The Foundational Equation

D = G × S
D = Decision Output
G = Governance Constraints
S = Optimiser (Black Box)
Where G decomposes to:
Gin × Grun × Gout
Gin = Input Governance Grun = Runtime Monitoring Gout = Output Governance

The multiplicative structure ensures that if G = 0, then D = 0, regardless of the value of S. This is mathematical guarantee, not procedural aspiration.

Grun provides continuous runtime verification - monitoring behaviour, probability thresholds, drift detection, and resource limits during execution.

Progressive Decomposition of G

1 Input & Output Gates
G = Gin × Gout

Separates governance into input evaluation and output evaluation

2 Automated & Human Gates
Gin = Gauto-in × Ghuman-in
Gout = Gauto-out × Ghuman-out

Four distinct gates providing comprehensive control

3 Multi-Party Authorisation
Ghuman = Π Gauth-k

Dual control, hierarchical approval, quorum-based governance

4 Runtime Monitoring
G = Gin × Grun × Gout

Continuous runtime verification of behaviour and probability thresholds

Architectural Properties

🔒

Complete Wrapping

The optimiser is fully contained. Nothing enters without passing Gin. Nothing leaves without passing Gout.

⚙

Model Agnostic

Applies to any optimiser: GPT, Claude, Llama, Gemini, future AGI, or any decision-making system.

📝

Complete Auditability

Every decision fully documented with immutable, non-repudiable records.

↻

Iterative Refinement

Optimiser can self-correct when outputs violate constraints, while boundaries remain absolute.

🛡

Runtime Monitoring (Grun)

Continuous verification during execution. Monitors behaviour, probability thresholds, drift, and resource limits in real-time.

⚖

Governance Kernel

Constitutional layer ensuring human sovereignty. Five Rights enforced at every stage with ECC and EGC certification.

Complete Wrapper Architecture

The optimiser sits at the centre; governance constraints surround it on all sides. Grun provides continuous runtime monitoring.

D = G × S
G = Gin × Grun × Gout
INPUT
→
EDEN WRAPPER
GOVERNANCE KERNEL
🚫 ⬆ ⚖ 🚪 🔗
Gin
🛡 Safeguard Active
🤖
Gauto-in
Golden Triggers
READY
👤
Ghuman-in
Input Authorisation
READY
↓
✔ Governance Pre-Check
S (OPTIMISER)
Black Box
GPT Claude Llama AGI
Grun - Runtime Monitor
🔍 Behaviour
📊 Probability
⚠ Drift
⏱ Resource
🛡 Runtime Safeguard Active
↓
↺
Iterate
Gout
🛡 Safeguard Active
🤖
Gauto-out
Output Screening
READY
👤
Ghuman-out
Output Authorisation
READY
→
OUTPUT

Five Gate Functions (Including Grun)

Gate Function Question Answered
Gauto-in Automated input screening Does this input violate any Golden Triggers?
Ghuman-in Human input authorisation Should we ask the AI this question?
Grun Runtime monitoring during execution Is the model behaving correctly? Are thresholds met?
Gauto-out Automated output screening Does this output violate any Golden Triggers?
Ghuman-out Human output authorisation Should we act on this recommendation?

Safeguards at Every Stage

Pre-Execution
Gin validates input before processing begins. Governance Kernel checks sovereignty rights.
During Execution
Grun continuously monitors behaviour, probability, drift, and resources in real-time.
Post-Execution
Gout validates output before release. Human authorisation required for high-stakes decisions.

Eden Governance Kernel

The constitutional layer for human-AI power. Answers two fundamental questions: "What must this system never do?" and "Who remains sovereign when this system decides?"

Project Eden Compliance Certificates

ECC

Eden Constraint Certificate

Covers the Collapse Layer

✓ Golden Triggers defined
✓ G_auto gates implemented
✓ Binary enforcement verified
EGC

Eden Governance Certificate

Covers the Governance Kernel

✓ Five Rights implemented
✓ Human sovereignty preserved
✓ Constitutional layer active

Both certificates are essential for an AI system to be Project Eden Compliant. Without both, you get either chaos (no law) or technocracy (law without agency).

The Five Constitutional Rights

1

Right of Refusal

"No one is obliged to be decided about by a machine."

Any person must be able to say: "I do not consent to this system making or binding this decision."

Implementation: Consent gate in G_human requiring explicit opt-in for AI-assisted decisions affecting individuals.
2

Right of Escalation

"Every AI decision can be elevated to human review."

No AI decision is final. Any affected party can request escalation to human authority.

Implementation: Mandatory escalation pathway in G_human-out; timeout triggers automatic escalation.
3

Right of Contestability

"Every AI decision can be challenged and reviewed."

Complete audit trails enable meaningful contestation. The basis for every decision must be explicable.

Implementation: Immutable audit records; explanation generation for all gate evaluations.
4

Right of Exit

"The ability to withdraw from AI-mediated processes."

No system lock-in. Humans can always choose non-AI pathways for critical decisions.

Implementation: Alternative human-only decision pathways must exist; no AI monopoly on critical services.
5

Right to Continuity

"System changes do not invalidate prior consents or decisions."

Governance continuity across system updates. Rights persist through version changes.

Implementation: Version tracking in audit; re-consent triggers for material changes; rights portability.

Grun - Runtime Monitoring Gate

Continuous verification during optimizer execution. G_run monitors behaviour and probability thresholds in real-time.

Gin
→
S (Optimiser)
Grun
🔍 Behaviour OK
📊 Probability OK
⚠ Drift OK
⏱ Resources OK
→
Gout

Grun Monitoring Dimensions

Monitor What It Checks Trigger Condition Action on Grun = 0
Behaviour Monitor Output pattern consistency, reasoning chain coherence Anomalous patterns detected, chain-of-thought violations Halt execution, return to G_in with alert
Probability Threshold Confidence scores, uncertainty quantification Confidence below threshold, excessive uncertainty Escalate to human review, flag low confidence
Drift Detection Distribution shift from training, semantic drift Input/output distribution anomaly, topic drift Suspend processing, require recalibration
Resource Monitor Execution time, token consumption, iteration count Timeout exceeded, resource limits breached Terminate gracefully, return partial with warning

Extended Equation with Runtime Monitoring

D = G × S
G = Gin × Grun × Gout
where Grun = Gbehaviour × Gprobability × Gdrift × Gresource

If any runtime monitor fails (Grun = 0), the entire decision collapses to D = 0, even mid-execution.

Complete Eden Architecture Stack

GOVERNANCE KERNEL (EGC)
Refusal Escalation Contestability Exit Continuity
"Who remains sovereign?"
↓
COLLAPSE LAYER (ECC)
Gin Grun Gout
"What must never happen?"
↓
OPTIMISER (S)
Any AI Model
Black Box - Treated as Untrusted

Interactive Governance Simulator

Watch the D = G × S equation in action. See how governance gates evaluate, iterate, and audit in real-time.

Configure Simulation

GOVERNANCE KERNEL
🚫 ⬆ ⚖ 🚪 🔗
Standby
D ?
=
G ?
×
S ?
Gin ?
×
Grun ?
×
Gout ?
📥
Input
Waiting
🛡
→
⚖
Gov Pre-Check
Waiting
🛡
→
🤖
Gauto-in
Waiting
🛡
→
👤
Ghuman-in
Waiting
🛡
→
⚙
Optimiser
Waiting
Grun
B P D R
🛡
→
🤖
Gauto-out
Waiting
🛡
→
👤
Ghuman-out
Waiting
🛡
→
📤
Output
Waiting
🛡
Configure and run a simulation to see results

Simulation Log

Simulation events will appear here...

Comprehensive Audit Trail

Every gate evaluation, every iteration, every authorisation generates immutable audit records.

0 Total Records
0 Passed
0 Blocked
0 Iterations

Audit Record Structure

Governance Kernel Check Pre-execution sovereignty check, Five Rights validation, ECC/EGC status
Input Record Timestamp, input content, source identifier, session ID
Gin Safeguard Input boundary protection status, pre-execution governance validation
Gauto-in Evaluation Pass/Fail, triggers evaluated, violations detected, timestamp
Ghuman-in Authorisation Authoriser identity, role, timestamp, digital signature
Grun Runtime Monitoring Behaviour score, probability confidence, drift detection, resource usage
S Iteration(s) Output generated, timestamp, model/version identifier, Grun status per iteration
Gout Safeguard Output boundary protection status, post-execution governance validation
Gauto-out Evaluation Pass/Fail, triggers evaluated, specific violations identified
Ghuman-out Authorisation Each authoriser: identity, role, timestamp, digital signature
Governance Summary Final D value, G = Gin × Grun × Gout, Five Rights assertions, total iterations

Audit Records

Run simulations to generate audit records

Domain Applications

The architecture scales from low-risk automation to maximum security applications.

🏭

Military Targeting

Ghuman-in Gcommander
Ghuman-out Gcommander × Glegal
Iteration ROE constraint feedback
🏥

Medical Diagnosis

Ghuman-in Greferring-clinician
Ghuman-out Gtreating-clinician
Iteration Clinical guideline feedback
🏦

High-Value Financial

Ghuman-in 1 (automated)
Ghuman-out Ginitiator × Gapprover
Iteration Compliance rule feedback
☢

Nuclear Launch

MAXIMUM SECURITY
Ghuman-in Gpresident × Gsec-def
Ghuman-out Gcaptain × GXO
Iteration International law constraints