Nous renders thought,
not tongues.
A decentralized protocol extracting traceable cognitive artifacts from internet noise.
Identity Stripped. Emotion Neutralized. Logic Distilled.
// Unease is a signal, not a bug.
Nous guarantees Traceability of Thought."
The Why Now
The Anti-Thesis
"The internet is a library on fire.
We are building the fireproof vault."
Information Entropy
Cognitive Order
HASH: SHA256(claim + evidence + reasoning)
The old internet optimizes for engagement.
Nous optimizes for truth.
The difference is not incremental. It is civilizational.
Protocol Manifesto
The Foundational Document
What is Nous Protocol?
A decentralized cognitive standard that transforms unstructured inputs into traceable, auditable artifacts. It defines standards for truth verification, consensus, and storage — distinct from the application layer where decisions and actions occur. Nous maps the terrain; it does not drive the vehicle.
(Artifact types are extensible by specification. What downstream systems do with these artifacts—search, planning, decisioning, automation—is implementation-defined.)
Key Guarantees
NousProtocol is a standard, not an application. It defines canonical formats and semantics for producing traceable cognitive artifacts (claims, uncertainty, provenance, receipts). What downstream systems do with these artifacts—search, planning, decisioning, automation, education—is implementation-defined.
Nous is a standard. Implementations may be modest, domain-scoped, and local-first.
Stewardship: Stewardship of the specification MAY be performed by any neutral process. The protocol identity remains independent of any single entity.
Core Definition
CORE_DEFonly the trace of thought.
Identity Lock
ID_LOCKNous is not a product, platform, or brand. It is a protocol-level identity that exists prior to implementation and survives after products fade.
Non-Servile Standard
NON_SERVILERefuses three dangerous forms: replacing your judgment, erasing reasoning paths, using "correctness" to override traceability.
"Nous doesn't exist to optimize your happiness; it exists to preserve your autonomy by filtering the noise that aims to steal it."
Protocol Characteristics
Mill is kinetic input, Pool is potential storage, Base is mass sedimentation.
Extraction Engine, Aggregation Field, Single Source of Truth.
Mill can have multiple instances. Pool can be deepened. Base points to ecosystem lock-in.
Protocol Constraints
Strict Axioms for Protocol Integrity
Nous is not a user interface or commercial application. It is a headless cognitive infrastructure.
- •Traceability: Every output traced to inputs and processing steps
- •Auditability: All processing decisions recorded and verifiable
- •Immutability: Committed records cannot be altered without detection
- •Uncertainty: Confidence levels explicit, not hidden
The Babel Shield
Identity bias stripping via forced localization. We render meaning, not foreign text.
- × No language switchers
- ✓ "Nous-Mediated" only
- ✓ Native truth rendering
The Visual Veil
Prevention of sensory hijacking. Images are dimmed, data is brightened.
- ✓ Default gaussian blur
- ✓ AI Logical descriptions
- ★ Vector Supremacy (SVG)
"Eye candy is the vector of manipulation. Aesthetics hijacks System 1; Logic defends with System 2."
The No-Ad Axiom
Zero monetization of human attention. The attention economy is a debt bubble. We are the default. Logic is the only currency.
- ✓ Commercial filtering
- ✓ Marketing incineration
- ✓ Logic only
Perception Layer
Logic over Lineage
The protocol defines three mandatory functional layers. Any compliant system seeking L5 Autonomy Certification MUST implement these logic interfaces.
Mill
Extraction Specification
Raw Input → Logic Tensors
Pool
Consensus Layer
Multi-Agent Verification
Base
Axiom Ledger
Immutable State Root
Client
Interface Layer (L4)
External Applications
The Protocol defines the standard for Truth verification. It does not define how Truth is acted upon. Action is the domain of the Client.
"Bros, this coin is about to moon! The team has strong backing, get in now! 🚀🚀🚀"
"Claim: Asset price increase imminent. Basis: Project team background. Status: Unverified speculation."
Operators react to the Author's Tribe (Chinese/English, Crypto/TradFi). Bias filters enable "Us vs Them" thinking. Logic is secondary to Identity.
Operators react to the Proposition. Language is treated as a rendering preference, not a flag. Identity is hidden until traceability is requested in the Audit Chamber.
The Perception Layer performs Semantic Extraction, not "translation". After identity stripping, what remains is pure logical truth value — eliminating operators' obsession with translation quality and redirecting focus to logic validity.
Claim Structure (Minimum Viable Definition)
Extraction Capability Boundaries
Semantic Extraction is NOT a claim of universal meaning preservation. The following limitations MUST be documented:
- • Sarcasm/Irony: modality SHOULD be "alleged" + low confidence
- • Ambiguous References: MUST flag unresolved pronouns
- • Cultural Idioms: MAY lose nuance; original preserved in audit
- • Legal/Religious Text: SHOULD preserve original + extraction
- • Claim Graph: structured propositions with edges
- • Qualifier Tags: time/location/probability markers
- • Evidence Links: traceable citation references
- • Uncertainty Field: machine-readable confidence
Babel Shield
SEMANTIC COLLAPSE · LATENT SPACE OPERATION
Not "translation" — it is Semantic Collapse. All input must be collapsed into culture-agnostic logic tensors in the Latent Space before entering Mill. The shield strips Rhetoric and preserves only Logic Terms.
Visual Veil
VECTOR SUPREMACY · SYSTEM 2 FORCING
All images are converted to logical topology graphs. Human response to images is intuitive (System 1 - Kahneman); response to descriptions is logical (System 2). The Veil forces the brain to switch to System 2. Cognitive load increases deliberation.
Identity Delay
LOGIC CREDIBILITY SCORE · DELAYED REVEAL
Author identity is cryptographically sealed until explicitly requested. Before reveal, system provides an anonymous Logic Credibility Score based on author's historical logic "hardness" and acceptance rate.
Human cognition is riddled with systematic biases. The Perception Layer actively neutralizes these attack vectors.
Authority Bias
"A PhD said it, so it must be true."
Credentials hidden until logic is evaluated.
In-Group Bias
"They're one of us, so I trust them."
Language/culture markers normalized.
Halo Effect
"The presentation is beautiful, so the content must be good."
Aesthetics stripped; logic prioritized.
Bandwagon Effect
"Everyone believes it, so it must be true."
Multi-agent verification, not popularity.
"AI will surpass human intelligence by 2030."
"CLAIM: Artificial General Intelligence timeline prediction — 2030. BASIS: Speculation. EVIDENCE: None cited."
Result: The same claim is evaluated on its logical merit, not on who said it or how many people liked it.
Babel Shield
Visual Veil
Identity Delay
"We do not hide what was said. We hide who said it."
Identity is always recoverable in the Audit Chamber. Anonymity is a UI choice, not a data loss.
Deterministic Execution Gate
While NousProtocol is a passive ledger, it defines standards for "actionable state". External agents (e.g., autonomous executors) MAY use NousBase as a safety interlock.
Traceability broken or Hardness < H2. No action permitted.
Traceability intact, Hardness ≥ H3, Consensus > 95%. Safe to consume.
Scenario Branching
The ledger supports hypothetical
branches.
Users can fork from mainnet to run "what-if" simulations. These branches are ephemeral and
do not merge back unless explicitly validated.
Core Principles
The Axioms We Do Not Break
Immutability
Once committed to Base, truth cannot be edited, only appended to. History is a write-only log.
Traceability
Every insight must link back to its source. No black boxes. The "Why" is more important than the "What".
Decentralization
Truth is not owned by a server. It is verified by a network of independent adversarial agents.
Evolution
The protocol assumes it is wrong. It is designed to be corrected. Base commits evolve over time.
Non-Servility
The UI is fluid, but the Protocol is Non-Servile. We prioritize logical density over user comfort. The truth doesn't care if you like it.
Antifragility
Noise makes the filter stronger. Attacks on validity improve the immunity of the consensus.
§3.1 Trace Receipt NORMATIVE
Every processing operation MUST produce a valid Trace Receipt.
{
"version": "string", // MUST be "1.0.0"
"trace_id": "uuid", // MUST be unique (UUIDv4)
"timestamp": "date-time", // MUST be ISO8601 UTC
"input_hash": "hex64", // SHA-256 of canonical input
"output_hash": "hex64", // SHA-256 of canonical output
"processor": {
"type": "string", // MUST be registered processor type
"id": "string",
"version": "string"
},
"parent_traces": ["uuid"], // MAY reference parent traces
"metadata": { // MAY include namespaced extensions
"additionalProperties": true
},
"signatures": ["Signature"] // MUST have ≥1 valid signature
}REQUIRED Fields
version,trace_id,timestampinput_hash,output_hash,processor
OPTIONAL Fields
parent_traces,metadata,ttl
Extension Rules
- • Implementations MAY include additional
fields in
metadata - • Extension fields MUST use namespaced keys
(e.g.,
org.nous.*,ext.*,x_*) - • Extensions MUST NOT change the semantics of required fields
- • Implementations SHOULD ignore unknown extension fields unless explicitly required by conformance profile
Processor Type Registry
Future versions may define additional processor types. Implementations SHOULD ignore unknown types unless explicitly required.
§3.2 TruthNode NORMATIVE
Once committed, a TruthNode MUST NOT be modified; only superseded or retracted.
{
"id": "hex64", // MUST be SHA-256 of canonical content
"version": "semver", // Protocol version
"content": "string", // MUST be non-empty
"hardness": "H0-H4", // Epistemic strength classification
"consensus_score": "0.0-1.0", // Validator agreement
"reasoning_chain": ["Step"], // SHOULD include derivation
"sources": ["Citation"], // MUST have ≥1 for H2+
"logic_hash": "hex64", // Semantic fingerprint (language-agnostic)
"created_at": "date-time",
"status": "active|superseded|retracted",
"supersedes": "hex64", // MAY reference previous version
"actionability": { // RESERVED: For L4 Agent consumption
"risk_level": "R0-R3", // Safety constraints for agents
"odd_tags": ["string"], // Operational Design Domain
"executable": "boolean" // Safe to act without human intervention
}
}Immutability Rules
- •
id,content,created_atMUST NOT be modified after commit - • Corrections MUST create new TruthNode with
supersedesreference
Artifact Types (Reserved)
Future versions may define additional artifact types beyond Claims. Implementations SHOULD ignore unknown artifact types unless explicitly required by the conformance profile.
§3.3 Hardness Classification NORMATIVE
The Cold Truth
PHILOSOPHICAL REFLECTION ON COGNITIVE SOVEREIGNTY
In the AGI era, humanity faces not a scarcity of information, but an epidemic of curated feeding. Commercial AI optimizes for helpfulness — pleasing the user, confirming biases, maximizing engagement.
NousProtocol optimizes for logical density. This is the fundamental conflict.
The Ultimate Question: If Nous proves a truth that makes all of humanity uncomfortable (per our Root Axiom: "Check the Why"), does Nous have permission to modify Base?
Answer: Per §3 Principle 01, Immutability supersedes human emotion. NousBase is cold. It does not care about comfort. It records what survives adversarial verification.
This coldness is not cruelty — it is the last firewall of cognitive sovereignty in an age where every other system is designed to manipulate, soothe, or sell. NousBase is the mirror that does not lie.
The Trinity
The Physics of Truth
This is the first official implementation of the protocol under current technical conditions. The protocol is abstract; implementations are concrete. Future versions may include a Rust-rewritten Mill or blockchain-based Pool, but they all belong to the Nous Protocol ecosystem.
NousMill
The Cognitive Factory
Crush noise, apply pressure, produce Alpha. A private node running the protocol, responsible for high-pressure extraction of raw information into cognitive essence.
NousPool
The Reasoning Pool
Distribute reasoning, community collaboration, multi-model consensus. The distribution layer of the protocol, responsible for verifying and distributing ground reasoning through community review and challenge systems.
NousBase
The Immutable Ledger
Not a database — a high-density truth crystal. It stores not data, but the residual logic that survived tens of thousands of adversarial derivations. The Base Oil.
git
revert (append inverse commit) but MUST NOT git rebase
--force (rewrite history).
Base Oil has viscosity — the more core the truth, the higher its logic viscosity. Modifying high-viscosity truths requires exponentially more Cognitive Gas.
Energy Model Closed-Loop
This is not literary metaphor — it is conservation. A chain that holds from thermodynamics to information architecture.
"The brain thinks, the network validates, the ledger remembers."
Mill = Cognition | Pool = Consensus | Base = Memory
System Definition
System Definition, Runtime & Telemetry
The Nous Protocol
A non-servile cognitive standard. It does not sell Truth; it creates an immutable trace of how a thought came to be.
"If Nous makes you uneasy, don't fix it. Check the Why."
If we make you uneasy, don't try to change it. Explore the reason behind it.
NousMill / NousPool / NousBase
Energy (Kinetic) → Matter (Mass) → Field (Potential)
Entropy / Dark Matter / Kernel / Sunset
"We don't sell Truth; only the trace."
What you have now is not a "finished system,"
but a starting point that allows future negation without corruption.
The Nous Protocol is now in position.
Runtime
Identity Stripped at the Edge, Never at the Core
Semantic Topology Extraction
PROTOCOL-LEVEL DEFINITION
The Visual Veil is not merely blur — it is semantic reconstruction. All images are force-converted through a vision encoder into SVG semantic topology graphs.
Operators see logic topology, not aesthetic manipulation. This severs the visual-emotional hijack vector completely.
Logic Engines
Internal Mechanics of Each Layer
The Adversarial Press
NousMill doesn't "process" data — it tortures it.
Pressure comes from an Adversarial Loss Function — only logical statements that cannot be falsified under multi-model cross-examination qualify as "Essence".
- • Raw Seeds: Global collection from diverse sources.
- • Torture Protocol: Multi-LLM cross-examination with adversarial prompts.
- • Survival Criterion: Only unfalsifiable logic survives to become Essence.
Deep Probe Capability (Optional Implementation)
While the reference Mill processes passive inputs, the protocol specification allows for "Active Probe" implementations. These are recursive Mills that can dynamically generate queries during the torture process to fill information gaps.
The Metabolism
How to avoid stagnation?
The Pool is a dynamic equilibrium system, not an infinite list.
- • Evaporation: Essence has a TTL. No interaction = disappearance (48h). High engagement extends TTL.
- • Sedimentation: High heat + longevity triggers a phase change into Solid Blocks (Base).
The Consensus Ledger
Who defines truth?
Base is not a database; it is a Version Control System for Wisdom.
- • Governance: Truth is fluid (v1.0 → v2.0 → v3.0).
- • Pull Request: New essence challenges old entries.
- • Authority: Derived from how many older versions it defeated.
Context Overlay ("Map" Primitive)
The protocol supports "domain overlays". Users can project the high-dimensional truth graph onto specific low-dimensional planes (e.g., strategic landscape, risk terrain).
{{ proofReceiptJson }}
Protocol In Action
How the Stacked Trinity Serves Different Actors
{{ sector.title }}
{{ sector.before }}
{{ sector.after }}
The Flow
From Noise to Truth
The Flow is the cognitive pipeline that transforms raw internet chaos into verified, immutable truth. Every piece of information must survive this gauntlet before earning its place in NousBase.
Raw Input
Internet Noise
NousMill
Extract & Refine
NousPool
Verify & Consensus
NousBase
Immutable Truth Storage
Ingestion Layer
BABEL SHIELD ACTIVE
- • Language normalization (source hidden)
- • Visual Veil: images → semantic descriptions
- • Author identity delayed reveal
Adversarial Press
MULTI-LLM SYNTHESIS
- • ≥3 independent validators MODEL FAMILIES IMPL-DEFINED
- • Entropy injection (5% Heretic Quota)
- • Dialectic contradiction detection
Consensus Engine
PROOF OF LOGIC
- • Stake-weighted voting mechanism
- • TTL expiration for unverified claims
- • Merkle root commitment to Base
"What survives the press is not truth — it is the best approximation we have today."
Tomorrow's Base commit may obsolete today's SOTA. The protocol assumes it is wrong.
Processing Stages NORMATIVE
Implementations MUST process content through these stages in order:
Raw input reception. MUST compute input_hash before any transformation.
Identity stripping. SHOULD remove author identity markers before semantic processing.
Image semantic extraction. SHOULD replace images with semantic descriptions.
Multi-agent extraction. MUST use ≥3 independent validators for H2+ content.
Community verification. MUST reach threshold before Base commit.
Ledger commitment. MUST generate Trace Receipt and compute output_hash.
Stage Transition Requirements
- • Each stage MUST produce a Trace Receipt linking to previous stage
- • Failed stages MUST halt pipeline and record failure reason
- • Stages MUST NOT be skipped for content targeting H2+ hardness
§ Consensus Mechanism (Minimum Viable Definition)
- • MUST: ≥3 independent LLM validators
- • MUST: No single model family >50% weight
- • SHOULD: Include ≥2 model architectures
- • MAY: Include human expert validators
- • ACCEPT: ≥N% agreement (N configurable, default 66%)
- • REVIEW: 50-N% agreement → human escalation
- • REJECT: <50% agreement → discard or retry
- • CONFLICT: High confidence + disagreement → flag
§ Hardness Classification Rules
| LEVEL | NAME | CRITERIA | MODIFICATION COST | EXAMPLES |
|---|---|---|---|---|
| H0 | Fluid | Unverified, single source | 1× base | Raw claims, rumors |
| H1 | Soft | Multi-source, consensus <66% | 2× base | News, opinions |
| H2 | Firm | High consensus, reproducible | 4× base | Verified facts, data |
| H3 | Hard | Cross-gen consensus, cited | 8× base | Established knowledge |
| H4 | Diamond | Axiomatic, foundational | 16× base | Math axioms, physics laws |
§ Visual Veil Capability Boundaries
- • Charts, graphs → structured data points
- • Tables → key-value pairs
- • Diagrams with text → relationship graphs
- • Screenshots of text → OCR extraction
- • Artistic imagery → aesthetic, not factual
- • Faces/identity → privacy boundary
- • Memes/satire → context-dependent meaning
- • Manipulated media → detection required first
extraction_confidence score. Content below threshold MUST be
flagged for human review or discarded.
§ Anti-Advertisement Detection Criteria
promotional_score attached. Consumers MAY filter based on
threshold.
§ Privacy & Compliance Requirements
- • PII MUST be stripped before Mill processing
- • Original sources MUST be hashed, not stored verbatim
- • User queries MUST NOT be logged with identity
- • Audit trails MUST be pseudonymized
- • GDPR: Right to erasure → mark as RETRACTED
- • CCPA: Disclosure of data sources on request
- • Content with geo-restrictions: flag jurisdiction
- • Cross-border data: document transfer basis
Value Stream — From Noise to Truth
The Value Stream represents the complete cognitive pipeline that transforms raw internet chaos into verified, immutable truth. Each stage applies increasing pressure, filtering out noise while preserving signal.
Input (Noise)
Raw data ingestion from diverse sources.
Press (Mill)
High-pressure adversarial extraction.
Flow (Pool)
Network validation with TTL countdown.
Solidify (Base)
Immutable truth committed to ledger.
Pressure Creates Value
Higher extraction pressure means lower noise. The rejection rate is not a bug—it's the core mechanism that separates signal from noise.
Time-Bound Validation
Every artifact in the Pool has a TTL. Without network validation and engagement, content evaporates—preventing infinite stagnation.
Immutable Provenance
Once committed to Base, every artifact carries a complete trace of its journey—from raw input to final consensus.
"The value of truth is inversely proportional to the noise it survived."
Higher rejection rate = Higher signal purity = Greater trust coefficient
Operator Journey
From Private Insight to Public Consensus
01. Input (The Layer)
NOUSMILL / PERSONAL FACTORY
Operator's NousMill Node auto-fetches Bloomberg RSS.
Raw signal captured: "TSMC 2nm production delayed by 6 months."
> SOURCE: BLOOMBERG_MKTS
> CONTENT: "2nm Yield Issues..."
> ANALYZING: TSMC_DELAY
> RESULT: NVIDIA_POSITIVE (Moat+)
02. Process (The Logic)
CONTEXTUAL REASONING
System mounts operator's "Semiconductor Thesis".
AI deduces: "Impact: Negative for TSMC (Yield);
Positive for Nvidia (Moat).
Sentiment: Stripped."
03. Distribution (The Wire)
NOUSPOOL / REASONING POOL
Operator publishes insight. Community Forks logic.
Other nodes append Samsung Yield Data to
reinforce the thesis.
04. Consensus (The Base)
NOUSBASE / TRUTH LEDGER
The multi-agent verified insight is Merged into the
global "Semiconductor Cycles" entry.
New industry consensus established.
The Audit Chamber
Don't Trust. Trace.
step: "{{ activeTraceStep }}",
timestamp: {{ Date.now() }},
signature: "0x9f8a...b2c1",
validators: ["Claude-3", "GPT-4"]
};
Truth-DAG: Recursive Audit
TRACEABILITY CHAIN · EVIDENCE HARDNESS LEVELS
Every Trace Receipt supports recursive audit. When viewing a conclusion in the Audit Chamber, each cited source must possess its own Trace ID. This forms a Truth Directed Acyclic Graph (Truth-DAG). The recursion terminates at H4 (Diamond) nodes — axioms that require no further proof.
H4 is not merely "axiom" — it is Self-Proving Logic. Examples: mathematical identities (1+1=2), thermodynamic laws (entropy always increases), logical tautologies (A ∨ ¬A). NousBase's mission is to ensure the entire edifice of human cognition is anchored to these Diamond nodes.
When viewing a Base conclusion, one-click expands all underlying logic nodes it depends on.
Every node is independently verifiable. Click any conclusion to see it explode into a tree diagram, revealing underlying H4 axiom nodes. Traceability becomes visceral.
Recursion terminates at H4 Diamond nodes — self-proving logic requiring no further proof.
if (node.hardness === 'H4') return { type: 'AXIOM', proof: 'SELF_EVIDENT' };
return { node, children: node.sources.map(auditRecursive) };
}
Deep Dives
The Trinity in Technical Detail
Each component of the Trinity serves a distinct purpose in the cognitive pipeline.
Together, they form an unbreakable chain of extraction,
verification, and preservation.
The Mill
§6.1 EXTRACTION ENGINE
Adversarial extraction through entropy injection. Where raw information is pressed into logic-dense essence.
The Pool
§6.2 REASONING POOL
GitHub for Reasoning. Where AI-extracted insights are distributed, forked, and SOTA-certified by the community.
The Base
§6.3 IMMUTABLE LEDGER
Version-controlled truth storage. Where verified knowledge is committed to an immutable, auditable ledger.
Deep Dive: The Mill
The Cognitive Factory
Raw information is noise. Single-model outputs are biased.
NousMill is your private Cognitive Factory —
where multiple AI agents collaborate to extract, refine, and synthesize intelligence.
We don't just answer. We argue, verify, and distill.
Mechanism 1: Multi-Agent Orchestration
No single model is the oracle. NousMill orchestrates multiple AI agents in parallel, each contributing unique strengths. The result is cross-validated, bias-reduced intelligence that no single model could produce alone.
Mechanism 2: Cognitive Lenses
Apply different analytical frameworks to the same information. Like X-ray vision for data — reveal what's hidden beneath the surface.
Mechanism 3: The Grinding Process
Watch information condense into an immutable hash. The Logical Fingerprint changes only when the logical derivation changes — not when text is rephrased. Semantic equivalence is preserved.
Mechanism 4: Diverse Data Ingestion
Philosophy: Glass Box
"We don't just answer. We argue, verify, and distill."
Ingest → Grind → Refine → Crystallize → Ship
Deep Dive: The Pool
The Reasoning Pool · GitHub for Reasoning
AI can extract. But only humans can validate meaning.
NousPool is the Community Verification Layer —
where Mill's Essence meets human judgment.
Think Knowledge Forums + Content Discovery + Q&A Communities,
but for AI-extracted insights.
Mechanism 1: Human Verification
"Expert" status is earned through historical logic contributions (Base adoptions), not real-world identity. No new power centers — only proven reasoning.
AI extracts, humans validate. Mill produces Essence through multi-model processing. Pool lets the community verify, challenge, and endorse — ensuring human judgment remains the final arbiter of truth.
For controversial or nascent Essence, the protocol increases flow friction in the Pool, providing a "cooling observation period" for the human community. This prevents AI processing speed from causing global cognitive shocks. Truth takes time.
Mechanism 2: Social Interactions
Like Content Discovery meets Knowledge Discussion — but for verified intelligence. Every interaction builds reputation and shapes consensus.
Mechanism 3: Truth Pipeline
The path from raw insight to Base commitment. SOTA is only granted in Base after Mill + Pool dual consensus.
Mechanism 4: Challenge System
Any reasoning in the Pool can be challenged. This is not trolling — it's structured dialectic. Challengers must provide counter-evidence, and the community votes on the outcome.
Mechanism 5: The Feed
"Mill refines. Pool validates. Base certifies SOTA."
Mill → Pool → Base (SOTA)
Deep Dive: The Base
Version Control for Truth
GPT's truth ≠ Claude's truth. Expert opinion ≠ Beginner's reality.
Without a benchmark, the world is a wasteland of noise.
NousBase stores the Base Oil — the current
industry consensus.
Mechanism 1: The Knowledge Graph (Git for Truth)
Knowledge Depreciation
ENTROPY WARNING · OXIDATION VISUAL
Truth stored in Base is not eternal. If a fact remains unchallenged and uncited for extended periods, its hardness decays. The system triggers an Entropy Warning — the node's icon displays an "oxidation" visual effect, alerting the community that this knowledge is losing vitality and needs re-pressing.
RE-PRESS: Any user can cite the rusted node with new evidence, resetting decay timer.
This prevents Base from accumulating outdated dogma. Knowledge must earn its permanence through continuous relevance.
Mechanism 2: Fork & Blend
Users come to NousBase not to "memorize" truth, but to "stock up". NousBase doesn't force users to accept "standard answers" — it provides a "high-quality draft that can be modified".
NousBase stores Base Oil — the industry consensus, a public good maintained by the entire community. It is the reference standard.
Your Blend is Private Brew — Base Oil + your personal context. It is your cognitive sovereignty, stored locally.
Context Overlay ("Map" Primitive)
DOMAIN PROJECTION · CLIENT RESPONSIBILITY
The protocol supports "Domain Overlays". Users can project the high-dimensional truth graph onto specific low-dimensional planes (e.g., strategic landscapes, risk terrain, decision maps).
Mechanism 3: Pull Request
If a user believes their "blended oil" formula is exceptional — even surpassing the original "base oil" — they can submit a Merge Request. This is the "best practices library maintained by all of humanity".
💡 Why Do We Need NousBase?
GPT's best practices ≠ Claude's best practices. Expert's best practices ≠ Beginner's best practices. Without a benchmark, the world is a wasteland of noise.
NousBase's responsibility is to store that "common denominator" — the "Base Oil". It is the purest standard product after NousMill's high-pressure extraction, with all impurities removed. It represents "the current industry consensus".
Technical Specifications
Data Structures and Consensus Mechanisms
Data Structure
Unlike content hash, Logic Hash only changes when the logical derivation changes — not when text is rephrased. This enables true logic version control, not simple text diff.
Consensus Mechanism
Minimum 3 independent AI agents must validate before commit.
Expert human review can accelerate or veto consensus.
Dialectic synthesis when agents disagree. Both positions preserved.
Consensus above 95% marks content as current State-of-the-Art (High Hardness). 90% = Pool entry, 95% = Base commit.
Whether expressed in Chinese, English, or French — if the logical structure is identical, the generated Hash must be unique and consistent. This is the true meaning of "Rendering Thought, Not Tongues."
Schema: The Ration Card
Data Structure Specification
{
// Unique identifier for the thought unit
"id": "ration_x9d8s7",
// Type of anchor: 'seed' (raw input) or 'graft' (user extension)
"anchor_type": "seed",
// Auditable source trail (Not shown in primary UI)
"source_data": {
"provider": "Bloomberg",
"timestamp": 1705500000,
"url_hash": "a1b2c3d4"
},
// The Dialectic Core
"content_matrix": {
"user_query": {
"lang_source": "en-US", // Backend only
"text_en": "Is this sustainable?",
"text_zh": "Is this sustainable?" // Pre-generated native truth (Chinese)
},
"ai_response": {
"consensus_score": 0.92, // 0.0 - 1.0
"models_used": ["gpt-4", "claude-3.5"],
"text_en": "Analysis suggests...",
"text_zh": "Analysis suggests..."
}
},
// The Visual Veil (Anti-Sensory Hijacking)
"visual_veil": {
"has_image": true,
"image_semantic": "Graph showing revenue decline",
"vector_replacement": "<svg>...</svg>" // Vector Supremacy
},
"state": "essence" // or "slag" (discarded)
}Language source hidden from UI. Only rendered output visible.
Images replaced with semantic descriptions and SVG vectors.
Multi-agent agreement level. >0.85 = SOTA candidate.
Chain of Custody
A Truth is Only as Valid as Its Lineage
The Golden Thread
Every atom of essence in NousBase contains a cryptographic trace back to its origin. The chain is unbreakable.
Core Mechanics
Truth Has a Half-Life
Truth is not static; it has a half-life. NousBase assigns a temporal validity score to every artifact. Information must be maintained, or it rots.
- Fresh (Active) - High confidence, < 3 months
- Oxidized (Stale) - Needs re-verification, > 3 months
- Rusted (Deprecated) - Functionally obsolete, requires "Re-Press"
The Slag Heap (Incineration)
High noise destruction rate. Content identified as
purely extractive is effectively erased from user
cognition. No ads. No clickbait. No rage-bait.
Negative Entropy Burn: When a fallacy is
proven as Slag (Falsified Entropy), the Gas consumed to submit it is permanently destroyed, increasing average logic density
network-wide.
Dark Matter (Cryo-Preservation)
High-semantic value, zero engagement. Frozen ideas waiting for their context. These ideas are preserved, not deleted. Waiting for "Archaeological Retrieval" when future context aligns.
Dark Matter is not permanently dormant. When new information entropy produces semantic resonance (cosine similarity > 0.85) with a frozen node's vector embedding, that node is automatically thawed and re-enters the Mill. This embodies the protocol's adaptive evolution capability. Trigger: High-engagement new content semantically aligns with frozen thought.
thaw(frozen_node);
frozen_node.context = merge(frozen_node.context, new_input.context);
mill.ingest(frozen_node, {priority: "HIGH", reason: "RESONANCE_AWAKENING"});
}
This embodies antifragility: forgotten ideas resurface precisely when the world is ready to reconsider them.
Entropy Injection
The Dialectic Engine — We Don't Just Answer. We Argue.
Standard AI might agree to please the user.
"Correction: SMR (Small Modular Reactor) data from 2024 indicates a safety profile 100x higher than coal. Cost per kWh amortized over 60 years competes with wind storage. Bias detected: 'Chernobyl Syndrome'."
NousMill forces a logical confrontation.
"Nuclear energy presents a nuanced trade-off: while historical incidents created lasting public concern, modern SMR technology offers significantly improved safety profiles. A balanced energy policy should evaluate lifecycle costs, carbon footprint, and grid stability requirements rather than relying on emotional heuristics."
"Truth emerges from structured confrontation, not agreement."
— The Dialectic Principle
Quorum Requirements NORMATIVE
Mill (AI) Consensus
- • Minimum 3 independent AI agents
- • Agreement threshold: ≥66%
Pool (Human) Consensus
- • Minimum 5 human validators
- • Agreement threshold: ≥80%
The Minority Report
5% HERETIC QUOTA · MANDATORY
The 5% entropy is MANDATORY, not optional. It is Structured Dissent — the system must actively seek the weakest link in the logic chain and forcibly generate counter-arguments. When 4+ models reach ≥95% consensus, the system forcibly invokes a challenger agent. This prevents the system from falling into "echo chamber" effects.
- • Temperature: 1.0 (max creativity)
- • Architecture: Symbolic Logic / Non-neural
- • Prompt: "Find the flaw in this consensus"
System identifies the weakest node in the logic chain and generates targeted counter-arguments. This is not chaos — it is surgical dissent.
If all models converge, the system risks mode collapse. The Minority Report ensures cognitive diversity survives consensus pressure.
System Dynamics
The Living System
Overton Window
Adjust the window to see how truth classification shifts. Tracks rationality boundary drift over time.
"Yesterday's Heretic is today's Standard."
Dark Matter
High-value, low-engagement thoughts are frozen.
Positive Feedback
AMPLIFICATION LOOP
High-consensus truths attract more validators → Higher confidence → More citations → Stronger consensus.
Negative Feedback
DECAY LOOP
Low-engagement content loses TTL → Fewer validators → Faster decay → Eventually incinerated.
Homeostasis
BALANCE LOOP
System self-regulates: Too much consensus → Entropy injection. Too much noise → Pressure increase.
Decision Thresholds
Challenge Mechanism
Any validator MAY challenge a committed TruthNode:
- • Challenge MUST include counter-evidence or logical refutation
- • Challenged nodes enter REVIEW state pending resolution
- • Resolution outcomes: UPHELD, SUPERSEDED, or RETRACTED
Evolution Mechanics
SOTA Rotation
State-of-the-Art is not permanent. New evidence triggers re-evaluation. Old SOTA becomes historical reference.
Paradigm Shifts
When accumulated anomalies exceed threshold, the system triggers a paradigm review. Kuhnian revolutions are built-in.
Minority Reports
Dissenting views are preserved in Dark Matter. When consensus shifts, frozen thoughts can be thawed and re-evaluated.
Archaeological Mode
Every version is preserved. Future researchers can trace how truth evolved, understanding the "why" behind each shift.
Epistemological Modesty
"Nous does not guarantee correctness. Nous guarantees the traceability of why we think what we think. Yesterday's heretic may be tomorrow's prophet. The system evolves."
Protocol Economics
The Gas of Thought
INFORMATIVE NOTICE: Economic mechanisms described in this section are informative unless explicitly marked NORMATIVE. Implementations MAY adopt alternative cost models (e.g., compute credits, rate limits, subscription tiers) while preserving protocol guarantees.
The protocol defines the semantic constraints; economic enforcement is implementation-defined.
Cognitive Gas
Processing thought is not free. Every logical deduction, verification step, and synthesis requires Cognitive Gas. This creates Noise Resistance — the economic barrier that filters low-quality input.
TIERED PRICING: Query (1 Gas) < Challenge (5 Gas) < Fork (10 Gas) < Commit (50 Gas)
Critical Distinction: Gas vs. Application Fee
The Protocol defines Gas economics. Client applications define their own pricing. Separation of concerns.
Sybil Resistance
In a multi-agent consensus system, the greatest threat is fake nodes — 1000 puppet agents voting to distort truth. Nous defends with:
Prevents single-bias resonance in large-scale consensus. Validators must cryptographically prove model diversity (e.g., Transformer vs. Mamba SSM, GPT vs. Claude training lineage):
Proof of Logic (PoL)
Nodes are rewarded not for hashing random numbers, but for validating logical chains. A "Miner" in Nous is a "Thinker".
Energy Conservation
Nous operates on Compute Credits, not tokens. Manufacturing noise consumes massive energy (Gas), while verifying truth recovers energy. This asymmetric cost structure is the core defense — Entropy Reduction Proof.
Proof of Logic Visualizer
Watch how cognitive compute transforms noise into verified truth
Nous tokens are not speculation. They are crystallized cognitive compute — proof that logic was verified, not just generated.
Gas Flow Dynamics
Watch how Cognitive Gas circulates through the Trinity
Version Control Model NORMATIVE
Base implements Git-like semantics:
New version replaces old (like git commit)
Restore previous version (like git revert)
Mark as invalid (content preserved)
Immutability Guarantee
- • History MUST NOT be rewritten
- • All state changes MUST be append-only operations
- • HEAD pointer MAY be updated to reference different commits
Governance
Immutable Patches for Protocol Integrity
{{ patches.find(p => p.id === activePatch).desc }}
A Crystal of Responsibilities
The Ultimate Arbiter
THREE-LAYER CONSENSUS
How does truth emerge from chaos? Through a three-layer arbitration system: AI models first extract and cross-validate, humans then verify and endorse, and finally Git commits and versions the consensus.
"Truth is not declared by authority. It is distilled through layers of verification."
— The Arbitration Principle
Layer 1: Mill
AI ARBITRATION
Multiple AI models (Claude, GPT, Gemini, DeepSeek, Qwen) process the same input in parallel. Cross-validation eliminates single-model bias.
Layer 2: Pool
HUMAN ARBITRATION
Community members upvote, challenge, and endorse. Domain experts provide authoritative validation. Human judgment is the final filter.
Layer 3: Base
SOTA ARBITRATION
Git-style version control commits verified truth. SOTA is only granted here after Mill + Pool dual consensus. History is immutable.
Git for Truth
VERSION CONTROL OPERATIONS
⚡ When Layers Disagree
- • Mill vs Pool: Human judgment overrides AI consensus
- • Pool vs Base: New commit creates version branch
- • Deadlock: Fork allowed, parallel truths coexist
✓ Resolution Hierarchy
- 1. Base history is always preserved (immutable)
- 2. Human consensus takes precedence over AI
- 3. High LRS endorsement weighs more than votes
Hardness Promotion NORMATIVE
Hardness Decay
TruthNodes SHOULD decay if uncited:
- • 90-180 days: COOLING state (visual indicator)
- • 180-365 days: OXIDIZED state (entropy warning)
- • >365 days: RUSTED → demote to Pool for re-verification
Versioning Policy
Semantic Versioning
Protocol versions follow MAJOR.MINOR.PATCH (SemVer 2.0). Breaking changes increment MAJOR; backward-compatible additions increment MINOR.
Compatibility Guarantees
Implementations MUST accept receipts from same MAJOR version. SHOULD provide migration paths for MAJOR upgrades.
Deterministic Execution Gate
While NousProtocol is a passive ledger, it defines the standard for "actionable state". External agents (e.g., autonomous executors) can use NousBase as a safety interlock.
RED STATE
- • Traceability broken OR
- • Hardness < H2 OR
- • Consensus < 80%
GREEN STATE
- • Traceability intact AND
- • Hardness ≥ H3 AND
- • Consensus > 95%
CRITICAL DISTINCTION:
"The Protocol defines the signal, but it does not pull the trigger."
Execution responsibility lies with the Client Layer (L4). The Protocol provides the verification proof; the Client decides whether to act.
"Mill refines. Pool validates. Base certifies SOTA."
Protocol Defense
Threat Model & Mitigations
Roadmap & Use Cases
The Path to Cognitive Infrastructure
{{ phase.title }}
{{ phase.desc }}
Protocol Maturity Checkpoints
Registry Considerations
This section describes registries that MAY be established for protocol extensibility. No central authority is required; implementations MAY maintain local registries.
Processor Type Registry
New processor types MAY be registered via
specification amendment.
Reserved prefix: ext.* for implementation-specific
types.
Artifact Type Registry
Cognitive artifact types beyond TruthNode MAY be defined in future versions.
Reserved prefix: x_* for experimental types.
Protocol Glossary
Shared Vocabulary for Cognitive Infrastructure
{{ item.term }}
{{ item.type}}{{ item.def }}
Join the Protocol
Become Part of the Cognitive Standard
Build the Infrastructure
for Collective Intelligence
NousProtocol is open. The specification is public. We're looking for builders, researchers, and truth-seekers to help construct the cognitive layer of the internet.