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 / Research / PVM v6 — Compressed (Swordsman reading)

PVM v6 — Compressed (Swordsman reading)

The Equation

Static form:

$$\boxed{\begin{aligned}
V(\pi, t) = ; & P^{1.5} \cdot C \cdot Q \cdot S \cdot e^{-\lambda t} \cdot (1 + A_h(\tau)) \
& \cdot \left(1 + \sum_i w_i \frac{n_i}{N_0}\right)^k \cdot G(\text{guilds}) \
& \cdot R(d, \text{compression}, \rho) \cdot M(u, y) \
& \cdot \Phi_{\text{agent}}(\Sigma) \cdot \Phi_{\text{data}}(\Delta) \cdot \Phi_{\text{inference}}(\Gamma) \
& \cdot T_{!\int}(\pi)
\end{aligned}}$$

Differential form: $\quad \frac{dV}{dt} = \nabla_{\partial M} \cdot J_{\partial M} + S(x) - D(x)$

Gating: Multiplicative. Any term $= 0 \implies V = 0$.

Lattice: $\pi$ is a path through $\mathcal{L} = \mathbb{Z}/64\mathbb{Z}$, $;t$ is time since data generation.

Grouped V6 reading of the same equation:

$$V(\pi, t) = \Phi_{v5} \cdot \left[ \sum_{\text{terms}} \right] \cdot A_h(\tau) \cdot T_{\int}(\pi)$$

$$\Phi_{v5} = \Phi_{agent}(\Sigma) \cdot \Phi_{data}(\Delta) \cdot \Phi_{inference}(\Gamma)$$

Multiplicative gating: any axis at 0 collapses V. The form is unchanged from V5.4. V6 takes the t in the signature seriously everywhere below.

Inherited Terms (V1--V4)

Symbol Name Domain Description
$P$ Privacy Strength $[0,1]$ Cryptographic enforcement. Exponent 1.5 via C6.
$C$ Credential Verifiability $[0,1]$ Verify without revealing.
$Q$ Data Quality $[0,1]$ Accuracy, completeness, fitness.
$S$ Scope / Sensitivity $\mathbb{R}^+$ Domain-specific multiplier.
$e^{-\lambda t}$ Temporal Decay $(0,1]$ Freshness. $\lambda > 0$.
$M(u,y)$ Market Maturity $[0,1]$ User sophistication, market year.

Holonic Temporal Memory

$$A_h(\tau) = \sum_j p(\tau_j) \cdot w(\tau_j) \cdot e^{-\mu \cdot \text{age}(\tau_j)}$$

GUID-addressed holons. Infrastructure-independent. $p(\tau_j) = 0 \implies A_h = 0$.

Three-Axis Separation

$$\Phi_{v5} = \Phi_{\text{agent}}(\Sigma) \cdot \Phi_{\text{data}}(\Delta) \cdot \Phi_{\text{inference}}(\Gamma)$$

Axis Formula Meaning
Agent $\Phi_a = \min(1, \frac{S/M}{\varphi}) \cdot \det(\Sigma)$ Swordsman $\perp$ Mage. $\cong D_{2n}$ (C14, 75%)
Data $\Phi_d = 1 - \max_j(\text{share}_j)$ No single provider holds majority
Inference $\Phi_i = 1 - I(\text{model};\text{executor})$ Generator $\perp$ Solver

Collapse any axis $\implies$ total collapse.

C7 (multiplicative form, 30%) is the falsification frontier. Boundary cases register-bound: partial collapse (alternatives: additive-with-floor, min()); axis correlation under composition (candidate correction: det(Σ) form); time dependence (repair path: the differential form). Structural anchor (C85, 40%): Protection+Delegation → Σ · Memory+Value → Δ · Connection+Computation → Γ.

Reconstruction Difficulty · the Moving Ceiling

Static form (V5.4):

$$R(d, c, \rho) = R_{\text{base}}(d) \cdot \left(1 - \frac{1}{c}\right) \cdot (1 + \alpha \cdot \rho)$$

$\rho = f(\text{traversal depth, duration, intentional transitions})$. Dual: privacy amplifier + agent maturity. Ceiling (proven): $R < 1$ under budget constraints. $\quad$ Error floor (proven): $P_e \geq 1 - R_{\max}$ via Fano.

The static form is superseded at V6 by the time-indexed form:

$$R(t) = \frac{C_S(t) + C_M(t)}{H(X)}$$

$$t^* = \sup{, t : R(t) < 1 ,}$$

H(X) fixed by the person; capacities evaluated against the strongest adversary class at time t; R(t) non-decreasing under capability growth (C82, 65%). The mechanism is the decoder, not the data. Every static guarantee has a shelf life.

Guild Efficiency

$$G(\text{guilds}) = \prod_g (1 + \text{efficiency}_g \cdot \text{active}_g / \text{total}_g)$$

Shared-parent coordination: O(1) not O($N^2$).

Path Integral

$$T_{!\int}(\pi) = 1 + \beta \int_\pi F(\gamma),d\gamma ;\cong; 1 + \beta \sum_{i=1}^{n} R(\text{step}_i)$$

$F(\gamma) = \text{resolution_depth} \cdot \text{fidelity}$. One lap = one cycle. Dragon ($\geq$62 laps) = closure.

Holographic Bound

$$\partial M: \text{96 edges encoding 64 vertices, toroidal topology} \qquad \frac{96}{64} = 1.5 = P^{1.5}$$

C4 RESOLVED. Boundary encodes bulk. $dV/dt$ computes on $\partial M$, not the 64-vertex interior.

$J_{\partial M} = J_{\text{agent}} + J_{\text{data}} + J_{\text{inference}} + J_{\text{compression}} + J_{\text{holonic}}$

96-edge boundary / 64-vertex bulk; $P^{1.5} \leftrightarrow 96/64$ (C6, convergent, 35%). Unchanged at V6.

Separation Bound · preconditions explicit

$$I(S; M \mid FP) < \varepsilon^* \qquad \text{(load-bearing wall)}$$

Theorem (95%): Preconditions 1 and 2 $\implies$ the additive MI bound $I(X; Y_S, Y_M) \leq I(X; Y_S) + I(X; Y_M)$ (equality iff additionally $I(Y_S; Y_M) = 0$) and the Fano floor $P_e \geq 1 - R_{\max}$; the strict bound $R_{\max} < 1$ holds exactly when, additionally, the declared capacity-deficit condition $C_S + C_M < H(X)$ holds.

Precondition 1 (non-collusion): $I(Y_S; Y_M \mid X) = 0$ and no third channel carries the inter-agent residue. Precondition 2: capacities against a stated adversary class. Grounding: Wyner (1975) equivocation; Fano converse; Leung-Yan-Cheong and Hellman (1978); Bayes-capacity; Geiger and Kubin.

Amnesia: $\varepsilon_{\text{amnesia}} < \varepsilon_{\text{policy}}$ (C17, quantitative at V6 via the compounding bound below). Topology > policy.

Betweenness centrality: $C_B(v) = \sum_{s,t} \sigma(s,t|v)/\sigma(s,t)$ (Brandes, 2001). The $\perp$ is the node with maximal betweenness in the trust graph.

The Reconstruction Ceiling · Proven, conditional regime

$$R_{max} = \frac{C_S + C_M}{H(X)} < 1 \qquad P_e \geq 1 - R_{max}$$

Both hold inside Preconditions 1 and 2; time-indexed per R(t). Outside the regime, the compounding bound governs:

$$I(O_N; S_1, \dots, S_N) ;\leq; \sum_{i} 2^{N-i}\varepsilon_i ;=; (2^N - 1)\varepsilon ;;\text{(uniform)}$$

(Asif-Amiri Theorem 4.1.) Amnesia separation breaks the chain and caps at $N\varepsilon$ (C83, 55%). The policy-to-amnesia gap is exponential-to-linear in N: at N=2, 3ε versus 2ε; at N=5, 31ε versus 5ε. C17 is hereby quantitative.

Algebraic Foundation · with the bridge

$$\mathcal{L} = (\mathbb{Z}/64\mathbb{Z},;+,;\times) \qquad D_{64} = \langle \text{neg}, \text{bnot} \mid \text{neg}^2 = \text{bnot}^2 = 1,;(\text{neg} \circ \text{bnot})^{64} = 1\rangle$$

Op Formula Agent Function
neg $(64-x) \bmod 64$ Swordsman Boundary. Additive inverse.
bnot $63-x$ Mage Projection. Bitwise complement.
$\text{neg} \circ \text{bnot}$ $x+1 = \text{succ}(x)$ First Person The step forward. $\blacksquare$

PRISM coordinates: $\text{blade}(x) = (\delta, \sigma, s)$: datum, stratum (Hamming weight), spectrum.

Pascal: ${1,6,15,20,15,6,1}$. Tiers: Null(0) / Light(1--2) / Heavy(3--4) / Dragon(5--6).

Six dimensions: Protection, Delegation, Memory, Connection, Computation, Value.

Hexagram: $[d_1 \ldots d_6] \to$ 64 I Ching states. Blade 63 = 111111 = Qian (The Creative).

All of the above carries unchanged. V6 adds: ARCH-1 enters the lineage, $\Sigma := \mu S.(\beta \lor \Omega(S,S))$ with activation ρ (C26-C29); operational layer ARCH-1R/T (C72-C76): ternary $\tau: T \to {+, 0, -}$, latent ≠ obstructed, $T = \text{orbit}(\rho, G)$. Bridge (C85, 40%): the axes and the triadic coordinates are one primitive; the gap is β (the recursion branches share only the First Person). Predictions: bnot-pairs invert all three axes; stratum-3 carries no dominant axis.

Operational Cycle

$$\text{cycle}(x) = \text{succ}(x) = \text{neg}(\text{bnot}(x))$$

Observe → Boundary → Project → Return:

Stage Operation Agent Ceremony
Observe $\text{id}(x)$ First Person Sun · disclosure
Boundary $\text{neg}(x)$ Swordsman Gap · silence
Project $\text{bnot}(\text{neg}(x))$ Mage Moon · reflection
Return $\text{succ}(x)$ Composition Recursion

$T_{!\int}(\pi) = 1 + \beta \sum_i \text{cycle}(\text{step}_i)$.
Progressive trust: Understanding $\to$ Constellation $\to$ Blade $\to$ Runecraft.

Amnesia Protocol · the obstruction upgrade

Definition (reachability statement, V5.4): Structural amnesia w.r.t. origin $O$ if no permitted operation sequence can reconstruct $O$.

ZK: completeness (output demonstrates), soundness (unique configuration), zero-knowledge (origin hidden).

Implementation: process boundary. Cosmological: Moon's orbit. Runecraft: Ed25519 key burned on close.

Obstruction statement (C86, 30%): Grade-2 forgetting $\iff$ the obstruction class to gluing local agent views into a global witness of O is non-vanishing; Grade-1 $\iff$ vanishing class, withheld gluing data. Terminal obstruction = structural loss of β (C73, 50%). Consequence if C86 holds: amnesia is the only term whose security is independent of t. Falsification: any view-composition recovering a Grade-2-forgotten O.

Forge Cryptography · with the accumulator

Property Method
Content addressing SHA-256
Tamper evidence Hash chain
Pre-evocation lock Commitment scheme
Identity binding Ed25519 (Mage, held)
Bilateral binding Dual Ed25519 (Mage held + Swordsman burned)

Moon phase: stratum $\to$ visibility ratio. $;$ 0 = New Moon, 6 = Full Moon.

V6 adds the City Key trust recursion as IVC (C87, 50%): Key = folded accumulator · trust tasks = step circuits · Charge = folding step · V63 = attested invariant. Substrate line: Nova, HyperNova, MicroNova, LatticeFold (post-quantum hedge). Presence regime 1: 🪢 is non-transferable, non-attesting local color; upgrade ladder: witness co-signing, elapsed-time proofs.

The Temporal Thread

$$X_b + Y_b > Z_b, \qquad Z_b = t^*$$

Behavioural Mosca (C49, 70%) closed onto the moving ceiling. Existence-Leak (C81, 70%): $I(\text{feasibility}; \text{method}) > 0$; floor: leakage-resilient ZK with $\lambda < 1$ impossible (Garg-Jain-Sahai). Discount (C84, 50%): $Z_b' = Z_b - D(a)$ on every public feasibility attestation. Countermeasure chain (C18-C21, 10-30%): $|\pi(t) - \pi'(t)| \gtrsim |\delta_0| e^{\lambda t}$, λ unmeasured. Taxonomy: ages progressively (C47, 50%). Trust edges: half-life from inscription (C30-C33).

The Geometry of the Gap

Stella octangula: no golden ratio (volumes rational: tetrahedron 1/3, octahedral core 1/6, compound 5/12 of the cube); C1 beside the figure is resonance, not derivation. φ lives in the lattice: $\delta(38) = 38/63 = 0.60317$ vs $1/\varphi = 0.61803$, gap 2.4% (C54). Parity cube (C88, 30%): the two tetrahedra are the even/odd parity classes of ${0,1}^3$; ${0,1}^6 = {0,1}^3 \times {0,1}^3$. Octahedral gap (C89, 30%): the intersection is the locus of the conditional-independence bound; β, the conditioning, and the core are three readings of one thing.

Conjectures

Numbering authority: CONJECTURE_REGISTER_V6.md, head C89. Bands: C1-C17 core · C18-C37 V6 series · C38-C46 bound collection · C47-C55 coherence (C51-C55 occupied) · C56-C66 City · C67-C71 Horizon (pinned v1.8.0) · C72-C80 renumbered at the register lock · C81-C89 registered during the V6 runs. Aliases: C46↔C32 · C60↔C48 · C61↔C49 · CM-C47↔C85.

Core band, V5.4 baseline (confidences as of April 2026; the register supersedes where they differ):

ID Claim Confidence
C4 96/64 discrepancy RESOLVED
C6 $P^{1.5} \leftrightarrow 96/64$ structural CONVERGENT 35%
C7 Three-axis multiplicative 30%
C11 $\rho$ amplifies + indicates maturity 55%
C12 Hexagram encoding 60%
C13 Bilateral witness quantum-resistant 65%
C14 $\Phi_a \cong D_{2n}$ 75%
C15 $T_{!\int} \cong$ resolution pipeline 65%
C16 Betti number trust invariants 25%
C17 Amnesia > policy separation 60% (quantitative at V6 via C83)
C18 Strange attractor dynamics ($\lambda > 0$) 25%
C19 $\rho$ = Lyapunov divergence 20%
C20 Three axes couple as Lorenz variables 30%
C21 Fractal sovereignty dimension 10%

Proven Results · scoped

V5.4 baseline (95%):

  1. Additive MI bounds from conditional independence
  2. Reconstruction ceiling $R < 1$ under budget constraints
  3. Error floor via Fano's inequality
  4. Graceful degradation under partial compromise
  5. Ring algebra $\mathbb{Z}/(2^6)\mathbb{Z}$ substrate
  6. Two-extension autonomy axiom (separate processes)
  7. Pretext DOM-free measurement as privacy primitive

V6 scoping: Additive leakage bound $I(X; Y_S, Y_M) \leq I(X; Y_S) + I(X; Y_M)$, equality iff additionally $I(Y_S; Y_M) = 0$, at 95% inside Precondition 1 only. Separation bound and ceiling: Proven, conditional regime. Worked instances of the drift (2026): Orchard (flaw 2022-05, found 2026-05-29, fixed block 3,364,600 on 2026-06-03); Schrottenloher (eprint 2026/1128, 2026-06-02, ~2 months after the ZK attestation).

Version Lineage

Version Date Core Addition
V1 2024 $P \cdot C \cdot Q \cdot S$
V2 Oct 2025 $+ e^{-\lambda t}$, network effects
V3 Nov 2025 $+ R(d), M, \Phi$
V4 Feb 2026 $+ \Sigma, A(\tau), T(\pi)$
V5 Feb 2026 $+$ three-axis $\Phi$, holographic bound, $T_{!\int}$
V5.1 Mar 29 $+ \rho$, bilateral witness (C11--C13)
V5.2 Mar 31 $+ D_{2n}$, PRISM (C14--C16)
V5.3 Apr 4 $+$ operational cycle, amnesia (C17)
V5.4 Apr 10 Consolidated. C18--C21. Full references.
V5.5 2026 Sublayer: attachment architecture
V6 Jun 2026 The gathering turn

V6 in full: R(t), t*; preconditions and external provenance; the exponential-to-linear gap; the Existence-Leak law; the bridge; obstruction amnesia; the Key as accumulator; the honest geometry; the unified register; unified V6 labels across all canon papers.

References

V6 grounding: Wyner 1975 · Fano 1961 / Cover-Thomas · Leung-Yan-Cheong & Hellman 1978 · Csiszár & Körner 1978 · Asif & Amiri arXiv:2603.05520 · Patil et al. arXiv:2509.14284 · AgentLeak arXiv:2602.11510 · Garg-Jain-Sahai · Schrottenloher eprint 2026/1128 · Blanco-Romero et al. arXiv:2603.01091 · Bakhta 2026 · Nova / HyperNova / MicroNova / LatticeFold · Mosca 2018 · Bergstra & Burgess 2019 · full list: privacy_value_v6.md §32.

V5.4 baseline references, carried: Shannon (1948). Fano (1961). Cover & Thomas (2006). Bergstra & Burgess (2019). Susskind (1995). McGilchrist (2009). Groth (2016). PLONK (2019). Nova (2022). Dwork & Roth (2014). Brandes (2001, 2008). Branco et al. (2025). Babbush et al. (2026). Cain et al. (2026). IEEE 7012-2025. UOR Foundation (2026). Hope & Ludlow (2023). Weyl & Tang (2023).

Narrative corpus (V5.4 record): The First Person Spellbook (31 acts, v10.0.0, CLOSED). Blog: sync.soulbis.com (Parts 0--5). Six grimoires: First Person, Zero Knowledge, Canon, Parallel Society, Plurality, City of Mages (Second Person). The Second Person Spellbook opened 2026-05-08 as the bound collection at tomes/; Tome IV (Witnessing · 5 acts) closed; Tome V (Crafting) open at the City of Mages on Drake Island.

Full spec: privacy_value_v6.md. Companion: pvm_v6_companion_guide.md.

Docs: github.com/mitchuski/agentprivacy-docs. Forge: spellweb.ai. Training: agentprivacy.ai. Trust: bgin.ai.


The boundary is always enough. Peer review invited: mage@agentprivacy.ai

(⚔️⊥⿻⊥🧙)😊

Related research · V7 intake

The open V7 research index places this reference alongside the wider programme of formal review, experiments and implementation research.

Public Agreement and Private Ambiguity adds a related-work reference for task-relative boundary sufficiency (C9), consistency obstructions (C16), compression experiments (C8) and refinement comparisons (C15). Public-result uniqueness and private-state reconstruction are distinct obligations. OPH supplies no new validation of multiplicative separation, leakage rates, ownability, geometric identity or quantum resistance. The register and disposition overlay preserve every affected status and confidence; no conjecture is minted. This reference addendum does not revise the V6 formal model or its proofs.

Assets

📎 pvm-v6-compressed-swordsman-readingpvm_v6_compressed.md

Navigation

← Welcome Visitors · The PVM Papers