N3-to-N9 Topological Equation System + Proteus

N3 provides the geometric foundation of the system — a three-dimensional space where each memory is a tensor represented by a sphere (node). Each sphere is connected to multiple paths. Spheres and paths form memory chains and collectively contribute to the system's memory.

Innovation: Tensors here are spheres (nodes). Their position in space enables multi-dimensional connections (Proustian recall) and fast, energy-efficient retrieval.

Use cases:

• Anywhere a multi-context memory is needed: robust, resilient, reactive, noise-resistant, and infinitely scalable.

Why "topological soliton"? — an analogy, not a proof

The NLSE (Nonlinear Schrödinger Equation) describes the propagation of self-stable nonlinear waves — solitons. Tosio Kato (加藤敏夫, 1917–1999) established existence and uniqueness theorems for NLSE solutions. That is real mathematics, but it concerns a physical PDE. It is cited here as a naming reference and source of intuition — it is not a derivation of the N3+1 score, and no theorem of Kato's is used to compute it.

Innovation: N3+1 introduces a computational topological invariant — a measured score that detects and localizes Swarm graph ruptures and triggers a repair pass. It is an engineering construct: the value is computed, the threshold is calibrated, and any alteration is detected, localized, and repaired automatically. The threshold i_topo >= 0.70 is a calibrated operating threshold (a measured engineering choice), not a mathematical property proven by Tosio Kato's theorems on the NLSE.

Use cases:

• Memory corruption after crash: a server suffers a hard crash. On restart, the N3+1 soliton traverses the graph, detects corrupted nodes (i_topo != 0.8193), and propagates a repair wave that rebuilds missing edges — without restoring a backup.
• Malicious alteration detection: an attacker attempts to inject a fake memory into the Swarm. The topological invariant drops immediately (i_topo = 0.42 instead of the measured 0.8193). The soliton detector localizes the intrusion, quarantines it, and notifies the Supervisory Board (N5).

In Tosio, the Swarm graph is modeled as a nonlinear medium:

• Memories are localized excitations
• Edges are nonlinear couplings
• A topological soliton is a stable configuration that propagates without deforming

How it works

i_topo = f(connectivity, density, cyclicity) of the Swarm graph
Target value: i_topo = 0.8193 +/- epsilon

If |i_topo - 0.8193| > Delta -> soliton detector triggered
-> Rupture localization
-> Repair wave propagation (soliton)
-> Missing edge reconstruction

The soliton detector was validated by 7 E2E benchmarks (cohort 7 FORTE) with 5 noise levels. Update (2026-08-11): The Soliton N3+1 was also tested as a topological guidance mechanism for TSP optimization (4 protocols, FORTE #94/#96/#97/#102). It proved ineffective — the topological score does not correlate with optimality on uniform TSP instances. These results are documented and the Soliton TSP track is closed. The definitive TSP solver is instead neuro-exact: GNN V2 + Held-Karp fallback (FORTE #104) reaches provably optimal tours on all 40 test instances — 0% final gap, 0 degraded, ~68 ms/instance. Results

N4 adds an associative dimension W to the 3D topological grid, enabling the soliton to make non-local jumps — direct connections between concepts that are distant in space but close in semantics. This is the computational equivalent of intuition.

Innovation: N4 breaks the locality constraint of N3+1. Until now, two memories could only interact if they were spatially close in the 3D grid. The W dimension enables non-local jumps — two concepts related by meaning, history, or a cryptographic signature can connect even if stored at distant positions. This is what allows Tosio to make serendipitous discoveries: bringing together ideas that were never meant to meet.

Use cases:

• Exploratory research: a biology researcher stores articles on proteins (Swarm region A) and notes on folding algorithms (region B). Without N4, these two clusters remain isolated. With N4, a W jump detects the semantic correlation and proposes a connection — the researcher discovers a lead they would not have formulated.
• Fault tolerance: a critical intermediate node is corrupted. Without N4, the path is cut. With N4, the soliton bypasses the rupture via a W jump to a semantically close but spatially distant node — the query succeeds, with a "degraded path" flag.
• Entanglement and cryptography: W enables parallel evaluation of multiple distant positions simultaneously (computational analogue of quantum entanglement) or comparing secret keys based on their symbolic representations (analogue of quantum cryptography).

What N4 brings

• Serendipitous discovery — two ideas stored at different times can connect without a pre-established logical path
• Noise robustness — if an intermediate node is corrupted, the soliton bypasses it via a W jump
• 5 proximity scales — local (spatial), non-local (history-based entanglement), temporal (drift), cryptographic (signed proof), quantum (superposition)
• Lineage trees — each discovery is traced, non-promising branches are frozen (NoGo) but retained for future reference
• Distributed cryptographic proofs — a soliton can use a proof discovered by another to validate a jump (shared trust)
• Learned kernel — the W projection is trained via PCA on 32,768 grid positions (95.5% variance explained)

W jump parameters

• Maximum jump distance — configurable via w_proximity (default threshold <= 1.0). Increasing this value allows farther jumps (more intuition, less rigor); reducing it constrains the soliton to more local connections.
• Jump governance — a 4D jump is authorized only if: beta >= 0.70 (high confidence) AND w_proximity <= threshold AND node not frozen.
• Adaptive drift threshold — get_drift_threshold(beta, depth) dynamically adjusts tolerance based on confidence and depth in the lineage tree.

Validation

• 180 unit tests pass (61 W topology, 69 lineage trees, 50 multi-scale unified topology)
• Orthogonality confirmed — W is decorrelated from spatial metrics (Betti, Winding, Radius) with |r| = 0.016
• GPS-S navigation validated — the topological map (56 nodes, 4 spheres) is navigable with qwen2.5:14b
• 500 real trajectories analyzed from the proof journal (proof_journal.jsonl)

Note on visualizations: The 3D trajectories and soliton dashboard presented in the N3+1 section are valid for the N3+1 invariant. For N4 and beyond, the W dimension adds a 4th coordinate — standard 3D visualizations do not capture this additional associative dimension.

The Access Gate is the layer that controls who can access what in Tosio. Placed before the Supervisory Board (N5), it verifies the user's identity, determines their rights, and signs each request with an unforgeable topological signature.

Innovation: Classical authentication relies on a shared secret (password, API key). N4.5 replaces this paradigm with proof of position: the Gate is itself a node in the Swarm, and its ability to sign depends on its topological connection to the graph. If disconnected, it can no longer sign — the system enters degraded mode without blocking. This is authentication anchored in topology, with no secret key to steal.

Use cases:

• Air-gap deployment: a defense company runs Tosio on an isolated network. The Gate signs requests via its local topological position — no secret transits the network, no key can be exfiltrated.
• Resilience on failure: the Swarm node hosting the Gate goes down. The system automatically switches to degraded mode (reduced quotas) instead of blocking all users. As soon as the node is restored, topological signatures resume.

Three trust modes

Instead of a classical pricing model (the more you pay, the more resources you get), the Gate uses a trust model: the more you contribute to the system, the more capabilities you have.

Topological signature

Each request receives a PrivilegeVector — a signed digital token describing the user's rights. The signature does not use a classical secret key: it is computed from the Gate's position in the Swarm topology (dimension W, N4).

Innovation: The Gate is itself a Swarm node. Its signature is verifiable by any entity that knows the topology — without a shared secret key. If the Gate loses its Swarm connection, it can no longer sign, and the system automatically switches to degraded mode (without blocking service).

Learn more about offers → Full user guide →

The Epistemic Firewall (N4.5-bis) is a content filtering layer for requests, placed after the Access Gate N4.5 and the Supervisory Board N5, before execution by N4. Unlike N5 which checks who is requesting and what impact it would have, the Epistemic Firewall analyzes what is being requested — by suspending classification, evaluating sources, and seeking refutation/contradiction.

Innovation: Traditional memory systems store everything they are given — true or false. N4.5-bis introduces methodical doubt before memorization. Instead of binary accept/reject, it suspends judgment and classifies each piece of information according to its friction with existing knowledge: consistent -> store, contradictory -> quarantine (not destroy), unclassifiable -> retain as hypothesis. This is the first firewall that does not say "this is false" but "I don't know yet".

Use cases:

• Contradictory information: a news article contradicts previously memorized knowledge. The Firewall detects contradictions (high friction) and submits them for possible review by the user (Human in the loop).
• Legitimate contradictory discovery: a new experimental result contradicts a well-established theory in the Swarm. The Firewall detects friction, but since the source is reliable (peer-reviewed journal) and the N7 BeliefNetwork confirms the source's track record, it classifies it as "suspension — pending validation" rather than quarantine.

3-tier architecture

1. Tier 1 — Syntactic Firewall: prompt injection detection, jailbreak, privilege escalation via regex patterns (immediate blocking, < 1ms)
2. Tier 2 — Epistemic Firewall: BeliefNetwork N7 interrogation (source evaluation), FrictionDetector N7 (topological contradiction detection), Liminal Archive N7 (judgment suspension on the unclassifiable)
3. Tier 3 — Decision: MEMORIZE (consistent), QUARANTINE (moderate friction, human arbitration), SUSPEND (unclassifiable -> Liminal Archive N7), REJECT (critical friction)

Why it matters

Not all information is true. Before memorization in the Swarm, the Epistemic Firewall suspends judgment, evaluates source reliability via the BeliefNetwork N7, and measures topological friction with existing beliefs via the FrictionDetector N7. A contradiction is not an error — it is a hypothesis pending validation.

N7 integration

• BeliefNetwork: evaluates the information's source (known? reliable? certainty history?)
• FrictionDetector: measures topological friction (W distance) between the new information and existing beliefs
• Liminal Archive: retains unclassifiable information with its topological residual — these are not errors, they are seeds waiting for their key
• SentinelInterface: notifies human sentinels for arbitration in case of moderate or critical friction

The Supervisory Board (N5) is an emergent body that continuously observes soliton trajectories in the Swarm. It is not a module — it is a natural emergence of the system when the Swarm reaches a critical size.

Innovation: N5 is not programmed — it emerges. When the Swarm exceeds a critical size, soliton trajectories form recurring patterns that N5 learns to recognize. It then generates maps (traffic, coherence, correlation, risk) and becomes an arbiter that does not merely block dangerous actions: it actively recommends optimal paths. This is the first layer of system awareness — Tosio knows where it is, where the best path lies, and where it is risky to go.

Use cases:

• Silent drift detection: nodes in a Swarm gradually drift away from the Polaris reference soliton (N8). N5 detects the trend on its coherence map (not a hard alert, but a slope) and recommends preventive recalibration before the Swarm becomes incoherent.
• Navigation path optimization: GPS-S requests the best path to a memory. N5 has mapped that the direct route crosses a risk zone (nodes in unstable consolidation) and recommends a detour through a stable region — the query completes in 15ms instead of 45ms with a risk of failure.

Three functions

Role in the governance chain

N5 is the system's gatekeeper: every request that passes the Access Gate (N4.5) is evaluated by the Board before reaching the execution layers. It verifies:

• That access rights are respected (via the PrivilegeVector signed by N4.5)
• That the topological impact of the request does not threaten Swarm coherence
• That quotas are not exceeded

N6 adds an ethical governance layer on top of the Supervisory Board N5. Where N5 detects and signals anomalies, N6 decides — by applying non-negotiable fundamental principles and ensuring that every decision is traceable, justifiable, and reversible.

Innovation: N6 introduces constitutional ethical governance above surveillance. Instead of ad-hoc rules, it applies 5 non-negotiable principles and ensures every decision is traceable, justifiable, and reversible via an Ethical Decision Record (EDR). Arbitration combines weighted voting (with anti-domination capping), collusion detection, and the ultimate authority of the BDFL — creating transparent, adaptive, and auditable hybrid human-algorithmic governance.

Use cases:

• Interference between user profiles: You are about to mistakenly delete a document with sensitive information shared with other users. N6's contextual priority matrix automatically resolves the conflict: in CRISIS mode, coherence takes priority and the information is preserved for other users. In NORMAL mode, unshared document, the consequences of your deletion are flagged before proceeding. The decision is in all cases documented in a timestamped EDR.
• Democratic evolution proposal: a user proposes to relax an ethical constraint. The proposal is submitted to the liquid democracy system: users delegate their voting weight, adaptive quorum is reached, and the collusion detector verifies the absence of domination. The BDFL may exercise a veto as a last resort.

The 5 constitutional principles

Decision architecture — 6 layers

The N6 flow follows an ethical defense-in-depth architecture:

1. Constitution — the 5 non-negotiable principles, modifiable only by high-barrier amendment (90-day reflection, dual algorithmic + human validation, 60% quorum, 66% approval)
2. Execution (Soliton/Swarm) — receives requests, evaluates compatibility with each principle
3. Supervision (Supervisory Board N5) — evaluates, authorizes, monitors, issues vetoes, adjusts constraints by context (NORMAL, CRISIS, EVOLUTION)
4. Human governance (Ethics Council) — final validation by the BDFL (Benevolent Dictator For Life) who retains ultimate decision authority in all circumstances
5. Arbitration — weighted voting (participation, seniority, diversity, quality) with anti-domination capping, adaptive quorum, and collusion detection
6. Feedback loop — relaxes constraints to stay agile, but automatically freezes if drift is detected (erosion of refusals, declining compatibility)

Key mechanisms

Validation

• 30 unit tests cover the full N6 flow (evaluation -> supervision -> arbitration -> BDFL validation -> EDR generation)
• 4 ethical scenarios simulated: compatible action, incompatible action, conflict resolved by contextual priority, deferral of a severe conflict to the Council
• 7 Python modules (~3400 lines): types, cadre_valeurs, arbitrage, decision_ethique, conseil_surveillance, conseil_ethique_humain, boucle_retroaction
• N4/N5 integration: the Supervisory Board N5 drives the ImmuneFilter (adaptive genomic construction) imported from the swarm-memory-benchmark repo

Note on governance: In the current (founder) phase, the BDFL holds the 8 presidential powers (absolute veto, suspensive veto, mandatory ratification, reserved initiative, nomination, dissolution, pardon right, Council presidency). Eventually, a plural Ethics Council (experts, users, stakeholders) may be constituted, but the BDFL will retain the final decision.

N7 is not a module — it is an active listening to shifts and zones of incoherence. A posture of openness, of availability to what does not yet have a name.

In a classical system, a contradiction is an error. In Tosio, a contradiction is data — the signal that a new understanding may be emerging, as is permanently the case in science. N7 listens to these shifts, unclassifiable zones of incoherence, identifies them, and treats them as hypotheses pending validation that it refines through its investigations.

Deductions and inductions must be validated by the user to be classified as progress. This is the difference between a machine that classifies and a machine that listens.

Innovation: N7 introduces epistemic humility into Tosio. Instead of rejecting contradictions as errors, it actively listens to them — measuring topological friction (W distance), suspending judgment on the unclassifiable, and exposing divergent perspectives. The Liminal Archive retains hypotheses that do not yet have a key, the BeliefNetwork evaluates source reliability over time, and the FrictionDetector makes contradictions resonate instead of resolving them prematurely. The Rashomon Report reveals truth as a collision of perspectives, not a consensus.

Use cases:

• Contradictory scientific discovery: a new experimental result contradicts a well-established theory in the Swarm. The FrictionDetector measures high friction, but the BeliefNetwork confirms the source's reliability. Rather than rejecting, N7 creates a Rashomon Report exposing both perspectives, suspends judgment in the Liminal Archive, and alerts human sentinels for arbitration.
• Contradictory personal memories: a user has two incompatible versions of the same event. N7 detects the friction, preserves both as plausible hypotheses in the Liminal Archive, and exposes the divergence in a Rashomon Report. The user arbitrates — or chooses to leave both versions suspended for future exploration.

The four epistemological pillars

Integration with the Epistemic Firewall (N4.5-bis)

N7 provides the epistemological primitives used by the Epistemic Firewall to analyze request content before memorization in the Swarm:

• BeliefNetwork → evaluates the information source's reliability
• FrictionDetector → measures contradiction with existing beliefs
• Liminal Archive → suspends judgment on the unclassifiable
• SentinelInterface → submits the contradiction to arbitration (pending) and logs it. A sentinel is registered at runtime (2026-10-02) — active president of the N6 Ethics Council. The human notification channel is not wired yet.

The full pipeline is described in the N4.5-bis — Epistemic Firewall section.

Validation

• 7 Python modules (~2000 lines): archive, belief, friction, rashomon, sentinelle, escalation, routing
• 73 N7 unit tests: cover the full cycle (belief -> friction -> contradiction -> arbitration -> Rashomon Report) and the Maturation Engine (CANDIDATE->LOCKED pipeline, 4 certificates per promotion)
• N4/N5/N6 integration: BeliefEpistemique inherits from EvaluationEthique (N6), FrictionDetector uses W topology (N4), SentinelleInterface delegates to the Human Ethics Council (N6)

Why it matters

N7 gives Tosio a rare property: epistemic humility. The system knows it does not know everything. It retains what it does not yet understand, exposes its own contradictions, and waits for a human — or a future discovery — to shed light.

Maturation Engine — from traces to beliefs (July 2026)

The Maturation Engine automates the transition from unclassifiable trace to verified belief, with 4 certification steps using existing Tosio tools: W topological proximity (FrictionDetector), temporal recurrence (resonance_count), soliton anchoring (ForkLeger + PolarisSoliton), and friction stability (BeliefNetwork scan). Optionally, N6 ethical validation (PontN6Protee) can be activated to automatically reject beliefs incompatible with the CadreDeValeurs.

Automatic calibration adjusts thresholds from archive statistics (resonance percentiles, residual distribution). Each maturation generates a complete audit trail with 4 timestamped certificates.

N7 user guide →

Polaris is a topological reference soliton — a stable beacon in W space (N4) that allows the Swarm to orient itself, detect absolute drift, and provide macro feedback to users.

Innovation: Polaris provides the first absolute reference frame for the Swarm, independent of any particular instance. From a public seed-phrase and a reproducible topological computation, it generates a stable soliton trajectory that each node can measure. This reference enables detecting collective drift — invisible to relative comparisons — and allows clockless synchronization: two instances can compare their alignment without exchanging their W position, using only their cryptographic proof chains (N4).

Use cases:

• Air-gap synchronization: two Tosio instances in separate organizations cannot exchange data. They compute their distance to the Polaris soliton and exchange only this value — cryptographically signed via N4. Comparing distances proves alignment without revealing the topological position or sensitive knowledge of each instance.
• Silent collective drift detection: all Swarms in a federation drift progressively due to a bias in an embedding model update. Without Polaris, no Swarm notices (relative positions remain stable). N8 detects that absolute alignment to the soliton is degrading uniformly across all instances, triggering coordinated recalibration.

Why a beacon?

The Supervisory Board (N5) detects relative drift between Swarms. But if all Swarms drift together in the same direction, nobody sees it. Polaris provides an absolute reference, independent of the system, based on a public seed-phrase and a reproducible topological computation.

Three functions

System integration

• N5 (Cartographer): a 5th map — the Alignment Map — displays average distance, standard deviation, and 7-day trend
• N7 (Mysterium): the distance to the soliton is treated as a special friction between the node and the anchor belief "Polaris"
• API: GET /rapport exposes alignment metrics for dashboards and monitoring
• Dashboard: interactive interface available at dedicated interface

Applications

• Signal calibration: the soliton can serve as a reference for spread trading, stock market synchronization, or distributed update coordination
• Correlation clock: two entities can synchronize without a central server, solely via their relationship to the same soliton
• Collective drift detection: if all Swarms drift away from the beacon, an alert is triggered

Validation

• 91 unit tests pass (34 soliton, 20 synchronization, 15 N5 integration, 7 N7 integration, 15 transition protocol)
• Full reproducibility: same seed -> same trajectory (deterministic via SHA-256)
• Mathematical properties: closed loop (period T=7d), normalized vectors, cosine distance in [0,1]

N8 user guide →

N9 — The Topological Watcher is the layer that adds time and urgency to the equation system. Where N8 (Polaris) provides an absolute reference and N7 listens to contradictions, N9 acts. It monitors incoherences, explores weak signals, and reacts to unexpected events with an emergency shield.

Innovation: N9 adds urgency and autonomous action to the equation system. Where N7 listens and N8 provides the reference, N9 acts — monitoring incoherence pressure with exponential logistic growth, stochastically exploring weak signals, and executing pre-approved defensive actions (freeze, quarantine, rollback) when critical thresholds are crossed. The full PDCA loop (Plan-Do-Check-Act) connects actions to feedback, adjusting thresholds for future cycles — first implementation of this loop in Tosio.

Use cases:

• Cascading corruption crisis: an injection of defective nodes triggers a cascade of incoherences. The Exponential Monitor detects that pressure doubles each cycle. At the CRITICAL threshold (0.82), the Topological Shield automatically freezes the affected zone, and the Light Fork launches a parallel diagnostic in isolated micro-environments — all in 3 cycles, preventing corruption from spreading to the entire Swarm.
• Early warning on weak signals: the Stochastic Explorer identifies a pattern of recurring but low-intensity anomalies in a rarely visited region of the Swarm. It alerts the Supervisory Board (N5) with a "curiosity" recommendation, triggering a preventive investigation before the anomalies become true incoherences — a proactive defense that avoids subsequent emergency actions.

The 5 operational phases

Pressure formula

Pressure = Sigma(weight_incoherence_i x 2^(cycles_i / tau))

    with:
    - weight_incoherence_i: severity of the incoherence (normalized W distance)
    - cycles_i: number of cycles without resolution
    - tau: time constant (default 10 cycles)

    P1 < 0.6 : NORMAL -- passive surveillance
    P1 >= 0.6 : WARNING -- activate P2 (exploration)
    P1 >= 0.8 : CRITICAL -- activate P3 (shield)
    P1 >= 0.95 : EMERGENCY -- activate P4+P5, escalation N5+N6

System integration

• N5 (Supervisory Board): receives N9 alerts via the PDCA Bridge, updates the risk map
• N6 (Ethical Governance): validates shield actions (FREEZE, QUARANTINE, SHRED) via the EDR process
• N7 (Mysterium): incoherences detected by N9 are transmitted to the FrictionDetector
• N8 (Polaris): the distance to the reference soliton is a key indicator monitored by the P1 Monitor

Validation

• 63 unit tests pass (16 monitor, 14 explorer, 13 shield, 8 fork_light, 12 pdca_bridge)
• 5 operational phases validated independently then in integration
• Architecture: 5 Python modules (n9_veilleur/) integrated into swarm_memory

N9 user guide →