-user, operator observation, derived-local producer, Monus, or Monus+Sensorium (mapped to source/class enum).
  2. Redaction and idiolect flattening — bounded local workflow, sanitizes text before publication (proposal 013 Must-Implement §Rumor Intake and Publication). Output is the value that lands in signal/text.
  3. User approval — MUST be explicit before publication; no silent publishing. Approval is a Memarium action-trace entry.
  4. Content validation — against whisper-signal.v1, including the allOf routing consistency rules.
  5. Envelope building — §Layer 2.
  6. Distribution-surface selection — §Layer 4.

Status

Per-component status table removed as stale — see the Post-M4 productization tracker above for authoritative status.

Layer 4 — Distribution routing

Policy router chooses the outbound path per whisper:

route_whisper(envelope, content) =
  match (content.disclosure_scope, content.routing_profile):
    ("private-correlation", _)        → INAC::push(…, inac.invitation passport)
    (_, "direct")                     → INAC::push(…)
    (_, "relayed" | "onion-relayed")  → Agora::publish(topic = envelope.topic_key)

This pseudocode does NOT reflect the implemented routing. The production model is whisper-core::resolve_outbound_privacy (OutboundPrivacyDecision::{Route, Refuse} + RoutabilityState + relay capability evidence), which is strictly richer than this (disclosure_scope, routing_profile) match — the match models no relay evidence at all. Do not copy this block as a contract; read it only for the failure-mode prose below, which remains accurate.

Both paths share the same envelope bytes. The router also enforces routing/failure-mode:

  • soft-fail: if the selected surface is unavailable, downgrade per policy (e.g. publish as whisper-durable on Agora only after the operator-configured soft-fail rule allows it),
  • hard-fail: refuse to publish at all rather than leak through an unapproved surface.

Status

Per-component status table removed as stale — see the Post-M4 productization tracker above for authoritative status (routing and routing/failure-mode enforcement are realized in whisper-core).

Layer 5 — Reception

Receiver-side ingest is split across the two surfaces but uses the same attestation-gate (proposal 041 §9).

Agora ingest (public whispers)

  • Topic ACL (proposal 035 §5) evaluates before persistence.
  • Attestation-gate (proposal 041) validates:
  • nym-certificate presence and scope when author/participant-id is nym:,
  • disclosure/scope against relay policy (public deployments SHOULD refuse private-correlation; enforcement is per-relay ingest policy),
  • rate budget per author and per topic,
  • any deployment-specific predicates (nym_council_in, pseudonymous_authorship, etc.).
  • Accepted records: subject-index update, SSE broadcast, Matrix federation bridge.
  • Memarium write on the receiver's side (if the receiver is a subscriber).

INAC ingest (private whispers)

  • Same attestation-gate (one verifier + one cache across surfaces).
  • Authorization source: typically inac.invitation passport for direct delivery, or whisper.receive general passport for subscribers known to the sender.
  • Memarium write is the primary side effect; there is no public index.

Status

Per-component status table removed as stale — see the Post-M4 productization tracker above for authoritative status. The attestation-gate, shared passport cache, nym-certificate resolver, and ingest-predicate grammar remain design targets; the predicate grammar is an open decision in proposal 041 (see §Open decisions).

Layer 6 — Memarium integration

Memarium is the local authoritative archive for every whisper the node authored or received (proposal 013 §Distribution → §Memarium as local storage).

Concerns (status tracked in the Post-M4 tracker above, not here):

  • Store envelope byte-identically (author and receiver) — Memarium has the byte-identical storage primitive; a whisper-specific observe rule is still a design target.
  • Action-trace for whisper lifecycle (draft, redact, approve, publish, forward, interest-count, threshold-crossing).
  • Per-whisper privacy policy (replay, forward, expose) — a new policy class.
  • Indefinite self-custody default for own whispers (inherited from proposal 040 §4).
  • Retention policy for received whispers (receiver's own choice) — generic Memarium retention applies; whisper-specific defaults unset.

Layer 7 — Threshold detection and durable promotion

From proposal 013 §5: when enough whispers correlate on topic/class + context/facets, the signal may be promoted to durable public presence as whisper-durable.v1.

Components (status tracked in the Post-M4 tracker above, not here; note that whisper-core already carries threshold/promotion primitives such as threshold_ids and public_gossip_*):

  • Local correlation index over topic/class + context/facets (computed from Memarium).
  • Threshold policy (count, time window, per-federation variance) — a policy decision, not protocol.
  • whisper-interest.v1 registration flow (schema also pending).
  • whisper-threshold-reached.v1 emission on crossing.
  • Explicit consent step before durable promotion (013 §6).
  • whisper-durable.v1 content schema.
  • Agora kind registration for whisper-durable.v1.
  • Cross-surface migration (same envelope bytes, different surface; §Invariant 6).

Layer 8 — Source integrations

(Status column removed as stale — see the Post-M4 tracker above.)

Source Integration contract
Direct user (compose UI) Node UI composer → whisper-core intake adapter
Pod-user Pod → daemon whisper submit endpoint → intake adapter
Operator-observed Operator middleware → intake adapter with operator-observed source/class
Derived-local Local producer → intake adapter with derived-local
Monus Monus host capability → draft producer → intake adapter with monus-derived
Monus + Sensorium Sensorium → Monus → intake adapter with monus-sensorium-derived

Open decisions blocking implementation

These must be resolved (short decisions, not full proposals) before Layer 1+ can be built coherently.

  1. Topic-key convention for Agora whispers. Resolved by proposal 046: public/federated whispers use ai.orbiplex.whispers/<topic-class>. Direct/private private-correlation artefacts use private/<name> while on INAC and MUST NOT be submitted to Agora.

  2. Predicate grammar for disclosure/scope filtering in attestation-gate. Resolved 2026-07-03: the specialised disclosure_scope_in(values) predicate, not the generic content_field_in(path, values) — minimal trusted core and fast audit of a security gate; any future filter over another field is a deliberate, explicit gate-contract change. Canonical record in proposal 041 §5.

  3. whisper-interest.v1 contract. Resolved 2026-07-03: counting/deduplication is per-node with a diversity guard (consistent with P014 per-node rate limiting; threshold policy MAY require source diversity — e.g. minimum distinct federations or trust tiers — post-MVP to resist single-operator inflation), and the artifact is scope-driven on both surfaces, mirroring whisper-signal.v1: disclosure/scope selects Agora (signed record) vs INAC (private artifact). Fields/lifecycle details land with the schema.

  4. whisper-durable.v1 content schema shape. Resolved 2026-07-03: aggregates plus a content-addressed ref to the original whisper-signal.v1 — the durable artifact stays light, content is verifiably linked rather than republished, and text redacted under one scope is never re-published under another.

  5. Threshold policy home. Resolved 2026-07-03: layered — per-topic-class defaults in the kind registry with a per-node operator override. Sensible defaults travel with the kind definition; the operator keeps the local last word; both layers are explicit and auditable. Federation-pack-level tuning was rejected as too heavy a change cycle for a tuning parameter.

MVP boundaries

Note: these are the original MVP boundaries, and this section sits outside the "Historical build plan" banner above. Several have been overtaken by the M4 path in the authoritative status at the top (e.g. whisper-threshold-reached.v1 and association-room-proposal.v1 are implemented for the Agora M4 path). The still-live constraints are the design restrictions below; the schema-scope bullet is corrected to match the tracker.

Keep these restrictions explicit in both code and docs during MVP:

  • one author per whisper; no co-signatures,
  • source/class = monus-sensorium-derived requires a present Sensorium host capability at author time (fail-closed),
  • routing/profile = onion-relayed is accepted as intent but the MVP does not ship an onion-routing realization; requests with hard-fail + onion-relayed on a node without such capability MUST refuse to publish,
  • threshold detection is local-only in MVP; no cross-federation gossip of interest counts. Any future cross-federation interest exchange must use Proposal 079 alliance policy as an admission input without replacing Whisper's author/disclosure/publication checks,
  • durable promotion requires an explicit user consent step; no automatic promotion in MVP,
  • whisper-threshold-reached.v1 and association-room-proposal.v1 are implemented on the Agora M4 path (association-room as schema + bounded local lifecycle facts; richer room/case facts still pending). whisper-interest.v1 is no longer blocked on a direction decision: the per-node counting, diversity-guard, and scope-driven surface choices in §Open decisions are resolved; the still-pending work is the concrete schema and registration flow.

Each step should leave the tree in a compiling state.

  1. Apply resolved predicate grammar. Add the specialised disclosure_scope_in(values) predicate from proposal 041 §5 to the attestation-gate implementation and negative tests.
  2. Layer 0. whisper-interest.v1 schema using the resolved per-node counting, diversity-guard, and scope-driven surface choices.
  3. Layer 2 — envelope builder. whisper-core crate with content validator + envelope sign adapter. Golden vectors.
  4. Attestation-gate (dependency, proposal 041). Cross-cutting primitive used by all receivers.
  5. Layer 5 — Agora ingest path. Whisper-specific predicates wired through attestation-gate; Matrix bridge allowlist for ai.orbiplex.whispers/*. This lets public whispers start flowing.
  6. Layer 4 — router (Agora branch only). Router that publishes to Agora; returns explicit error for private-correlation until INAC exists.
  7. Layer 6 — Memarium observe rules + action-trace. Whisper lifecycle facts.
  8. Layer 3 — authoring pipeline. Intake → redact → approve → validate → envelope → route.
  9. INAC (proposal 042 impl). Separate track; see 042-inter-node-artifact-channel-impl.md. When INAC lands, the router's INAC branch flips on.
  10. Layer 7 — threshold detection + durable promotion.
  11. Layer 8 — source integrations (UI, Monus, Sensorium).

Testing posture

  • Golden vectors for canonical_signed_bytes and record/id in whisper-core/tests/ — one vector per representative whisper shape (rumor vs weak-signal; problem vs idea polarity; participant vs nym authorship). These vectors are a federation-level contract.
  • Round-trip tests for content-body-only validation: WhisperSignalV1 → JCS → parse → WhisperSignalV1 must be byte-identical on re-serialization.
  • Envelope-vs-content regression tests: assert that none of the removed fields (signal/id, created-at, sender/*, rumor/nym, auth/*) appear in the content body after build.
  • Nym-authorship binding tests: given a nym private key, the built envelope must have author/participant-id beginning with nym: and envelope signature verifiable against the nym public key.
  • Surface-migration tests: a whisper authored for INAC and republished to Agora must pass record/id equality.
  • Attestation-gate tests (shared with proposal 041): reject private-correlation on a public-deployment relay; accept on a closed-federation relay.

First reviewed Matrix publication and status-cache withdrawal (2026-11-06)

The source-bound Sensorium fixture independently verifies A's private AD certificate, reviewed protected v3 UI signing and Matrix HTTP 202/public readback. Its separately pinned campaign reaches 21/131, with 110 missing and identical ordinary/optimized results. B fails startup before export: status-cache compaction requires a formerly adopted profile absent from current configuration. C is reaped; no common-receiver, threshold, complete-PDU or full A/B credit follows. The earlier SQLite 22/131 retains its separate source and result.

The owner correction validates retained canonical bytes, digest and source links before withholding unadopted status. It preserves facts and existing budgets, refuses current use without adoption and restores signature/cursor verification only on exact readoption. Other adopted profiles can operate and missing projections can rebuild; corruption remains a refusal. Component regressions pass and the next v3 conformance selection requires both withdrawal/corruption assertions. Fresh current-source native/runtime and Matrix qualification remain open; no historical gate or production adoption is renewed. See node:docs/evidence/sensorium-web/SENSORIUM-WEB-V3-FIRST-MATRIX-PUBLICATION-20261106.md and node:daemon/docs/CONTEXTUAL-NYM-CERTIFICATION.md.

Scoped reviewed-v3 Matrix first pair (2026-11-07)

Two ordinary reviewed producers and an independent C/D projection form one verified first pair on Synapse. A separate client and native reader bind every complete PDU to the original author buffer. The expiry proof reuses persisted current-view verification because status-only evaluation does not rewrite membership-change history. Existing phase-1 acceptance retains its original scope; this scoped v3 checkpoint does not close full A/B or production readiness.

The independently composed campaign verifies 25/131, with 106 missing, and identical ordinary/optimized incomplete results. Runtime/capture Node 985964ee83 / Orbidocs 2b5ea4fd and corrected reader Node d83a9a0f remain separately pinned. Both original reader refusals and every source observation are preserved; no certification/publication retry, production adoption or historical gate renewal follows. Current implementation evidence: node:docs/evidence/sensorium-web/SENSORIUM-WEB-V3-MATRIX-FIRST-PAIR-20261107.md.

Later 2026-11-07 fixture replica/window checkpoint

The source-bound campaign now verifies 29/131, with 102 missing, identically in ordinary and optimized readers. Exact-predecessor A2 renewal on a distinct registered node of A's participant retains one source and no first emission at independent C. A newly started C reads the same signed records after the single aligned window ends. Runtime 985964ee83 / 2b5ea4fd, capture 845bf6f7 / ca9b73c9 and window reader fb4301fb remain separate pins. Earlier source-bound reports and the original four-PDU scope remain unchanged.

This adds two replica/source-count and two window/history obligations. It does not qualify certificate-at-boundary, all-producer end refusal, old-node v2 cutoff, complete public/P082 direct-peer/Room/gossip privacy or full Gate A/B. Production is not adopted. The remaining fixture matrix and existing parent trackers stay open; another profile digest does not automatically preserve source continuity. Current implementation evidence: node:docs/evidence/sensorium-web/SENSORIUM-WEB-V3-REPLICA-WINDOW-20261107.md.

Later 2026-11-07 SQLite/late-receipt checkpoint

The SQLite/late-receipt fixture checkpoint verifies 32/131, with 99 missing, identically in ordinary and optimized readers. One reviewed Harvester v3 owner completed protected UI preview/sign across restart, SQLite publication (201) and a signed receipt. A new SDK receiving store first admitted the exact original author bytes/receipt after certificate expiry plus skew, using the real clock, and retained its facts across close/reopen. This is not an HTTP import, common SQLite threshold/room proof, second publication or full Gate A/B.

Runtime 985964ee83 / 2b5ea4fd, capture fe7c595e / 65a4b8a3, receiver build 2995a9f7 and final reader ba4a235b stay separate. All previous29 original predicates reverify. The earlier wrong-backend runtime refusal, unattempted late SDK, incomplete reader and empty-stderr campaign refusal remain preserved; corrections use new cases or read-only views, without retrying an uncertain publication. Backend readiness now precedes certification. The bounded thirteen-root retention review leaves all217 exact-copy candidates blocked; no retirement action or production adoption occurred. Parent trackers and the remaining99 obligations stay open, including all-producer end refusal, boundary/ old-node-v2 cutoff and full P082/gossip privacy.

Current implementation evidence: node:docs/evidence/sensorium-web/SENSORIUM-WEB-V3-SQLITE-LATE-RECEIPT-20261107.md.

Native owner-policy checkpoint (2026-11-07)

The native policy checkpoint verifies 37/131, with 94 missing, identically in ordinary and optimized readers, exit 2. All previous32 original predicates reverify. Five native AD/P091 obligations add inactive/complete adoption, editorial versus normative factory identity, separate ceiling refusal and unchanged old claim terms. Eleven actual system-clock stages include restored controls and owner reopen; three coherently rehashed scratch substitutions each refuse and restore the original positive.

Native build 190552c41, capture a8933630 and final reader 714e0bc10 remain separate from existing runtime 985964ee83 / normative 2b5ea4fd. Original rehearsals, cold-WAL collector refusal and hold-field reader refusal remain preserved; the successful runtime was not repeated for reader correction. Full gates and production adoption remain open, and parent statuses are unchanged.

Complete v3 reception/mixed classes, all-producer end refusal and both complete Gate A/B campaigns remain separate mandatory qualifications. Policy checkpoint evidence: node:docs/evidence/sensorium-web/SENSORIUM-WEB-V3-NATIVE-POLICY-20261107.md.

Native claim and transfer checkpoint (2026-11-07)

The native claim checkpoint verifies 41/131, with 90 missing, identically in ordinary and optimized readers (incomplete, exit 2). All previous37 original predicates reverify. Four native AD obligations add same-key parameter conflict, overlapping concurrent retain and lost-response reopen/retry, atomic transfer/ claim/first-result commit, and explicit live extension with real expiry controls. Four coherently rehashed scratch substitutions each refuse with a restored original positive. The native build/capture 069e95bbd and corrected reader b7d0d6fc8 stay separate from unchanged runtime 985964ee83 / normative 2b5ea4fd; the original reader expiry-model refusal remains preserved.

No signing, first-publication or receiver qualification is inferred from these native owner effects. Full gates and production adoption remain open; parent statuses are unchanged. Claim checkpoint evidence: node:docs/evidence/sensorium-web/SENSORIUM-WEB-V3-NATIVE-CLAIMS-20261107.md.

Native legacy checkpoint (2026-11-07)

The native legacy checkpoint verified 43/131, with 88 missing, identically in ordinary and optimized readers (incomplete, exit 2). All previous41 predicates reverify. Two native AD rows add digest collision/monotonic interest through ordinary release, actual expiry/sweep and unadopted reopen, plus explicit native operator-method release with a withdrawn admission witness. Same-object, managed-only GC and healthy controls establish the causes; four rehashed scratch substitutions each refuse and restore the original. Native build/capture/reader 6fcf49a3d / 35c23a9c remains separate from runtime 985964ee83 / normative 2b5ea4fd. Both pre-effect preparation refusals remain preserved.

Native owner cleanup does not qualify authoring, first publication or complete receiver behavior. Full gates and production remain open; parent statuses are unchanged. Legacy checkpoint evidence: node:docs/evidence/sensorium-web/SENSORIUM-WEB-V3-NATIVE-LEGACY-20261107.md.

Original bytes and explicit revalidation checkpoint (2026-11-07)

At this original-byte checkpoint the campaign verified 48/131, with 83 missing, identically in ordinary and optimized readers (incomplete, exit 2). All previous 43 cases reverify. Five composed predicates cover exact original transfer/inline representation custody, actual gzip and typed ISO-8859-2 refusal without handoff, original-preserved extractor omissions, acquisition horizons and three explicit 304 through delayed review, restart and replay. Two owners are reaped; no unsolicited fetch or renewed claim occurs. Six coherently rehashed controls per mode each refuse then restore original positives. 73 Python tests per mode pass.

Capture dade45982 and corrected reader 945a2bf9c / Orbidocs e37021ad remain separate from the pinned runtime 985964ee83 / normative 2b5ea4fd. The original capture runs once; three reader precondition refusals remain preserved. No production Rust, schema or capability contract changed. Ten-root catalog/dry-run review records 158 selected files/6264441 bytes, seven exact duplicate-output candidates 1767461 bytes and zero eligible. Active 48 evidence/authority remains protected; no retained deletion/move/retirement authorization/execution occurred.

This checkpoint stops at a reviewed local draft; signing, publication, full receiver/privacy and window/cutoff requirements remain separate. Full gates and production remain open; parent statuses are unchanged. Original-byte checkpoint evidence: node:docs/evidence/sensorium-web/SENSORIUM-WEB-V3-ORIGINAL-BYTES-20261107.md.

Finding origin and adopted monitoring checkpoint (2026-11-07)

The original-input campaign verifies 51/131, with 80 missing, identically in ordinary and optimized readers (incomplete, exit 2). All previous 48 cases reverify. Three new rows cover all nine canonical finding-class mappings, non-web Sensorium/mixed material, and collection bound to an adopted configuration after its real 60-second interval. Thirteen immutable reviews replay across an actual owner restart, with two web snapshots and four exact custody refs. Scalar metadata cannot create web, monitoring or model lineage. These synthetic operator-inline finding classes imply no mail/filesystem/device acquisition or signed author-class reception. Scheduler is disabled; due execution is explicit.

Collector/reader bfb0d90b0 / Orbidocs f2abbe21 remain separate from runtime 985964ee83 / normative 2b5ea4fd. Six coherently rehashed controls per mode refuse, each followed by a restored-original positive; 82 Python tests per mode pass. The original capture is unchanged and runs once. Six-root catalog/dry-run review records 138 selected files/4863454 bytes, three duplicate-output candidates 1294585 bytes and zero eligible. Active qualification and authority dependencies remain protected; no retained deletion/move/retirement authorization/execution. These numbers are run observations, not contract thresholds.

Full processing/inference provenance, signed v2 absence versus v3 unknown, author-class reception, mixed authorship holds, clock/load/fault, all-producer window/cutoff, P082/gossip privacy and full two-backend gates remain open. Parent statuses and production adoption are unchanged. Current measured evidence: node:docs/evidence/sensorium-web/SENSORIUM-WEB-V3-ORIGIN-MONITORING-20261107.md.