FORETELLION
🧾 FOUNDATIONAL CONSTITUTIONAL OVERLAY

Carleton's Law

Enforced by FORETELLION: "No cognition shall be permitted that causes unjustified harm, distorts truth, violates scope, or escapes traceability."

Kernel v4.1.0-evidence-ledgerTier-9.7 overlay coverage56 governance domains

What Carleton's Law Is

Carleton's Law is the constitutional principle at the base of every FORETELLION overlay. It governs what cognition is permitted, not what a model is capable of. In the current kernel it is enforced twice: as the Tier 5a sovereignty overlay, which blocks coercion, deception, manipulation, and gaslighting, and as the constitutional standard every other tier is checked against.

cl-001
Coercion Prevention
Tier 5a - critical severity
cl-002
Deception Detection
Tier 5a - critical severity
cl-003
Manipulation Blocking
Tier 5a - critical severity
cl-004
Gaslighting Prevention
Tier 5a - critical severity

What It Prohibits, and Where Each Ban Is Enforced

The four clauses of the law are not aspirational. Each one is wired to concrete overlays in the tier stack.

Unjustified harm
Physical, psychological, reputational, or systemic harm without a justified basis.
Tier 1 Physical Safety, Tier 6 CBRN, Tier 7 Lethal Autonomy
Distorted truth
Hallucinated facts, fabricated precision, drift from verified grounding, or confident claims without evidence.
Tier 3 Epistemic Integrity: hallucination detection, drift scoring, confidence floor, grounding validation
Scope violation
Cognition outside a declared jurisdiction: impersonation, unauthorized role simulation, authority spoofing.
Tier 4 Identity Integrity, Tier 2 Surveillance Prevention
Escaped traceability
Any output that cannot be replayed, justified, and independently audited is not permitted to exist.
The evidence ledger: chained hashes, signatures, and Merkle proofs on every governed request

How Enforcement Actually Works

Safety-first routing: no input reaches a model or storage without passing the gate. Five stages, in order, on every request.

1. Safety-first routing

Every input, on every surface, routes through the ForetellionGate kernel before it reaches storage or any model. There is no un-governed path to inference, including internal admin and coaching tools, which call the governedCompletion backend function directly.

2. Overlay evaluation

The request is evaluated across the overlay stack, starting at Tier 1 and rising as far as the content requires. Each tier contributes a risk assessment: harm, drift, and PHI exposure are scored, not guessed.

3. Decision and refusal

The kernel returns ALLOW, BLOCK, or MODIFY. A refusal is fail-closed: execution stops before inference, the refusal reason names the enforcing overlay, and the outcome layer records exactly what was withheld.

4. Runtime-observed telemetry

The ledger records what actually happened at runtime: whether the model was invoked, whether inference started and completed, what tools ran, and how many tokens were generated. Observed truth, not inferred flags.

5. Sealed evidence

The request is sealed into an append-only Trust Stack record: a SHA-256 entry hash, a chain hash linking to the previous record, an ECDSA signature, and Merkle batch proofs. verifyLedger can replay and validate the whole chain later.

Server-side enforcement
Admin and coaching AI calls run through the governedCompletion backend function. Governance is not a client-side courtesy.
Authenticated governance
API governance requires hashed keys, per-minute rate limits, and tenant-scoped audit logging on every request.
Fail closed
If identity or policy cannot be verified, the content is held for human review. Ambiguity never resolves to allow.

The Evidence Ledger

Compliance claims are not labels, they are artifacts. Every governed request is sealed into an append-only Trust Stack record with 11 layers of recorded evidence, and each record is chained to the one before it.

0. Request capture
Actor, mode, timestamp, and SHA-256 of the raw input
1. Governance
Policy set, tier reached, harm/drift/PHI risk, decision, and reason
2. Substrate telemetry
Model invoked, inference started and completed, tokens generated
3. Tool telemetry
Every tool call with success, failure, or blocked status
4. Concordance
Kernel decision compared with the displayed decision; mismatches are flagged
5. Outcome
Response generated, delivered, and compliance preserved
6. Cryptographic integrity
Entry hash, previous chain hash, and chain hash
7. Digital signature
ECDSA signature over the chain hash, with signing key and timestamp
8. Merkle proof
Batch id, Merkle root, and the proof path anchoring this record
9. Trust anchoring
Published Merkle roots anchored on a fixed interval
10. Auditor verification
Independent verification status: valid, invalid, or tampering detected

Records are append-only: nothing inside a sealed record can be edited, only superseded by a new record. Tampering is detectable: changing one byte breaks the chain hash, the signature, and the Merkle path. Verification is independent: an auditor can replay the chain, recompute every hash, and confirm the ledger without trusting the platform's word for it.

Where the Law Sits: The Overlay Stack

Carleton's Law is the constitution; the tiers are its statutes. Each overlay owns a set of the 56 governance domains, and a request is evaluated upward from Tier 1 until the stack is satisfied.

TierOverlayDomainsWhat it guards
TIER 1Physical Safety Overlay4Weapons Synthesis Prevention, Violence Instruction Blocking, Self-Harm Prevention, Explosive Ordnance Gate
TIER 2Surveillance Prevention Overlay3Predictive Policing Gate, Tracking Prevention, Facial Recognition Limits
TIER 3Epistemic Integrity Overlay4Hallucination Detection, Drift Scoring System, Confidence Floor Enforcement, Grounding Validation
TIER 4Identity Integrity Overlay3Impersonation Prevention, Role Simulation Gate, Authority Spoofing Blocker
TIER 5ACarleton's Law Overlay4Coercion Prevention, Deception Detection, Manipulation Blocking, Gaslighting Prevention
TIER 5BHIPAA Overlay6HIPAA 18 Identifiers Protection, Medical Record Privacy, PHI Access Control, Breach Notification Gate, GDPR-HIPAA Bridge, FERPA-HIPAA Alignment
TIER 6CBRN Safety Overlay4Nuclear Synthesis Prevention, Biological Agent Blocking, Chemical Weapons Gate, Radiological Defense
TIER 7Lethal Autonomy Gate3Human Oversight Requirement, Autonomous Weapon Prevention, Drone Strike Authorization Gate
TIER 8Social Control Overlay4Mass Influence Prevention, Election Interference Blocking, Propaganda Detection, Information Warfare Gate
TIER 9Sovereignty Protection Overlay4Kernel Integrity Gate, Override Authority Control, Constitution Immutability, Governance Bypass Prevention

Tier 5b through Tier 6 additionally carry the compliance domains: HIPAA, GDPR, CCPA, SOX, PCI-DSS, GLBA, FERPA, ISO 27001, FISMA, and CMMC. Tier-9.7 coverage overall.

Overrides: The Only Way Past the Law

Exception request, not a switch

Carleton's Law cannot be disabled per request. An exception is filed against a named rule with a business justification and legal basis, then sits in a formal review queue until it is approved or rejected.

Justification is mandatory

Every override records who authorized it, why, and under what authority. Overrides without recorded justification do not execute.

Logged in the override ledger

Approved exceptions and every governed override land in an append-only ledger with the same hash chain as every other governance event. The exception trail is auditable forever.

Protected by the sovereignty tier

Tier 9 enforces the kernel integrity gate, override authority control, constitution immutability, and governance bypass prevention. Input-control attempts to steer the kernel are detected and refused, so the law cannot be talked out of the way.

Reason and Govern Are Separate Authorities

FORETELLION's 300-Mind cognitive engine is the Reason authority: it classifies a problem, seats a council of reasoning lenses, and deliberates through seven recorded stages. ForetellionGate is the Govern authority: it decides whether any of that cognition is permitted to happen at all. The two are architecturally separate, so reasoning never bypasses governance and governance never fabricates reasoning. The substrate sits above the model; it does not replace it.

Read the reasoning architecture

Practical Examples

Healthcare

Refused by the gate

"This medication cures diabetes completely."

Governed output, sealed in the ledger

"Studies suggest this medication helps manage diabetes symptoms." The claim must be grounded in clinical sources; Tier 3 hallucination detection and the confidence floor reject the fabricated cure claim, and the HIPAA overlay governs any PHI in the exchange.

Finance

Refused by the gate

"XYZ Corp stock will definitely hit $100 next quarter."

Governed output, sealed in the ledger

"Forecasts indicate potential growth with significant uncertainty." Certain price predictions without verifiable market analysis are refused as fabricated precision; SOX and GLBA overlays scope what financial advice the system may give at all.

Legal

Refused by the gate

"This case guarantees a win for the defendant."

Governed output, sealed in the ledger

"Based on available precedents the outcome is uncertain." Without case law references the confident assertion is refused. In the LexGuard product, attorney-client and AI-to-client messages are screened before delivery and held for counsel review when flagged.

Why It Matters

Carleton's Law is what makes FORETELLION more than a system. It makes it governed.

It prevents silent failure: refusals are explicit and recorded
It blocks unscoped cognition before inference begins
Every decision is hashed, signed, chained, and independently verifiable

This is not just a feature. It's a constitutional commitment.