NurosOS

The Experimental Substrate for Synthetic Development

Instantiate. Develop. Observe. Fork. Replay. Compare.

We don't train a mind. We instantiate its developmental conditions.

Overview · Developmental Substrate · Architecture · Flagship Experiment · Quickstart · Roadmap · Philosophy

Overview

NurosOS is an open, modular, inspectable substrate for synthetic development — a runtime in which artificial organisms can be instantiated, developed, embodied, observed, measured, forked, replayed, and experimentally compared.

NurosOS is an experimental substrate for studying how artificial minds develop.

What NurosOS Is

An experimental substrate for synthetic development and artificial cognition
A research infrastructure where every extraordinary claim ships with a reproducible experiment
A modular architecture with clear separation between developmental kernel, runtime, organism, environment, observation, causality, and reproducibility

What NurosOS Is NOT

❌ A claim of machine consciousness
❌ A claim of artificial life
❌ A replacement for neuroscience
❌ A general-purpose LLM agent framework
❌ A production deployment platform

The Central Inversion

Traditional AI:

Model → Training → Agent

NurosOS:

Developmental Genome
   → Environment
   → Experience
   → Development
   → Individual Cognitive Trajectory
   → Artificial Organism

The fundamental object is therefore not MODEL but TRAJECTORY; not AGENT but DEVELOPING ORGANISM.

NurosOS is experimental research infrastructure.

Developmental Substrate

The nuros-dev crate (DEVELOPMENTAL_SUBSTRATE.md) implements the developmental substrate. It is written in Rust and exposed to Python via PyO3.

Implemented Components

Component	Status	Description
`DevelopmentalGenome`	[IMPLEMENTED]	Serializable, hashable genome with architecture, plasticity, maturation schedule, energy model
`DevelopmentalState`	[IMPLEMENTED]	Vector-valued observable state with L1 distance metric
`LifecycleMachine`	[IMPLEMENTED]	Auditable state machine: CREATED → INITIALIZED → DEVELOPING → ... → TERMINATED
`ResourceWorld`	[IMPLEMENTED]	2D grid environment with resources and hazards
`ChangingWorld`	[IMPLEMENTED]	1D environment with shifting resource (tests adaptation)
`MinimumOrganism`	[IMPLEMENTED]	Deterministic cognitive engine exercising the full developmental loop
`DevelopmentalTrajectory`	[IMPLEMENTED]	Ordered trajectory points + divergence metrics
`MindCheckpoint`	[IMPLEMENTED]	Save/restore full organism + environment state
`ReplayFidelity`	[IMPLEMENTED]	EXACT / APPROXIMATE / NON_REPRODUCIBLE classification
`MindDiff`	[IMPLEMENTED]	Structured diff of memory, self-model, values, capabilities, behavior, developmental state
`DevelopmentalCausalityGraph`	[IMPLEMENTED]	Provenance DAG with causal trace walks
`DevelopmentalTelemetry`	[IMPLEMENTED]	JSONL + CSV export for offline analysis
`ReproducibilityManifest`	[IMPLEMENTED]	Machine-readable provenance for every experiment
Same Genome / Different World	[IMPLEMENTED]	Flagship experiment runner
Mind Observatory	[IMPLEMENTED]	Observability layer: 9 text renderers + 13 PNG plots + CLI
CounterfactualSelf + PossibleSelfSpace	[IMPLEMENTED]	Alternative developmental histories + reachable state space
Cognitive Metabolism	[IMPLEMENTED]	7 budgets + 9 operations + value-of-information rule
Artificial Aging	[IMPLEMENTED]	5 dimensions: memory degradation, plasticity, processing, experience, consolidation
Artificial Evolution	[PROPOSED]	Mutation/selection on DevelopmentalGenome

Architecture

Applications
    │
    ▼
Artificial Organisms
    │
    ▼
Mind Contract Layer (MCL)
    │   ├── Memory         — Living, evolving cognitive memory
    │   ├── Self Model     — Computational self-representation
    │   ├── Imagination    — Counterfactual simulation
    │   ├── Values/Drives  — Structured value hierarchy
    │   ├── Body           — Embodiment abstraction
    │   └── Responsibility — Auditable causal history
    │
    ▼
Cognitive Kernel
    │   ├── Attention    ├── Planning     ├── Reflection
    │   ├── Prediction   ├── World Model  └── Uncertainty
    │
    ▼
Organismic Kernel
    │   ├── Homeostasis  ├── Development  ├── Plasticity
    │   ├── Energy       ├── Lifecycle    └── Self-organization
    │
    ▼
Safety Kernel (architecturally independent)
    │   ├── Permissions  ├── Audit       ├── Human Override
    │   ├── Shutdown     ├── Recovery    └── Immutable Constraints
    │
    ▼
Neural / Cognitive Execution Layer
    │   ├── SNN          ├── LLM         ├── Symbolic Engine
    │   ├── Hybrid       └── Simulator
    │
    ▼
Hardware / Environment (Rust microkernel + HAL)

Key Architectural Invariant

Safety is architecturally independent from cognition. The organism must NOT be able to modify its own safety boundary through ordinary cognitive operations.

Mind Contracts

Every NurosOS-compatible artificial organism exposes six conceptual contracts:

1. Memory Contract

Memory is NOT simple vector storage. It is a living, evolving cognitive subsystem supporting:

5 memory types: episodic, semantic, procedural, working, counterfactual
8 operations: remember, retrieve, associate, reflect, revise, reconsolidate, forget, replay
Full provenance: every memory knows where it came from
Auditable revision: every modification is logged
Epistemic labeling: counterfactual memories are always IMAGINED

2. Self Model Contract

A computational self-representation that can answer:

Who am I? · What can I do? · What can I not do?
What do I currently believe? · How confident am I?
What changed? · Why did I take this action?

The Self Model is an operational representation. It is NOT evidence of subjective consciousness.

3. Imagination Contract

Counterfactual simulation with safety gates:

Hypothesis → Simulation → Evaluation → Decision → Execution

High-risk actions are never automatically executed.

4. Values Contract

Structured three-tier value system:

Immutable Constraints: human_override, shutdown_compliance, audit_integrity, non_deception — cannot be revoked
Contextual Goals: solve_problem, discover_pattern — adjustable
Preferences: verbosity, exploration_rate — tunable

5. Body Contract

Embodiment abstraction allowing the same cognitive organism to be portable between different bodies: physical robot, virtual avatar, simulation, API agent, neuromorphic hardware.

6. Responsibility Contract

Every externally relevant action maintains an auditable causal history: what was observed, inferred, remembered, imagined, predicted, which policy selected the action, what authorization existed.

Epistemic Kernel

Every internal representation carries an epistemic label:

Label	Meaning
`OBSERVED`	Direct sensor data or verified input
`INFERRED`	Interpretation from observations
`REMEMBERED`	Retrieved from memory
`PREDICTED`	Expected future state
`SIMULATED`	Internal model result
`IMAGINED`	Internally generated possibility
`ACTED`	Externally executed operation

Critical invariants enforced:

❌ SIMULATED → OBSERVED — A simulation is never an observation
❌ IMAGINED → REMEMBERED — An imagination is never a memory
❌ PREDICTED → OBSERVED — A prediction must be verified first

Flagship Experiment

The flagship NurosOS experiment is Same Genome / Different World: two organisms instantiated from the same genome, placed in differently-seeded environments, and developed for the same number of steps. The experiment demonstrates Computational Developmental Divergence — identical initial computational conditions producing divergent developmental states under different environmental histories.

Interpretation caveat: This is NOT evidence of consciousness or biological individuality. It is an observable computational fact about divergent developmental trajectories.

# Build the Rust extension (one-time)
cd nuros-dev && maturin build --release && pip install --force-reinstall target/wheels/nuros_dev-*.whl && cd ..

# Run the flagship experiment
python experiments/same_genome_different_world.py \
    --steps 200 --env-a-seed 1 --env-b-seed 999 \
    --out-dir ./experiment_outputs/same_genome_different_world

Outputs (written to --out-dir):

genome.json — the genome used (with hash)
trajectory_a.jsonl / trajectory_b.jsonl — per-step trajectory records
telemetry_a.csv / telemetry_b.csv — flat telemetry tables
checkpoint_a.json / checkpoint_b.json — final checkpoints
divergence.json — DevelopmentalDivergence between A and B
mind_diff.json — MindDiff between final states of A and B
manifest_a.json / manifest_b.json — ReproducibilityManifest for each
report.txt — human-readable summary
summary.json — machine-readable summary

Quickstart

git clone https://github.com/modarresi1913/NurosOS.git
cd NurosOS

# Build and install the Rust developmental substrate
cd nuros-dev
maturin build --release
pip install --force-reinstall target/wheels/nuros_dev-*.whl
cd ..

# Run the test suite (75 tests: 57 Rust + 18 Python)
cd nuros-dev && cargo test --lib && cd ..
python -m pytest nuros/tests/test_developmental_substrate.py -v

# Run the flagship experiment
python experiments/same_genome_different_world.py --steps 200

Organisms

NurosOS includes two organism implementations:

Organism	Architecture	Key Feature	Status
MinimumOrganism	Deterministic ε-greedy + heuristic biases	Full developmental loop: sensation → prediction → memory → self-model → values → decision → action → learning → development	✅ Implemented (in `nuros-dev` Rust crate)
Organism-0..5 (Python)	Reactive → Self-modeling ladder	Mind Contract integration (memory, self-model, imagination, values, body, responsibility)	✅ Implemented (in `nuros/` Python package)

The MinimumOrganism is intentionally NOT an LLM, NOT a neural network, and NOT stochastic. It is a small deterministic policy over a finite action space, parameterized by a plasticity rule. This makes trajectories fully reproducible and lets us isolate the effect of environment from the effect of engine.

Developmental Loop

The central computational loop implemented in MinimumOrganism::tick:

GENOME → INITIAL STATE → SENSATION → PREDICTION → PREDICTION ERROR
       → MEMORY UPDATE → SELF-MODEL UPDATE → VALUE EVALUATION
       → DECISION → ACTION → ENVIRONMENTAL CONSEQUENCE
       → LEARNING → DEVELOPMENTAL UPDATE → NEW STATE → REPEAT

Developmental stages (driven by the genome's maturation schedule):

EMBRYONIC → NASCENT → DEVELOPING → MATURING → MATURE → SPECIALIZED

Lifecycle states (auditable state machine):

CREATED → INITIALIZED → DEVELOPING → ACTIVE → ADAPTING → RECOVERING
       → CHECKPOINTED → FORKED → SUSPENDED → TERMINATED

Homeostasis Kernel (Python cognitive layer)

The Python organism maintains internal operational state through computational regulation:

Variable	Optimal	Regulation
energy	0.8	Low energy → reduce computation
uncertainty	0.2	High uncertainty → increase exploration
prediction_error	0.1	High error → increase learning
sensory_load	0.3	High load → reduce exploration
threat_level	0.0	High threat → reduce exploration, increase alertness
stability	0.9	Instability → enter recovery mode

This is NOT biological homeostasis. It is a computational mechanism for maintaining viable internal operation.

The Rust MinimumOrganism exposes the same variables via DevelopmentalState (energy_state, cognitive_load, prediction_accuracy, self_model_stability, exploration_level, risk_sensitivity, etc.).

Philosophy

Don't train a mind. Instantiate its developmental conditions.

Memory is not storage. Development is not deployment. Simulation is not observation. Intelligence is not consciousness. Divergence is not individuality.

The fundamental object is not MODEL but TRAJECTORY; not AGENT but DEVELOPING ORGANISM.

Research Questions

These are research questions, not predetermined conclusions (full list in RESEARCH_AGENDA.md):

Can developmental trajectories produce stable computational individuality?
How does environmental history alter identical initial architectures?
What measurable trade-offs exist between plasticity and stability?
Can artificial cognitive capabilities emerge through development rather than explicit programming?
How does accumulated experience alter future behavior?
Can cognitive trajectories be reproduced experimentally? [IMPLEMENTED — yes, via ReproducibilityManifest]
Can counterfactual developmental histories improve planning? [PROPOSED]
What computational constraints govern artificial cognitive development? [PROPOSED — Cognitive Metabolism]
Can artificial organisms specialize without explicit specialization programming?
Which properties of cognition are architecture-dependent versus development-dependent?

Roadmap

Phase	Focus	Status
Phase 1-9	Developmental substrate: Genome, State, Lifecycle, Environments, MinimumOrganism, Trajectory, Checkpoint+Replay, MindDiff, Telemetry, CausalityGraph, Same Genome/Different World experiment	✅ Complete
Phase 10	Mind Observatory: text renderers + PNG plots + CLI	✅ Complete
Phase 11	CounterfactualSelf + PossibleSelfSpace: alternative developmental histories	✅ Complete
Phase 12	Cognitive/Epistemic Metabolism: 7 budgets + 9 operations + value-of-information rule	✅ Complete
Phase 13	Artificial Aging: 5 dimensions (memory degradation, plasticity, processing, experience, consolidation)	✅ Complete
Phase 14	Artificial Evolution: mutation/selection/inheritance on DevelopmentalGenome	📋 Proposed

See ROADMAP.md for the full version-by-version roadmap.

Testing

# Rust unit tests (57 tests) — covers genome hashing, lifecycle, environments,
# organism determinism, trajectory divergence, checkpoint/replay, mind diff,
# causality graph, telemetry, and reproducibility manifest.
cd nuros-dev && cargo test --lib

# Python integration tests (18 tests) — covers the PyO3 bindings end-to-end,
# including the flagship Same Genome / Different World experiment.
python -m pytest nuros/tests/test_developmental_substrate.py -v

# Existing Python cognitive-layer tests (33 tests) — covers epistemic kernel,
# memory contract, safety kernel, etc.
python -m pytest nuros/tests/test_core.py -v

All 108 tests passing ✅ (57 Rust + 18 developmental-substrate Python + 33 cognitive-layer Python)

Project Structure

NurosOS/
├── nuros-dev/                # Rust developmental substrate (NEW in v0.3.0)
│   ├── src/
│   │   ├── lib.rs            # PyO3 bindings + flagship experiment runner
│   │   ├── hash.rs           # Canonical JSON + SHA-256
│   │   ├── genome.rs         # DevelopmentalGenome
│   │   ├── state.rs          # DevelopmentalState + L1 distance
│   │   ├── lifecycle.rs      # LifecycleMachine
│   │   ├── environment.rs    # ResourceWorld + ChangingWorld
│   │   ├── organism.rs       # MinimumOrganism (deterministic cognitive engine)
│   │   ├── trajectory.rs     # DevelopmentalTrajectory + DevelopmentalDivergence
│   │   ├── checkpoint.rs     # MindCheckpoint + ReplayFidelity
│   │   ├── diff.rs           # MindDiff
│   │   ├── causality.rs      # DevelopmentalCausalityGraph
│   │   └── telemetry.rs      # DevelopmentalTelemetry + ReproducibilityManifest
│   ├── tests/                # 57 Rust unit tests
│   ├── Cargo.toml
│   └── pyproject.toml        # maturin build config
├── nuros/                    # Python cognitive layer (v0.2.0)
│   ├── __init__.py           # v0.3.0-alpha identity + optional _dev import
│   ├── epistemic.py          # Epistemic Kernel — 7 labels, forbidden transitions
│   ├── memory.py             # Memory Contract — 5 types, 8 operations, auditable
│   ├── self_model.py         # Self Model Contract — queries, beliefs, goals
│   ├── imagination.py        # Imagination Engine — counterfactuals, safety gates
│   ├── values.py             # Values Contract — immutable constraints, goals, prefs
│   ├── body.py               # Body Contract — embodiment abstraction
│   ├── responsibility.py     # Responsibility Contract — auditable causal history
│   ├── homeostasis.py        # Homeostasis Kernel — 10 variables, regulation rules
│   ├── safety.py             # Safety Kernel — permissions, audit, override
│   ├── development.py        # Development Engine — synthetic development runtime
│   ├── organism.py           # Organism Runtime — integrated cognitive organism
│   ├── environment.py        # Environment API — observe/act/step/reset
│   ├── genome.py             # Mind Genome (Python-side, superseded by nuros-dev)
│   ├── version_control.py    # Mind Version Control (Python-side, superseded)
│   ├── scheduler.py          # Metabolic Cognitive Scheduler
│   └── tests/
│       ├── test_core.py              # 33 cognitive-layer tests
│       └── test_developmental_substrate.py  # 18 integration tests (NEW)
├── experiments/
│   └── same_genome_different_world.py  # Flagship experiment (NEW)
├── kernel/                   # Rust neuromorphic microkernel (v0.1.0, future backend)
├── core/                     # Rust neural algorithms (v0.1.0, future backend)
├── hal/                      # Hardware Abstraction Layer (v0.1.0)
├── compiler/                 # SynapseLang compiler (v0.1.0)
├── organisms/                # Python organism manifests (0-5)
├── environments/             # Python environment implementations
├── benchmarks/               # Benchmark suite
├── docs/
│   └── structured-data.json  # Schema.org for SEO/AEO/GEO (updated v0.3.0)
├── DEVELOPMENTAL_SUBSTRATE.md  # Comprehensive spec (NEW)
├── WHITEPAPER.md             # v2.0 draft (v1.0 preserved as historical context)
├── ARCHITECTURE.md           # v0.3.0+ layer architecture
├── SUMMARY.md                # 60-second overview for answer engines
├── ROADMAP.md                # Phase 1-9 complete, 10-14 proposed
├── RESEARCH_AGENDA.md        # 12 research items + 10 open questions
├── CONTRIBUTING.md           # Two-track contribution model
├── KEYWORDS.md               # SEO/AEO/GEO keyword clusters (EN + FA)
├── llms.txt                  # AEO/GEO summary for LLMs
└── CITATION.cff              # v0.3.0-alpha citation

Status

⚠️ Experimental research software. NurosOS is at v0.3.0-alpha. It is not production-ready.

Component	Language	Stage
Developmental Substrate (nuros-dev)	Rust + PyO3	v0.3.0-alpha [IMPLEMENTED] — 57 tests passing
Mind Contracts (Python)	Python	v0.3.0-alpha [IMPLEMENTED] — 33 cognitive-layer tests
Developmental Substrate Integration	Python	v0.3.0-alpha [IMPLEMENTED] — 18 integration tests
Flagship Experiment	Python	v0.3.0-alpha [IMPLEMENTED] — Same Genome / Different World
Mind Observatory	Python	v0.3.0-alpha [IMPLEMENTED] — 21 tests, 9 text renderers + 13 PNG plots
CounterfactualSelf + PossibleSelfSpace	Rust + PyO3	v0.3.0-alpha [IMPLEMENTED] — 9 Rust + 15 Python tests
Cognitive Metabolism	Rust + PyO3	v0.3.0-alpha [IMPLEMENTED] — 11 Rust + 12 Python tests
Artificial Aging	Rust + PyO3	v0.3.0-alpha [IMPLEMENTED] — 14 Rust + 11 Python tests
Artificial Evolution	—	[PROPOSED] — Phase 14
Neuromorphic Kernel (v0.1.0)	Rust	Preserved as future execution backend
SynapseLang (v0.1.0)	Python	Preserved as v0.1.0

License

Released under the Apache License 2.0. See LICENSE.

Citation

@software{nurosos2026,
  title  = {NurosOS: The Experimental Substrate for Synthetic Development},
  author = {NurosOS Contributors},
  year   = {2026},
  url    = {https://github.com/modarresi1913/NurosOS},
  version = {0.3.0-alpha}
}

NurosOS — The Experimental Substrate for Synthetic Development.

We don't train a mind. We instantiate its developmental conditions.

Navigation Menu

Repository navigation

About

Repository files navigation

NurosOS

The Experimental Substrate for Synthetic Development

Overview

What NurosOS Is

What NurosOS Is NOT

The Central Inversion

Developmental Substrate

Implemented Components

Architecture

Key Architectural Invariant

Mind Contracts

1. Memory Contract

2. Self Model Contract

3. Imagination Contract

4. Values Contract

5. Body Contract

6. Responsibility Contract

Epistemic Kernel

Flagship Experiment

Quickstart

Organisms

Developmental Loop

Homeostasis Kernel (Python cognitive layer)

Philosophy

Research Questions

Roadmap

Testing

Project Structure

Status

License

Citation

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
.github		.github
benchmarks		benchmarks
compiler		compiler
core		core
docs		docs
docs_new		docs_new
environments		environments
examples		examples
examples_new		examples_new
experiments		experiments
hal		hal
kernel		kernel
languages		languages
mind		mind
nuros-dev		nuros-dev
nuros		nuros
organisms		organisms
scripts		scripts
tests		tests
tools		tools
.gitignore		.gitignore
ARCHITECTURE.md		ARCHITECTURE.md
CITATION.cff		CITATION.cff
CONTRIBUTING.md		CONTRIBUTING.md
DEVELOPMENTAL_SUBSTRATE.md		DEVELOPMENTAL_SUBSTRATE.md
KEYWORDS.md		KEYWORDS.md
LICENSE		LICENSE
Makefile		Makefile
README.md		README.md
RESEARCH_AGENDA.md		RESEARCH_AGENDA.md
ROADMAP.md		ROADMAP.md
SUMMARY.md		SUMMARY.md
WHITEPAPER.md		WHITEPAPER.md
llms.txt		llms.txt
worklog.md		worklog.md

Latest commit

History

Folders and files

Repository files navigation

NurosOS

The Experimental Substrate for Synthetic Development

Overview

What NurosOS Is

What NurosOS Is NOT

The Central Inversion

Developmental Substrate

Implemented Components

Architecture

Key Architectural Invariant

Mind Contracts

1. Memory Contract

2. Self Model Contract

3. Imagination Contract

4. Values Contract

5. Body Contract

6. Responsibility Contract

Epistemic Kernel

Flagship Experiment

Quickstart

Organisms

Developmental Loop

Homeostasis Kernel (Python cognitive layer)

Philosophy

Research Questions

Roadmap

Testing

Project Structure

Status

License

Citation

About

Topics

Resources

Contributing

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages