YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Marlborg-WORM
Self-Modifying Sovereign Agent
Hardware-enforced cognitive strain protection. Quantum to silicon.
Marlborg-WORM is a self-modifying computation engine that explores what happens when an autonomous agent can rewrite its own rules while a hardware monitoring layer observes the computational strain of that process in real-time.
The system is not merely a software agent.
It is a full-stack research implementation spanning from quantum circuit descriptions down to 7nm ASIC tapeout constraints.
The central question:
Can a self-modifying system be made to observe and constrain its own transformation without an external authority?
The Principle
THE ATTACKER'S EFFORT BECOMES THEIR DEFEAT.
MORE STRAIN.
MORE ENTROPY.
FASTER LOCKOUT.
Any attempt to inject, probe, or reverse-engineer the system generates computational work.
That work is observable.
That observation is enforced in hardware.
The harder an attacker pushes, the faster the system recognizes the threat and closes the boundary.
Architecture
Quantum (Q#, Circom, Lean 4) β What it computes
Clash / Haskell β Hardware specification
SystemVerilog / Verilog / BSV β Synthesizable RTL
SVA + SymbiYosys β Formal verification
WDDL + Jitter Engine β Side-channel resistance
7nm SDC + UPF + DRC β Physical implementation
Rust + C β Runtime monitoring + networking
Common Lisp + Janet β The VM itself
Lean 4 β Mathematical proof of correctness
Docker β Deployment
The architecture is intentionally deep.
Each layer adds a different kind of guarantee.
Cognitive Strain Model
The system continuously computes cognitive entropy:
H_cog = H_base + H_trunc + H_hash + H_marlborg
Where:
H_base = 0.10 nats (constant baseline)
H_trunc = N Γ 0.00001665 nats per operation
H_hash = 0.005 nats penalty when hash integrity is removed
H_marlborg = ΞR Γ 0.0005 nats per rule installed
ICP = max(0, H_cog β H_safe)
Thresholds:
Safe Limit β 0.20 nats
Warning β 0.30 nats
Critical Lockout β 0.40 nats
440 Rules β ACCESS PERMANENTLY DENIED
The strain monitor lives in an always-on power domain.
It cannot be bypassed by clock glitching, power collapse, or voltage fault injection.
The Execution Pipeline
RULE CHANGE
β
REWRITE ENGINE
β
EXECUTION LOAD
β
STRAIN OBSERVATION (hardware, always-on)
β
THRESHOLD CHECK
β
ACCEPT / REJECT / LOCKOUT
The system does not merely check whether a rule is syntactically valid.
It checks whether the act of processing that rule produces a strain signature consistent with legitimate operation.
Security Layers
| Layer | Mechanism | Defeats |
|---|---|---|
| Cryptographic | Ed25519 + SHA3-256 + WORM chain | Forgery, replay, state corruption |
| Zero-Knowledge | Circom ZK-SNARKs (ICP auth guard) | Information leakage during auth |
| Hardware | Always-on strain monitor (7nm ASIC) | Bypass, clock glitch, power collapse |
| Side-Channel | WDDL + jitter engine (2^20 DPA traces) | Power analysis, timing attacks |
| Radiation | TMR + pseudo-ELT (300 krad TID) | SEU, cosmic ray bit-flips |
| Formal | Lean 4 proofs + SVA assertions | Logical errors, specification gaps |
Formally Verified Properties
The following have been proven mathematically:
Convergence
Trace distance contracts by Ξ± β€ 1/2 per cycle.
(Banach fixed-point theorem.)
Real-time compliance
Worst-case jitter: 150ns < 1000ns deadline.
850ns margin for crypto computation.
Metastability freedom
Isolation asserts before power collapse.
Releases only after power stability confirmed.
Chain integrity
Append-only WORM chain with cryptographic hash linkage.
No deletion. No rewrite. No forgetting.
Involution
Quantum walk is its own inverse.
(Fβ wormhole walk proof.)
The WORM Chain
Write Once Read Many.
OPERATION
β
HASH (SHA3-256)
β
APPEND TO CHAIN
β
LINK TO PREVIOUS
β
SEAL
The chain cannot be edited.
Every rule installation, every state transition, every access attempt is permanently recorded.
The system cannot forget what it has done.
Self-Modification Under Constraint
Marlborg-WORM allows rules to modify other rules.
This is deliberate.
The research question is not whether self-modification is possible.
The research question is whether self-modification can be made observable and constrained without removing the capability entirely.
The answer explored here is:
Allow modification.
Observe the modification.
Measure the cost of the modification.
Reject modifications that exceed the strain envelope.
Record everything regardless.
Hardware Implementation
The system is designed to be physically realizable.
Target: TSMC N7FFC (7nm FinFET)
Core voltage: 0.72V
IO voltage: 1.8V
Frequency: 100 MHz
Core area: 0.16 mmΒ²
Total power: 14.2 mW (active)
Sleep power: 1.82 mW (strain monitor only)
Power savings: 87.2% during idle
TID tolerance: > 300 krad(Si)
SEU rate: < 1e-10 errors/bit/day
The strain monitor remains powered during all sleep states.
There is no moment when the system is not watching.
Build
# VM (requires SBCL + Janet)
sbcl --load src/primitives.lisp
# Hardware (requires Clash + Yosys)
clash --verilog hardware/clash/SovereignShiftTruncator.hs
yosys -p "read_verilog hardware/*.v; synth"
# Formal verification
lean4 quantum/JitterRealTime.lean
lean4 quantum/ShadowWalk.lean
# Docker (monitoring daemon)
docker build -f deploy/Dockerfile -t marlborg-strain-monitor .
Research Status
This is a research implementation.
The system explores ideas at the intersection of:
- self-modifying computation
- hardware security
- formal methods
- quantum information theory
- cognitive load modeling
Not every component is production-ready.
The architecture is the contribution.
The Name
Marlborg
The WORM is Write Once Read Many.
The combination is intentional.
A self-consuming process that cannot erase its own history.
Copyright
Copyright BEL ESPRIT D ACCORD TRUST HOLDINGS INC.
See LICENSE for the governing terms.
the attacker's effort becomes their defeat.
more strain. more entropy. faster lockout.
verified by design. trusted by hardware.

