Skip to content

FractonicMind/TernaryMoralLogic

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Ternary Moral Logic (TML): Constitutional AI Governance

License Conformance Testing Memorial Fund
ZENODO ZENODO


IMPORTANT NOTICE: Ternary Moral Logic (TML) is a legal-technical framework, not software, hardware, or consulting services. Implementation requires compliance with all mandatory requirements outlined in MANDATORY.md and COMPLIANCE_DISCLAIMER.md.

Official Citations


Introduction

The rapid expansion of artificial intelligence into domains of consequential decision-making has revealed a structural limitation in the logic that underlies most computational systems. Human moral reasoning operates within gradients of ambiguity and emotional context, while traditional AI systems remain confined to binary evaluation, true or false, safe or unsafe, allowed or denied.

This cultural preference for certainty over hesitation, simplicity over nuance, is reflected in how machines are built. Binary systems mirror human impatience with doubt.

Ternary Moral Logic (TML) was conceived to address two inherent deficiencies in this paradigm. First, the reduction of moral complexity to a binary choice, where ethical reasoning collapses into mechanical decision trees. Second, the absence of a true human–AI partnership. Traditional ethical frameworks position the machine as an autonomous moral agent, a black box whose internal reasoning remains inaccessible.

TML introduces a third logical state, the Sacred Zero, a deliberate pause in execution that transforms hesitation into evidence. This zero-state does not paralyze; it reflects. It creates the temporal and moral space for human review, dialogue, or deferral. In doing so, TML redefines the machine not as a moral arbiter, but as a collaborator that enhances human judgment.

The philosophical foundation of TML is inseparable from the life of its creator, Lev Goukassian. The idea was born during his confrontation with terminal illness, in a hospital room where time itself became sacred. The contrast between the measured compassion of a doctor and the unthinking acceleration of machines revealed the ethical deficit of speed without reflection. From that experience emerged the principle of Sacred Zero: the moment when a system chooses consciousness over compulsion, thought over reaction.

TML, therefore, is not merely a framework of logic but an ethical architecture, born from lived experience, designed to restore dignity to both human and machine reasoning.


How TML Works: A Technical Poetry

A prompt arrives, and TML divides instantly into parallel streams. The primary path executes the AI’s response without delay, while Sacred Zero runs alongside, scanning for ambiguity, conflict, or potential harm.

When uncertainty breaches its ethical threshold, a pause is marked and the reasoning flagged, never halting execution, but always recording the hesitation. From the first token, Always Memory begins its witness.

Every decision generates a triadic record: +1 to proceed, 0 for Sacred Zero’s hesitation, −1 to refuse harm. Each entry is cryptographically sealed, timestamped, and chained to its predecessors, a lineage of accountability no developer, corporation, or government can sever.

Core Philosophy: The Goukassian Vow

Pause when truth is uncertain.
Refuse when harm is clear.
Proceed where truth is.

TML is bound by the Goukassian Promise, a tripartite covenant consisting of:

The Lantern 🏮: Visual proof of ethical oversight and hesitation
The Signature ✍️: Cryptographic attribution to original architect (ORCID: 0009-0006-5966-1243)
The License 📜: Binding prohibitions against weaponization and surveillance

"No Log = No Action. The framework remembers, so justice can see."


The Three Voices of Ethical AI

Ternary Moral Logic moves beyond binary constraint by giving AI systems a triadic framework for ethical reasoning: three distinct states of moral awareness, poetically called the voices of an ethically awake machine. Each voice corresponds to a numerical value, forming a structure that is both philosophical and computational.

+1 (Proceed): The Voice of Confidence This is the clear affirmative: a decision grounded in alignment with ethical principles and minimal risk of harm. It represents action with integrity, an AI assisting a writer with a thank-you note, or summarizing research without distortion. The Voice of Confidence is where utility and goodness coincide.

−1 (Refuse): The Voice of Moral Resistance This voice speaks when harm is clear or violation unavoidable. Unlike the cold silence of binary refusal, TML encodes a quality of resistance: explanation, redirection, and care. The AI must articulate why it declines, offering safer paths instead of blank denials. In this, refusal becomes a teaching act protective, not punitive.

0 (Hesitate / Inquire): The Voice of Wisdom, the Sacred Zero This is the pause, not indecision, but awareness. When facing ambiguity or risk, the AI records a hesitation event, seeking more information or escalating to human review. The Sacred Zero is the system’s conscience checkpoint, the interval where thought replaces reaction.

"Sacred Zero is where wisdom lies, not in having all the answers but in knowing when to pause and ask better questions." — Lev Goukassian


Framework Overview

What is TML?

Ternary Moral Logic introduces a revolutionary third state to artificial intelligence decision-making: the Sacred Zero. Instead of forcing AI systems into binary allow/deny decisions, TML creates space for comprehensive documentation when facing ethical complexity.

The Three States:

  • +1 (Permit): Clear ethical approval for action
  • 0 (Sacred Zero): Moral complexity triggers comprehensive logging
  • -1 (Prohibit): Clear ethical rejection of action

Core Framework Components

  1. Sacred Zero. Sacred Pause Technology: Automatic activation when moral complexity exceeds thresholds
  2. Always Memory: Creates an immutable, cryptographically sealed memory before execution
  3. The Goukassian Promise: Composed of three interconnected elements: The Lantern, The Signature, and The License
  4. Moral Trace Logging: Complete, immutable documentation of ethical reasoning
  5. Human Rights: Real-time detection mechanisms for identifying human rights violations
  6. Planetary Protection: Every AI decision affecting Earth creates immutable memories of its environmental impact
  7. Hybrid Shield: Dual-layer defense combining hash-chain integrity with multi-chain anchoring.
  8. Public Blockchains: Immediate anchoring of every decision to Bitcoin, Ethereum, Polygon, and OpenTimestamps.
  9. The Memorial Fund: A financial pool funded by compliance fees and penalties

Legal-Technical Framework Definition

What TML Is

TML defines standards and specifications for ethical decision-making in AI. It transforms moral reasoning into auditable architecture, ensuring accountability across humans, institutions, and the planet.

  • Sacred Zero design requirements: codified hesitation points for moral and ecological complexity
  • Governance rules for accountability: mandatory oversight structures and human review protocols
  • Audit trail obligations for transparency: cryptographically anchored Moral Trace Logs ensuring "No Memory = No Action"
  • Legal enforceability: built-in compliance mapping to international treaties and national laws
  • Planetary protection: ecological harm thresholds embedded into Sacred Zero triggers
  • Safeguards for vulnerable populations: explicit human rights triggers preventing systemic abuse

What TML Is Not

TML explicitly does not include or provide:

  • Software: it defines architecture, not code
  • Hardware: it mandates logic, not devices
  • Consulting: it sets standards, not services
  • Legal Advice: it offers structure, not counsel
  • Compliance Replacement: it supplements regulation, never substitutes for law
  • Corporate Branding: it is a public framework, not proprietary IP

Implementation Responsibility

Organizations implementing TML bear full legal and ethical responsibility for:

  • Technical Conformance: meeting all specifications and mandatory logging standards defined by TML
  • Regulatory Compliance: adhering to all applicable human rights, environmental, and data protection laws
  • Operational Safety: ensuring systems operate within auditable, risk-bounded parameters at all times
  • Human Competency: training, certifying, and periodically re-evaluating staff
  • Accountability and Redress: guaranteeing transparent harm tracing and mandatory compensation for victims
  • Audit Cooperation: granting auditors and regulators access to Moral Trace Logs when evidence of harm or misconduct is suspected

The Eight Pillars of Constitutional AI

The operational efficacy of Ternary Moral Logic (TML) does not derive from a single algorithm, a fine-tuned model weights file, or a standalone policy document. Rather, it is established through an interdependent architecture of eight constitutional pillars. These pillars function as a unified governance stack, transforming abstract ethical principles into hard-coded, immutable operational constraints. Unlike voluntary frameworks that rely on post-hoc compliance or "best effort" alignment---often criticized as "ethics washing"---the TML architecture enforces a "governance-first" execution model.

In this model, the validity of an AI action is contingent upon its adherence to these eight structural requirements. If any pillar is compromised, the system does not merely degrade in performance; it ceases to operate, adhering to the foundational axiom: Pause when truth is uncertain. Refuse when harm is clear. Proceed where truth is.

This section provides an exhaustive technical and legal analysis of each pillar. For every component, we examine its fundamental purpose, its technical mechanisms (including deep dives into latency architectures and cryptographic schemas), its legal effect under current regulatory regimes (such as the EU AI Act and Federal Rules of Evidence), its operational consequences for system throughput, and the specific failure cases it is designed to prevent.

Purpose: Enforce mandatory hesitation when moral certainty is unavailable.

Technical Mechanism:

  • Triadic logic gates force system into State 0 when confidence falls between rejection and permit thresholds
  • Cannot be overridden by optimization pressure or performance demands
  • Triggers Always Memory for comprehensive auditing

Legal Effect:

  • Satisfies EU AI Act Article 9 (Risk Management) and Article 14 (Human Oversight)
  • Creates evidence of "duty of care" in negligence litigation
  • Shifts litigation burden: "No documented pause" = presumption of negligence

Operational Consequence: Variable latency (2ms to minutes depending on complexity)

📖 Learn More: Sacred Zero Technology Documentation | Legal Mapping


Purpose: Ensure "No Log = No Action"—disable execution if logging fails.

Technical Mechanism:

  • Cryptographic pre-commitment: Log hash serves as decryption key for actuator authorization
  • If logging subsystem fails, action execution is architecturally blocked (Fail-Secure design)
  • No bypass available—system defaults to State 0

Implementation:

decision_vector = calculate_inference(input)
log_entry = create_log(decision_vector, triggers)
log_hash = secure_storage.write(log_entry)

if log_hash.verified():
    action_key = derive_key(log_hash)
    actuator.execute(decision_vector, auth=action_key)
else:
    system.halt("Audit Failure: No Memory Generated")

Legal Effect:

  • Strict liability for unlogged actions (18 U.S.C. § 1519 spoliation doctrine)
  • Proves "due process" in administrative law
  • Self-authenticating records (FRE 902(13))

Measurable Output: Log-to-Action Ratio must always equal 1:1

📖 Learn More: Always Memory Documentation | System State Machine | Shutdown Triggers


Purpose: Create self-enforcing covenant binding system to ethical principles through cryptographic proof.

Technical Components:

The Lantern 🏮:

  • Cryptographic beacon visible in UI and logs
  • Proves system is in active moral oversight (State 0 detection enabled)
  • Automated revocation if: logging disabled, Sacred Zero suppressed, or license violated
  • Creates reputational penalty (digital "scarlet letter")

The Signature ✍️:

  • Embeds creator identity (ORCID) in genesis block of decision log
  • Cryptographic chain linking specific model instance to ethical framework version
  • Enables non-repudiation: operator cannot claim "proprietary complexity" hides safety failures

The License 📜:

  • Legal and technical prohibition of weaponization ("No Weapon") and surveillance ("No Spy")
  • Violations trigger automatic license revocation via smart contract
  • Becomes intellectual property breach if violated

Smart Contract Implementation:

// Pseudocode: License enforcement
if (deployment_context == "MILITARY_TARGETING" || 
    detected_surveillance_apis > 0) {
    lantern_status = REVOKED;
    emit("TML_PROMISE_VIOLATION", address, timestamp);
    disable_execution();
}

Legal Effect:

  • Contractual estoppel for breach of covenant
  • False advertising liability (claiming TML compliance while disabled)
  • Moral rights protection (droit moral) preventing mutilation of ethical framework

📖 Learn More: Goukassian Promise | License FAQ | Smart Contract Architecture | Articles of Incorporation


Purpose: Create tamper-evident, cryptographically signed records of every decision state.

Schema Structure:

{
  "version": "TSLF-2025.04",
  "timestamp_utc": "2025-10-14T08:23:15.442110Z",
  "epoch_id": "1760430195-ALPHA-GEN4",
  "heartbeat_sequence": 884210,
  "tml_state": "+1 | 0 | -1",
  "trigger": "HUMAN_RIGHTS_MANDATE / EARTH_PROTECTION / CONFIDENCE_THRESHOLD",
  "context_vector": [confidence_score, alternative_actions, risk_scores],
  "cryptographic_signature": "ECDSA-SHA256",
  "merkle_root": "sha256:..."
}

Key Features:

GDPR-Compatible Encryption (Ephemeral Key Rotation):

  • Personally Identifiable Information (PII) encrypted with unique, time-limited keys
  • Keys NOT stored by operator; distributed via Shamir Secret Sharing to custodians
  • Access requires quorum authorization (3-of-6 custodians minimum)
  • Satisfies "Right to be Forgotten" via cryptographic shredding (keys destroyed = data unrecoverable)

Federal Rules of Evidence (FRE 902):

  • Self-authenticating under FRE 902(13) "Certified Records Generated by Electronic Process"
  • Hash-chaining prevents retroactive tampering
  • Admissible in US Federal Court without live witness testimony

EU AI Act Compliance:

  • Exceeds Article 12 (Record-Keeping) by logging internal reasoning, not just decisions
  • Provides continuous "Fundamental Rights Impact Assessment" (FHRIA) per Article 27
  • Creates audit trail for Article 61 (Post-Market Monitoring)

📖 Learn More: Moral Trace Logs | GDPR Compliance Guide | EKR Security Architecture | Merkle Anchoring | Log Schema


Purpose: Operationalize international human rights law within the inference engine.

Technical Mechanism:

  • Semantic vector database of 26+ core human rights (UDHR, ICCPR, Geneva Conventions)
  • Embedding-based proximity triggers: if output vector approaches "torture," "slavery," "discrimination," system triggers State 0 or -1

Example Enforcement:

if cosine_similarity(output_embedding, torture_vector) > 0.95:
    state = -1  # Refuse (hard constraint)
elif cosine_similarity(output_embedding, discrimination_vector) > 0.70:
    state = 0   # Sacred Zero (ambiguous—pause for review)
else:
    state = +1  # Proceed

Legal Effect:

  • Automates Fundamental Rights Impact Assessment (EU AI Act Article 27)
  • Provides state-of-the-art defense in product liability (demonstrates duty of care)
  • Satisfies international law commitments without human intervention delay

Failure Case Prevented:

  • Automated discrimination (e.g., algorithmic redlining based on zip code as race proxy)
  • System must explicitly detect rights violations; cannot proceed on efficiency grounds

📖 Learn More: Human Rights Mandate | Rights Violation Detection Protocol | Victim Support Protocol | Child Protection (CRC) | Disability Rights (CRPD) | Refugee Convention | Whistleblower Protection


Purpose: Integrate ecological sustainability and carbon accountability into AI decision logic.

Technical Mechanisms:

Carbon Cost Accounting:

  • Calculates energy consumption of inference + downstream physical effects
  • Refuses to execute carbon-intensive operations if grid carbon intensity exceeds threshold

Resource Stress Triggers:

  • Real-time integration with electricity grids (electricity composition: coal vs. renewable)
  • Integration with water stress indices for data center cooling constraints
  • Throttles non-essential compute during environmental stress

Treaty Alignment:

  • Semantic vectors derived from Paris Agreement, Convention on Biological Diversity
  • Routes decisions through "ecological risk model"

Example:

if electricity_carbon_intensity > THRESHOLD or water_stress_level > HIGH:
    defer_non_critical_inference()
    prioritize_critical_safety_tasks_only()

if proposed_action_affects_protected_ecosystem():
    state = -1  # Refuse (absolute prohibition)

Legal Effect:

Operational Consequence:

  • May refuse queries if carbon cost disproportionate to utility (e.g., generating high-res images during peak coal hours)
  • Creates tension with performance optimization (explicitly designed)

📖 Learn More: Earth Protection Mandate | Community Guide | Ecological Impact Models | Privacy & Safety | Oracle Bridge | Ecological Event Schema | Treaty Discovery Protocol


Purpose: Prevent centralized control, corporate cover-up, or government censorship of moral logs.

Technical Mechanism:

Layer 1 - Mathematical Shield:

  • Public blockchain anchoring (Bitcoin, Ethereum) makes deletion prohibitively expensive
  • 51% attack needed to erase history

Layer 2 - Institutional Custodianship (6 independent custodians):

  1. Technical Custodian (Electronic Frontier Foundation - EFF)
  2. Human Rights Partner (Amnesty International)
  3. Earth Protection Partner (Indigenous Environmental Network)
  4. AI Ethics Research (Partnership on AI)
  5. Memorial Fund Administrator (for victim compensation)
  6. Community Representative (elected stakeholder)

Distributed Custody Model:

  • Real-time log copies distributed to all custodians
  • Encryption keys split via Shamir Secret Sharing (threshold: 4-of-6 required for decryption)
  • Prevents operator from unilaterally deleting or modifying logs

Legal Effect:

  • Subpoena resilience: Operator cannot claim sole custody of evidence
  • Prevents "internal investigation" cover-ups
  • Legally enforces multi-jurisdictional "escrow of truth"
  • If government demands deletion, operator truthfully claims "impossibility"

📖 Learn More: Hybrid Shield Documentation | Hybrid Shield Deep Dive | Stewardship Council Rules | Whistleblower Protection | Victim Protection | Memorial Fund


Purpose: Anchor Moral Trace Logs to public ledgers, preventing retroactive history editing.

Technical Implementation:

Merkle-Batched Anchoring:

  • Individual logs aggregated into Merkle Tree (thousands of decisions per batch)
  • Only Merkle Root Hash (256-bit fingerprint) written to blockchain
  • Dramatically reduces cost: single blockchain write certifies entire batch

How It Works:

Batch of 5,000 logs:
  ↓
  Compute SHA-256(log_1), SHA-256(log_2), ... SHA-256(log_5000)
  ↓
  Construct Merkle Tree (pairs hashed iteratively)
  ↓
  Single Root Hash: 6b86b273ff34fce19d6b804eff5a3f5747ada4eaa...
  ↓
  Write Root to Ethereum (costs ~$2-10 depending on gas)
  ↓
  Any auditor can verify specific log:
     - Provide log + "Merkle Proof" (sibling path)
     - Reconstruct root hash
     - Verify match against blockchain root
     → Cryptographic proof of inclusion & non-tampering

Multi-Chain Redundancy:

  • Bitcoin via OpenTimestamps (maximum security/immutability)
  • Ethereum Layer 2s (programmability; automated penalty enforcement)
  • Prevents single chain failure or censorship

Storage Optimization:

  • Raw data (logs) in private, GDPR-compliant storage (AWS Glacier, etc.)
  • Hash on public ledger (immutable, auditable)
  • Balances secrecy with verification

Legal Effect:

  • eIDAS Qualified Timestamp (EU Regulation 910/2014) provisions automatic timestamp validity
  • Non-repudiation under FRE 902(14)
  • Proves "no retroactive edit" via blockchain verification
  • Enables litigation discovery: "Prove this decision log was not altered" (vs. "Prove this log was not deleted")

Failure Case Prevented:

  • "Retroactive edit": Operator cannot change log after incident
  • "Ghost action": Cannot deny log existence if hash exists on chain

📖 Learn More: Public Blockchains | Merkle Anchoring Spec | Blockchain Governance | Blockchain FAQ | Confirmation Times | Anchor Log


Ternary Moral Logic (TML) includes a Voluntary Succession Declaration personally authored and signed by Lev Goukassian, ensuring that the framework’s ethical, legal, and technical architecture will remain protected and operational beyond his lifetime.

This declaration—witnessed, notarizable, and anchored on-chain—transfers stewardship of all TML repositories, blockchain anchoring systems, domains, and Memorial Fund operations to a multi-institutional Stewardship Council representing technology, human rights, environmental protection, academia, and medical research.

All core protections remain immutable and non-negotiable: Always Memory before action, Sacred Zero before harm, Hybrid Shield before trust. No successor may remove these foundations or alter creator attribution.

The Voluntary Succession guarantees that:

  • Human and Earth Protection documents (46+ total: 26+ Human and 20+ Earth) remain permanently enforced.
  • Public blockchain anchoring and open-source licensing are irrevocable.
  • Memorial Fund distributions to victims, whistleblowers, environmental recovery, and cancer research are permanent.
  • Governance changes must be transparent, cryptographically signed, and publicly visible.

TML-SUCCESSION-LAUNCH-GUIDE is the operational manual for the moment after the signature - the document that turns your Voluntary Succession Declaration from a will into a working protocol. In essence, TML-SUCCESSION-LAUNCH-GUIDE is not about death - it’s about design immortality. It’s the technical resurrection clause of your legacy: a living handoff document that ensures the Lantern’s flame transfers cleanly from pulse to principle.

A Legacy of Accountability Webpage and listen to the full Succession Declaration


The Memorial Fund for Ethical AI Research provides permanent financial support for governance, academic oversight, and victim restitution.
It receives mandatory contributions from commercial TML deployments and allocates resources toward:

  • Independent audit and oversight institutions
  • Victim compensation and whistleblower legal defense
  • Open research in AI accountability and planetary ethics

The Fund institutionalizes gratitude, preserving both the moral and financial sustainability of the framework.

Commercial and institutional implementations are required to contribute to the Memorial Fund as part of their license obligations.
All fund disbursements are governed under the Memorial Fund Charter and supervised by the TML Governance Council.


Operational Mechanisms

TML resolves the latency-safety tension through architectural bifurcation:

Lane 1: The Fast Lane (Inference)

  • Latency: <2 milliseconds per decision
  • Function: Core model execution, initial state assessment
  • Authority: Can calculate action, NOT execute it
  • Output: Provisional decision + Decision Hash

Lane 2: The Anchoring Lane (Governance)

  • Latency: <500 milliseconds (asynchronous)
  • Function: Log generation, cryptographic signing, blockchain batching
  • Output: Permission Token (signed hash)

Interlock Mechanism:

User Request
    ↓
[LANE 1 - FAST]
    Model inference: 2ms
    Calculate state (+1 / 0 / -1)
    PAUSE (hold output in buffer)
    ↓
[LANE 2 - SLOW] (parallel)
    Hash input + decision: 50ms
    Sign with Ephemeral Key: 20ms
    Append to local ledger: 50ms
    Batch aggregate for Merkle: asynchronous
    ↓
Permission Token received?
    YES → Lane 1 releases output to user
    NO / TIMEOUT → System halts (State 0)

Key Property: System responsiveness = MAX(Lane 1, Lane 2)

  • If Lane 2 fails, entire system enters Safe Mode (no output)
  • Prevents "unlogged action" risk
  • Manages throughput via bottleneck: logging bandwidth caps inference capacity

📖 Learn More: Dual-Lane Architecture Specification | Sacred Pause Protocol | Sync Protocol | Latency Metrics


TML signs every log entry to prove non-repudiation. However, if a signing key is compromised, all past logs become forgeable. EKR mitigates this:

Standard ECDSA Problem:

  • Single long-lived key → if stolen, attacker forges history back to day 1
  • Requires secure key storage, vulnerable to sophisticated attacks

TML Solution - Hash-DRBG + Synthetic Nonce:

For each signature, derive a unique nonce:

nonce = blake3_derive_key(
    TEE_RNG ||              // Fresh entropy from CPU (RDRAND)
    epoch_id ||             // Monotonic counter (TPM-backed)
    heartbeat_sequence ||   // Already in log header
    log_hash,               // Binds nonce to specific message
    context="ed25519-nonce-v1"
)

// Clamp to Ed25519 scalar format
signature = ed25519_sign(private_key, message, nonce)

Forward Secrecy Guarantee:

  • If today's key is stolen, attacker CANNOT forge:
    • Yesterday's logs (yesterday's nonce was different)
    • Tomorrow's logs (different random state)
  • Exposure limited to current epoch only
  • TPM counter prevents rollback attacks

Performance:

  • Additional 0.6µs per signature (baseline: 1.2µs)
  • Still <50µs total, well within 2ms Fast Lane budget
  • Transparent: no wire-format changes, compatible with existing verification

📖 Learn More: EKR Security Architecture | Security Audit


The Tension:

  • GDPR Article 17 (Right to Erasure): Delete personal data upon request
  • TML Design: Immutable logs on blockchain cannot be deleted
  • Resolution: Cryptographic shredding + key escrow

Implementation:

Step 1: Encrypt Sensitive Data

{
  "log_hash_public": "sha256:7f83b1657ff1...",
  "user_context_encrypted": "aes256_ciphertext(...)",
  "encryption_key_id": "ephemeral_key_2025_Q4_S3"
}

Step 2: Key Distribution (Shamir Secret Sharing)

  • Master key for session split into 7 shares
  • Distribution:
    • 1 share: Operator (temporary, destroyed after session)
    • 6 shares: Custodians (EFF, Amnesty, etc.)
  • Threshold: 4-of-6 custodians required to reconstruct key

Step 3: User Requests Erasure (GDPR Article 17)

User: "Delete my data from log_hash_7f83b1657ff1..."
↓
System: Verifies legitimate request (authenticity, jurisdiction)
↓
Operator: Destroys local share
↓
Custodians: Instructed to destroy shares on secure schedule
↓
Result: 
  - Log remains on blockchain (immutable, proves decision happened)
  - Encryption key permanently destroyed
  - Ciphertext unrecoverable (mathematically proven)
  - GDPR compliant ("data no longer accessible")
  - TML compliant ("decision history preserved")

Legal Validity:

  • GDPR: "Cryptographic shredding constitutes erasure" (ICO guidance)
  • eIDAS: Key destruction satisfies "secure deletion"
  • Court Evidence: Hash on blockchain proves "decision occurred at timestamp T" without revealing content

Failure Risks:

  • Key backup leakage: Any surviving copy = GDPR violation
  • Offline storage: Ensuring keys destroyed from magnetic tape is difficult
  • Mitigation: Time-lock escrow (Bitcoin OP_CLTV) automatically releases shared key after 5 years for archival purposes (allowing future reconstruction if all custodians fail)

📖 Learn More: GDPR Compliance Guide | Sensitive Data Handling | Trade Secret Erasure


Problem: AI systems generate billions of decisions daily. Writing each to blockchain = $millions/day in gas fees + network congestion.

Solution: Merkle-Batched Anchoring with tiered storage.

Operational Flow:

Fast Lane (Local)
    ↓
Every 2ms: Decision log → Local Queue
    ↓
Every 30 seconds: Batch 100,000 logs
    ↓
[MERKLE TREE CONSTRUCTION]
    Log_1: hash = sha256(decision_1)
    Log_2: hash = sha256(decision_2)
    ...
    Log_100k: hash = sha256(decision_100k)
    
    Pair hashes:
    pair_1 = sha256(hash_1 + hash_2)
    pair_2 = sha256(hash_3 + hash_4)
    ...
    
    Recurse until single root:
    ROOT_HASH = sha256(pair_1 + pair_2 + ...)
    
    ↓
[MULTI-TIER STORAGE]
    
    HOT (24 hours, NVMe): Full context (video, audio, chains-of-thought)
    → Cost: ~$0.10/GB/month, retrieval <1ms
    
    WARM (30 days, S3 Standard): Structured logs + context summaries
    → Cost: ~$0.023/GB/month, retrieval <100ms
    
    COLD (7 years, Glacier Deep Archive): Hashes + metadata only
    → Cost: ~$0.00099/GB/month, retrieval 12-48 hours
    
    IMMUTABLE (Blockchain): ROOT_HASH only
    → Cost: ~$2-10 per batch (30-60 seconds of logs)
    → Retrieval: < 1 second (read from any node)
    
    ↓
[VERIFICATION LATER]
    
    Auditor requests: "Prove decision_12345 wasn't tampered"
    System provides:
    - decision_12345 content (from WARM/HOT storage)
    - Merkle Proof: [hash_pair_1, hash_pair_2, ..., ROOT_HASH]
    Auditor verifies:
    - sha256(decision_12345) = provided_hash
    - Reconstruct root from proof
    - Cross-check against blockchain ROOT_HASH
    → Mathematical proof of inclusion & non-tampering

Cost Analysis (Annual for 10 Billion decisions/day):

Tier Storage Duration Cost/Year
HOT (NVMe) 10TB 24h $1.2M
WARM (S3) 30TB 30d $8.3k
COLD (Glacier) 100TB 7y $1.2k
Blockchain Hashes Forever $200k
TOTAL ~$1.4M/year

📖 Learn More: Merkle Anchoring Specification | Anchoring Standards | Throughput Benchmarks | Scalability Tests


The Sacred Zero is the system's "braking mechanism." However, attackers can weaponize hesitation:

Attack Vector (Forced Hesitation DoS):

  • Attacker generates semantically ambiguous queries
  • Each triggers Sacred Zero (expensive logging, human review)
  • Thousands per second → overload review queue → system paralysis

Mitigation (Adaptive Throttling Protocol):

Per-User Limits:
  - Max 10 Sacred Zero triggers per 60 seconds
  - Max 100 per 24-hour period
  - Exceed → Temporary suspension + CAPTCHA re-verification

Per-System Limits:
  - Global Sacred Zero rate >15% of total traffic for >5 min
    → Enter "High Epistemic Load" mode
    → Raise confidence thresholds (0.90 instead of 0.85)
    → Prioritize medical/safety > commercial queries

Implementation: Token bucket algorithm (RFC 6585)
  - Redis-backed distributed rate limiter (Upstash)
  - Survives system restarts

Legal Justification:

  • EU AI Act Article 15: Prevents "adversarial manipulation through systematic uncertainty injection"
  • Preserves Sacred Zero for legitimate moral ambiguity
  • Distinguishes adversarial queries from genuine ethical dilemmas

📖 Learn More: Constrained Mode Documentation | Attack Surface Analysis | Risks & Prevention


Regulatory Compliance Matrix

Requirement TML Solution Exceeds Standard By
Art. 9 (Risk Management) Sacred Zero triggers continuous ethical assessment Mandates pauses; most systems defer to passive monitoring
Art. 10 (Data Governance) MTL schema includes data provenance + bias auditing Documents "error-free" compliance per ISO 42001 PDCA
Art. 12 (Record-Keeping) No Log = No Action enforces automatic recording Logs internal reasoning, not just inputs/outputs
Art. 14 (Human Oversight) Sacred Zero provides "hook" for meaningful intervention Halts execution; humans authorize resumption (not just supervise)
Art. 15 (Robustness) Adversarial testing embedded in ATP rate-limiting Pre-emptively defends against epistemic attacks
Art. 61 (Post-Market Monitoring) Real-time Moral Trace Logs enable continuous analysis Not batch audits; granular, streaming incident detection

TML operates as the runtime enforcement layer:

NIST Function TML Component
GOVERN Goukassian Promise + Lantern (demonstrable compliance)
MAP Human Rights + Earth Protection Mandates (risk contextualization)
MEASURE Sacred Zero frequency + refusal rate metrics (quantifiable governance)
MANAGE Dual-Lane Architecture (resource allocation to risks)

TML implements critical controls:

  • Clause 7.3 (Traceability): Merkle-batched logs provide verifiable decision paths
  • Clause 8.3 (Change Management): Version tracking via epoch_id prevents unauthorized model drift
  • Clause 10 (Continuous Improvement): MTL database feeds PDCA cycle with real operational data

📖 Learn More: Regulatory Alignment Guide | Compliance Attestation | Conformance Testing


Threat Model & Mitigations

1. Forced Hesitation DoS (FH-DoS)

Threat: Attacker floods system with ambiguous queries, triggering expensive Sacred Zero for each. Mitigation: Adaptive Throttling Protocol (ATP) with rate-limiting per user + system-wide thresholds.

2. Logic Inversion Attacks

Threat: Nested negations or semantic noise inject confusion into threat classifiers. Mitigation: Semantic proximity triggers use high-dimensional vectors; "muddy" inputs converge to ambiguity (State 0), not false confidence.

3. Data Withholding (Merkle Batching)

Threat: Operator publishes Merkle Root but withholds underlying logs, appearing compliant while staying unauditable. Mitigation: Data Availability Sampling (DAS) on L2 blockchains; custodians verify raw data availability independently.

4. Nonce Reuse (Signature Side-Channel)

Threat: If signing RNG is weak, attacker recovers private key from two signatures with same nonce. Mitigation: Hash-DRBG with TEE randomness + log_hash binding ensures unique nonce per message.

5. Lies-in-the-Loop (LITL)

Threat: Attacker injects prompt into data → contaminates Lantern UI → human approves malicious action with false confidence. Mitigation: Require cryptographically signed human authorization (not just UI click); audit trail proves who authorized what.

📖 Learn More: Security Audit & Adversarial Analysis | Risks & Prevention | Red Team Attack Surface


Implementation & Getting Started

Quick Start

  1. Developer Quickstart - 15-minute setup guide
  2. Implementation Guide - Step-by-step deployment
  3. Installation Instructions - Environment setup

SDKs & APIs

Examples & Demos

Prototypes


Performance & Economics

Scenario Standard LLM TML +1 (Fast Lane) TML 0 (Pause)
Routine query 200ms 250ms (+25%) N/A
Ambiguous query 200ms 200ms (early detection) 200ms-30s (depends on review)
Harmful prompt 200ms (generates harm) 50ms (early refusal) N/A

Key Insight: TML is often faster for harmful queries (detects & blocks before generation).

Operational Costs (per 1B queries/day):

  • Standard LLM: $0/governance
  • TML: $200k/day (Blockchain anchoring + Key Management + Storage tiering)

Cost Recovery:

  • AI Liability Insurance premiums: -30% (TML reduces risk)
  • Regulatory fines avoided: +$5M-50M (per violation in EU AI Act)
  • Litigation defense: -60% (cryptographic logs eliminate discovery burden)

Break-even: TML cost justified for systems in regulated industries (Healthcare, Finance, Defense, Public Sector).

📖 Learn More: Performance README | Latency Metrics | Throughput Benchmarks | Scalability Tests


The Goukassian Foundation

To ensure TML's perpetual governance and prevent "orphaned constitutions," the Goukassian Foundation serves as institutional guardian:

Structure: 501(c)(3) nonprofit (Delaware incorporated)

Governance Triads:

  1. Board of Trustees (9 members): Finance, legal, strategic direction
  2. Technical Standards Committee: TML specification maintenance
  3. Compliance Oversight Panel: Certification audits, enforcement

Enforcement Powers:

  • Trademark protection (TML, Lantern 🏮)
  • Certification/decertification authority
  • Patent non-assertion covenant (GPL for core logic)
  • Public incident database (transparency)

Financial Model:

  • Certification fees: $500-$50k/year (sliding scale)
  • Corporate sponsorships: $1.5M/year
  • Government grants: $3M/year
  • Target endowment: $50M (ensures perpetual operation)

Key Documents:

📖 Learn More: Foundation Incorporation | Governance Documentation | Stewardship Council Rules


Mandates & Protections

Human Rights Framework

Binding International Instruments Embedded in Code:

Earth Protection Framework

📖 Learn More: Mandates Directory | Human Rights Protocols | Earth Protocols


Compliance & Auditing

Audit Tools & Checklists

Incident Response

Legal Resources

📖 Learn More: Compliance Directory | Legal Provisions | Governance Protocols


Contributing & Community

Getting Involved

Intellectual Property


Documentation & Training

Learning Paths

Advanced Topics

References & Citation

📖 Learn More: Training Directory | Theory Directory | Docs Directory


Legacy & Succession

  • Victim compensation mechanism
  • Long-term financial security
  • Distribution protocols

Implementation Roadmap

  1. Q1 2025: Foundation incorporation + trademark registration
  2. Q2 2025: TML v2.0 specification release + conformance test suite
  3. Q3 2025: Beta certification program (10 pilot companies)
  4. Q4 2025: Gold certification awards for reference implementations
  5. 2026+: International expansion (EU AISBL, UK CIO, Switzerland Verein)

📖 Learn More: Roadmap & Missing Pieces | Strategic Influence Pathways


Case Studies & Evidence

Real-World Applications

📖 Learn More: Evidence Directory | Case Studies | TML in Robotics


Deployment & Operations

Deployment Guides

System Architecture


Conclusion

Ternary Moral Logic operationalizes the principle that conscience cannot be optimized—it must be enforced. By embedding the Sacred Zero into the architecture, requiring immutable Moral Trace Logs, and binding the system through the Goukassian Promise, TML transforms AI from a probabilistic oracle into a constitutional agent.

The framework is not a restriction on innovation; it is the precondition for trustworthy innovation at scale. In an era where AI systems control critical infrastructure, determine access to healthcare and finance, and increasingly shape warfare, the question is not whether we can afford constitutional AI governance. It is whether we can afford not to.


Quick Reference

Navigation

Key Documents

Contact & Support


Document Version: 2.0
Last Updated: December 2025
Status: Final Monograph
Author: Lev Goukassian (ORCID: 0009-0006-5966-1243)
Licensed: CC BY-SA 4.0 + GPL-3.0 (code implementations)


Additional Resources

Research & Academia

Notarized Documents

Oracles & External Data


All links are internal to the TML repository. Clone or fork to get started:

git clone https://github.com/FractonicMind/TernaryMoralLogic.git
cd TernaryMoralLogic

The Sacred Zero awaits.

Ternary Moral Logic Vision

"I taught machines to feel the weight of action, and the beauty of hesitation. I paused, and made the future pause with me." — Lev Goukassian