The WISP Architecture

1The Problem WISP Solves

From single-signal authentication to multi-modal coherence

Most identity systems treat identity as a binary signal: the user is, or is not, who they claim to be. The decision is made at one moment, against one piece of evidence, and after that moment the system trusts the assertion until the next authentication event. This shape works when the cost of being wrong is low and when the user's coherent identity is unlikely to drift between checks.

For governed AI systems, neither condition holds. The cost of acting on a wrong identity decision can be severe — irreversible governance decisions, sensitive memory access, autonomous action with downstream effects. And the user's coherent identity — the structured composition of who they are across narrative, role, time, and context — is not a single signal that can be checked once. It is a manifold that must be continuously evaluated.

WISP is the substrate primitive that addresses this problem. It encodes identity as a multi-modal composite object — narrative, role, temporal, and location vectors — that the governance pipeline can evaluate as a geometric coherence question rather than a binary lookup.

2The Structure of WISP

Multi-modal composite as a single keyed object

A WISP comprises at minimum four orthogonal modality vectors:

• Narrative vector (v_narrative) — a semantic embedding of the user's coherent narrative. Not a transcript; a high-dimensional embedding that preserves narrative coherence as geometric proximity.
• Role vector (v_role) — an encoding of the user's current operating role and the authority that role conveys. Specializations exist for specific classes (v_role_user, v_role_doctor, etc.).
• Temporal vector (v_time) — a multi-scale Fourier encoding of when the user is acting (daily, weekly, yearly cycles, plus a linear epoch term).
• Location vector (v_location) — a geometric encoding of where the user is acting (physical, network, or logical context as appropriate).

Each modality vector is an independent geometric object. The composite WISP holds all four — and any additional modality vectors the deployment requires — under one keyed entry. Typed edges (has_narrative, has_role, has_temporal, has_location) connect the WISP to its constituent modalities, allowing the governance pipeline to address any modality independently or the composite as a whole.

3Authentication as Similarity

Cosine proximity to historical WISP, threshold gated

Authentication in the WISP framework is not a credential lookup. It is a geometric proximity question: how close is the currently-observed WISP to the historical WISPs the system has recorded for this user?

Because identity is multi-modal, the similarity calculation is per-modality, weighted by the recency of each modality's last observation:

This formulation has two important properties. First, it produces a continuous trust signal, not a binary decision — the system can know how sure it is, not just whether it is sure. Second, it gracefully degrades when one modality is unavailable: the weights renormalize and the available modalities carry the decision.

4Continuous Coherence Evaluation

Identity as ongoing manifold rather than discrete event

Because WISP is geometric, identity coherence can be evaluated continuously. The governance pipeline produces an updated trust signal whenever new evidence updates any modality vector. A user who was strongly authenticated five minutes ago and is now showing a substantially shifted narrative vector is, in this framework, no longer strongly authenticated — the system has the information needed to notice that.

This continuous evaluation is the substrate for the four-state authorization model:

• PERMIT — coherence is strong, action authorized.
• DELAY — coherence is uncertain, action held pending additional evidence.
• ESCALATE — coherence is low or anomalous, action surfaced for human review.
• DENY — coherence has failed, action blocked.

DELAY and ESCALATE are not error states. They are the system explicitly acknowledging that it cannot make a clean decision. This is in keeping with the STAN Truth primitive — the system's outputs must be traceable to coherent intent. When intent cannot be coherently established, the system is required to surface that uncertainty rather than guess.

5Persistent Semantic Memory Graph

WISPs as nodes, modalities as typed edges, history as the manifold

WISPs are stored in a persistent semantic memory graph. Each WISP observation becomes a node. Each modality is a typed edge. Historical WISPs accumulate as the user's behavioral manifold — the geometric record of who they have been over time, against which current observations can be compared.

The graph supports drift detection, role inference, coherence scoring, and provenance reconstruction. It is not append-only in the simple sense — modalities decay according to configurable half-lives (older evidence carries less weight) and stale modalities can be pruned. But every modality update is hash-chained into the governance decision record (SIG-C10 in the primitives companion), so the historical trajectory can always be reconstructed for audit purposes.

6Drift Detection

Semantic drift function as a continuous health signal

One of the substrate's ongoing health signals is the semantic drift function — a measure of how much the user's identity manifold has shifted between observations:

Drift values near zero indicate stable identity. Larger drift values indicate change — which may be benign (the user genuinely moved into a different context) or anomalous (the user's identity has been compromised or is being impersonated). The substrate does not categorize drift; it surfaces drift values to the governance pipeline, which combines drift with other signals to produce the four-state decision.

The drift function is computed per modality and aggregated. Drift in one modality (e.g., location shifted but narrative consistent) is treated differently from drift across modalities. The substrate's pre-computed health signal lets governance decisions inspect not just that the user has drifted but how they have drifted.

7Cryptographic Binding to Execution

From identity decision to action authorization

The governance decision produced by the WISP pipeline is not a passive recommendation. It is cryptographically bound to specific authorized actions through the Knox warrant primitives. A PERMIT decision produces a one-time warrant whose preimage authorizes exactly one downstream operation. The warrant's validity decays continuously (SIG-T07, thermodynamic warrant validity), so even an authorized action cannot be replayed indefinitely.

The cryptographic binding has three properties that matter for governed AI: non-replay (each authorization is consumed exactly once), decay (authorizations lose validity continuously over time, independent of exercise), and provenance (every action carries a cryptographic anchor back to the governance decision that authorized it).

The Knox architecture paper, forthcoming, describes the cryptographic substrate in detail. The companion already publishes the formal primitives.

8Verifiable Provenance Without Central Callback

Downstream verification through cryptographic artifact, not oracle

A practical concern in any cryptographically-anchored identity system is the cost of verification. If every downstream system must call back to a central authority to verify each WISP-anchored authorization, the central authority becomes a bottleneck and a single point of failure.

The WISP framework addresses this through verifiable provenance tokens — cryptographic artifacts that a downstream system can verify against the WISP's public anchor without round-tripping to Foundation infrastructure. The verification is a deterministic cryptographic check; the trust anchor is the published WISP public key (or Merkle root, depending on deployment) that the downstream system already possesses.

This is what makes WISP suitable for substrate-class deployment: the system's correctness is verifiable in distributed contexts, not gated through a central oracle.

9Mathematical Anchoring

Every claim has a formal signature

The descriptions above are summaries of formal mathematical specifications. Every primitive cited has a corresponding entry in the Mathematical Primitives Companion, where the formal definitions are stated rigorously with implementation pointers.

Key primitives referenced in this paper:

• SIG-T01 through SIG-T12 — temporal primitives (Fourier encodings, half-life decay, thermodynamic warrant validity, multi-modal temporal trust fusion).
• SIG-C01, C02, C10 — cryptographic primitives (hash chain generation, preimage verification, governance decision chain).
• SIG-G01 through SIG-G06 — geometric primitives (helix coordinates, centroid, variance, curvature).

10Status and Disclosure

What is public; what is forthcoming

The WISP architecture as described above is publicly disclosed through the issuance of US Patent 19/308,229. The patent specification covers the multi-modal identity authentication and governance system, the composite WISP object, the persistent semantic memory graph, authentication via vector similarity against threshold, the four-state governance decision, specific modality vector embeddings, drift detection mechanisms, verifiable provenance tokens, transformer-based embedding of narrative and role, and the temporal directionality encoding (clock-face manifold).

Subsequent disclosures — drift detection mechanisms with cryptographic anchoring, reinforcement-decay update rules, and adjacent extensions — are in continuation-in-part filings whose publication timeline is governed by USPTO process. The Foundation publishes architecture papers as the underlying disclosures issue.