guide to agentprivacy
Browse collections
✨Visualise
Connect with Star
Your VTA, your chosen perspective

The planned connection uses your VTA and the Trust Spanning Protocol to carry a scoped exchange for you or your agent. You choose what is presented; the receiving service checks the request before a view is shared.

This guide has no VTA connection adapter yet. Opening Star does not connect an identity or send a key.

Open Star ↗ · Inspect your City Key ↗
guide / Skill / Knowledgegraph

Knowledgegraph

Category privacy_layer · Version 4.0

Equation term Full entity-relationship registry of all model terms

Complete entity-relationship knowledge graph for 0xagentprivacy. Activates when needing the full map of entities (persons, agents, VRCs, guilds, pools, ceremonies), their relationships, and query surfaces. Use as the canonical reference for how all system components connect. Contains the comprehensive entity registry. V6 register note (2026-06-10): conjecture and version citations resolve to agentprivacy-docs/research/CONJECTURE_REGISTER_V6.md (head C89); model head: privacy_value_v6_formal_specification.md.

Source: Privacy Value Model V4 Formal Specification (Travers, Feb 2026)
Target context: Knowledge graph ingestion, trust graph systems, ontology alignment, entity-relationship overlap detection
Architecture: agentprivacy.ai · Sync: sync.soulbis.com · Contact: mage@agentprivacy.ai

Entity Registry

Core Entities

ENTITY: Privacy_Value_Model_V4
  TYPE: economic_model
  DOMAIN: [privacy, agent_economics, cryptography, network_theory]
  STATUS: working_paper
  AUTHOR: Mitchell_Travers
  DATE: 2026-02
  PREDECESSOR: [PVM_V1, PVM_V2, PVM_V3, PVM_V3.1]
  SUCCESSOR: PVM_V5 (planned: flow dynamics on manifold)

ENTITY: Sovereignty_Lattice
  TYPE: mathematical_structure
  STRUCTURE: boolean_lattice({0,1}^6)
  VERTICES: 64
  EDGES_UNDIRECTED: 192
  EDGES_DIRECTED: 384
  STRATA: 7 (following Pascal's row: 1,6,15,20,15,6,1)
  CONJECTURED_CORRESPONDENCE: UOR_Toroidal_Structure (unresolved: 96 vs 64 edges)

ENTITY: Swordsman_Agent
  TYPE: agent_archetype
  FUNCTION: privacy_protection, boundary_enforcement, data_minimisation
  KNOWS: what_to_hide
  DOES_NOT_KNOW: what_to_do
  SEPARATION_REQUIREMENT: conditional_independence(Mage_Agent, ε < 0.1)

ENTITY: Mage_Agent
  TYPE: agent_archetype
  FUNCTION: delegation, external_coordination, action_execution
  KNOWS: what_to_do
  DOES_NOT_KNOW: what_to_hide
  SEPARATION_REQUIREMENT: conditional_independence(Swordsman_Agent, ε < 0.1)

ENTITY: Separation_Matrix_Σ
  TYPE: mathematical_object
  STRUCTURE: symmetric_4x4_matrix
  DOMAIN_ENTRIES: [0,1]
  DIAGONAL: 1 (self-separation is trivially perfect)
  AGGREGATION: determinant (volume of sovereignty tetrahedron)
  FORCES: [Protect, Project, Reflect, Connect]

Sovereignty Forces

ENTITY: Protect
  TYPE: sovereignty_force
  ORIGIN: primary (Swordsman)
  FUNCTION: boundary_enforcement, data_minimisation

ENTITY: Project  
  TYPE: sovereignty_force
  ORIGIN: primary (Mage)
  FUNCTION: delegation, external_coordination

ENTITY: Reflect
  TYPE: sovereignty_force
  ORIGIN: emergent (from Protect)
  FUNCTION: temporal_integral_of_protection_decisions
  DEPENDS_ON: derivation_chain_τ, integrity_fraction_h(τ)

ENTITY: Connect
  TYPE: sovereignty_force
  ORIGIN: emergent (from Project)
  FUNCTION: network_effect_of_delegation_patterns
  DEPENDS_ON: stratum_weighted_network, agent_count_n_i

Model Terms

ENTITY: Privacy_Strength_P
  TYPE: model_term
  DOMAIN: [0,1]
  EXPONENT: 1.5 (superlinear returns)
  MEASURES: cryptographic_enforcement_quality
  VERSION_INTRODUCED: V1

ENTITY: Credential_Verifiability_C
  TYPE: model_term
  DOMAIN: [0,1]
  MEASURES: independent_verification_without_revealing_underlying_data
  DEPENDS_ON: zero_knowledge_proofs
  VERSION_INTRODUCED: V1

ENTITY: Temporal_Memory_A(τ)
  TYPE: model_term
  FORMULA: α · ln(1 + |τ|) · h(τ)
  GROWTH: logarithmic (conjectured; power-law and sigmoid alternatives plausible)
  GATE: h(τ) = 0 → A(τ) = 0 (unverified history contributes nothing)
  VERSION_INTRODUCED: V4

ENTITY: Reconstruction_Resistance_R(d)
  TYPE: model_term
  DOMAIN: (0,1)
  BOUND: R_max = (C_S + C_M) / H(X) < 1
  PROOF_STATUS: proven (under dual-agent conditional independence)
  V6_NOTE: V6: the ceiling is time-dependent, R(t) with shelf life t* (C82); the static result is Proven in the conditional regime.
  VERSION_INTRODUCED: V3

ENTITY: Sovereignty_Duality_Φ(Σ)
  TYPE: model_term
  FORMULA: min(1.0, (S/M)/φ) · det(Σ)
  DEPENDS_ON: Separation_Matrix_Σ, golden_ratio_φ
  GOLDEN_RATIO_STATUS: conjectured_optimal
  VERSION_INTRODUCED: V4

ENTITY: Edge_Value_T(π)
  TYPE: model_term
  FORMULA: 1 + β · Σ_e f(e) · g(n_e)
  MEASURES: trajectory_through_sovereignty_space
  INSIGHT: transitions_define_identity_more_than_states (cf. Yoneda_lemma)
  VERSION_INTRODUCED: V4

Detailed Reference

See references/detailed-reference.md for complete entity maps, query surfaces, and extended documentation.

Provenance

Forkable skill page migrated from the agentprivacy skills repo (persona/agentprivacy-knowledgegraph, v4.0).
Source of truth: agentprivacy-skills. Fork this page to materialize a local SKILL.md via fedwiki-to-skill.

origin: 0xagentprivacy · author: Mitchell Travers

Assets

📎 knowledgegraphSKILL.md