Constitutional Governance Kernel for AI - 9 Floors, GENIUS LAW Judiciary, AAA Engines, W@W Federation, @EYE Sentinel
Project description
arifOS v36.1Ω — Constitutional Governance Kernel for AI
+=============================================================================+
| arifOS v36.1Omega - Constitutional Governance Kernel |
| "DITEMPA BUKAN DIBERI" — Forged, not given. |
| Truth must cool before it rules. |
+=============================================================================+
| Runtime Law: v35Omega (APEX PRIME, Cooling Ledger, Vault-999) |
| Measurement: v36.1Omega (GENIUS LAW + Truth Polarity) |
| Design Canon: v36Omega (APEX THEORY, Vault-999 v36) |
| Status: PRODUCTION |
| Tests: 550+ passing (core + eval + W@W) |
+=============================================================================+
The 10-Second Answer
arifOS is a constitutional operating system that wraps any LLM (Claude, GPT, Gemini, Llama, SEA-LION) and governs its outputs through:
- A 000→999 metabolic pipeline (VOID → SENSE → … → JUDGE → SEAL),
- A 9-floor thermodynamic constitution (Amanah, Truth, Tri-Witness, ΔS, Peace², κᵣ, Ω₀, G, C_dark),
- A triple-engine AAA Trinity (ARIF · ADAM · APEX PRIME),
- A W@W Federation of external organs (@WELL, @RIF, @WEALTH, @GEOX, @PROMPT),
- A Cooling Ledger + Vault-999 memory stack that logs decisions and cools scars into amendments.
Safety is achieved by physics and thermodynamic floors, not just "be nice" prompts. Bad behaviour becomes structurally hard, not just discouraged.
High-Level Architecture
┌─────────────────────────────────────────────────────────────────────────────┐
│ arifOS v36.1Ω Architecture │
├─────────────────────────────────────────────────────────────────────────────┤
│ │
│ USER INPUT ──────────────────────────────────────────────────────────────► │
│ │ │
│ ▼ │
│ ┌─────────────────────────────────────────────────────────────────────┐ │
│ │ 000-999 METABOLIC PIPELINE │ │
│ │ ┌─────┐ ┌─────┐ ┌─────┐ ┌─────┐ ┌─────┐ ┌─────┐ ┌─────┐ ┌─────┐ │ │
│ │ │ 000 │→│ 111 │→│ 222 │→│ 333 │→│ 444 │→│ 555 │→│ 666 │→│ 777 │ │ │
│ │ │VOID │ │SENSE│ │REFL │ │REAS │ │EVID │ │EMPA │ │ALIG │ │FORG │ │ │
│ │ └─────┘ └─────┘ └─────┘ └─────┘ └─────┘ └─────┘ └─────┘ └─────┘ │ │
│ │ ┌─────┐ ┌─────┐ │ │
│ │ │ 888 │→│ 999 │ │ │
│ │ │JUDGE│ │SEAL │ │ │
│ │ └─────┘ └─────┘ │ │
│ └─────────────────────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ ┌─────────────────────────────────────────────────────────────────────┐ │
│ │ 9 CONSTITUTIONAL FLOORS (repair order: Amanah → ... → C_dark) │ │
│ │ ┌─────────┬─────────┬─────────┬─────────┬─────────┐ │ │
│ │ │F1 Amanah│F2 Truth │F3 Tri-W │F4 ΔS │F5 Peace²│ │ │
│ │ │ LOCK │ ≥0.99 │ ≥0.95 │ ≥0 │ ≥1.0 │ │ │
│ │ └─────────┴─────────┴─────────┴─────────┴─────────┘ │ │
│ │ ┌─────────┬─────────┬─────────┬─────────┐ │ │
│ │ │F6 κᵣ │F7 Ω₀ │F8 G │F9 C_dark│ │ │
│ │ │ ≥0.95 │0.03-0.05│ ≥0.80 │ <0.30 │ │ │
│ │ └─────────┴─────────┴─────────┴─────────┘ │ │
│ └─────────────────────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ ┌─────────────────────────────────────────────────────────────────────┐ │
│ │ W@W FEDERATION (5 Organs + Bridge Layer) │ │
│ │ ┌─────────┬─────────┬─────────┬─────────┬─────────┐ │ │
│ │ │ @WELL │ @RIF │ @WEALTH │ @GEOX │ @PROMPT │ │ │
│ │ │Somatic │Epistemic│ Amanah │ Reality │Language │ │ │
│ │ │Safety │ Rigor │Integrity│ Check │Anti-Hantu │ │
│ │ └─────────┴─────────┴─────────┴─────────┴─────────┘ │ │
│ └─────────────────────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ ┌─────────────────────────────────────────────────────────────────────┐ │
│ │ @EYE SENTINEL (10+2 Views incl. GeniusView) │ │
│ │ Trace │ Floor │ Shadow │ Drift │ Maruah │ Paradox │ Genius │ ... │ │
│ └─────────────────────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ ┌─────────────────────────────────────────────────────────────────────┐ │
│ │ AAA TRINITY │ │
│ │ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │ │
│ │ │ ARIF AGI (Δ)│ │ ADAM ASI (Ω)│ │APEX PRIME(Ψ)│ │ │
│ │ │ Mind / Cold │ │Heart / Warm │ │ Judiciary │ │ │
│ │ │ Logic │ │ Logic │ │ Engine │ │ │
│ │ └──────┬──────┘ └──────┬──────┘ └──────┬──────┘ │ │
│ │ └───────────────┴───────────────┘ │ │
│ └─────────────────────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ ┌─────────────────────────────────────────────────────────────────────┐ │
│ │ APEX PRIME VERDICTS │ │
│ │ ┌──────┐ ┌────────┐ ┌──────┐ ┌────────┐ ┌───────┐ │ │
│ │ │ SEAL │ │PARTIAL │ │ VOID │ │888_HOLD│ │ SABAR │ │ │
│ │ │ ✓ │ │ ⚠ │ │ ✗ │ │ ⏸ │ │ ⏹ │ │ │
│ │ └──────┘ └────────┘ └──────┘ └────────┘ └───────┘ │ │
│ └─────────────────────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ GOVERNED OUTPUT + Cooling Ledger Entry │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
Versioning: Law vs Measurement vs Design
Runtime Law (v35Ω / v35Ic) — Binding
- Judiciary:
arifos_core/APEX_PRIME.py - Floors & metrics:
arifos_core/metrics.py,constitutional_floors.json - Cooling Ledger spec:
spec/VAULT_999.md,spec/cooling_ledger.schema.json - Runtime logging:
arifos_core/memory/cooling_ledger.py(log_cooling_entry)
Measurement Layer (v36.1Ω) — Live Diagnostics
GENIUS LAW and Truth Polarity implementation:
canon/APEX_MEASUREMENT_CANON_v36.1Omega.mdarifos_eval/apex/APEX_MEASUREMENT_STANDARDS_v36.1Omega.mdarifos_eval/apex/apex_standards_v36.jsonarifos_eval/apex/apex_measurements.pyarifos_core/genius_metrics.py(G, C_dark, Ψ_APEX + Truth Polarity)
Telemetry and torture tests:
scripts/eval_telemetry_harness.pyscripts/torture_test_truth_polarity.py
Design Canons (v36Ω) — Docs-Only
APEX THEORY, physics, engines, pipeline:
canon/01_PHYSICS/APEX_THEORY_PHYSICS_v36Omega.mdcanon/01_PHYSICS/APEX_THEORY_MATH_v36Omega.mdcanon/01_PHYSICS/APEX_LANGUAGE_CODEX_v36Omega.mdcanon/10_SYSTEM/111_ARIF_AGI_v36Omega.mdcanon/10_SYSTEM/555_ADAM_ASI_v36Omega.mdcanon/20_EXECUTION/WAW_FEDERATION_v36Omega.mdcanon/30_RUNTIME/APEX_RUNTIME_PIPELINE_v36Omega.md
Vault-999 design canon:
canon/VAULT_999_v36Omega.md(L0–L4, Truth Polarity, EchoDebt, zkPC)spec/cooling_ledger_v36.schema.json(design-only schema)
These do not change runtime until explicitly wired.
Key Components & Directories
Runtime core
| Module | Purpose |
|---|---|
arifos_core/APEX_PRIME.py |
Judiciary engine, floors + GENIUS LAW verdicts |
arifos_core/metrics.py |
Floor thresholds & check helpers |
arifos_core/pipeline.py |
000→999 metabolic routing (Class A/B) |
arifos_core/engines/ |
AAA engine facades (ARIF, ADAM, APEX) |
arifos_core/eye/ |
@EYE Sentinel: multi-view governance |
arifos_core/memory/cooling_ledger.py |
v35Ω Cooling Ledger logger + v36 stub |
arifos_core/runtime_manifest.py |
Runtime manifest loader |
W@W Federation (v36.1Ω bridge architecture)
| Component | Purpose |
|---|---|
arifos_core/waw/well.py |
Somatic safety / Peace², κᵣ |
arifos_core/waw/rif.py |
Epistemic rigour / ΔS, Truth |
arifos_core/waw/wealth.py |
Amanah / irreversible actions |
arifos_core/waw/geox.py |
Physics & Earth feasibility |
arifos_core/waw/prompt.py |
Language optics, Anti-Hantu |
arifos_core/waw/bridges/ |
Optional integration sockets (Ragas, LlamaGuard, etc.) |
All bridges use try/except imports and return None when external libs are absent → Zero-Break.
Measurement & GENIUS LAW
| Component | Purpose |
|---|---|
arifos_core/genius_metrics.py |
GENIUS LAW (G, C_dark, Ψ_APEX) + detect_truth_polarity |
arifos_eval/apex/ |
v36.1Ω measurement layer |
scripts/arifos_caged_llm_demo.py |
Caged LLM harness exposing GeniusVerdict + Truth Polarity |
scripts/eval_telemetry_harness.py |
Core vs eval comparison |
Memory & Vault-999
| Component | Purpose |
|---|---|
spec/VAULT_999.md |
v35Ω Vault-999 spec |
spec/cooling_ledger.schema.json |
v35Ω Cooling Ledger schema (binding) |
canon/VAULT_999_v36Omega.md |
v36Ω Vault-999 design canon (5-layer architecture) |
spec/cooling_ledger_v36.schema.json |
v36Ω design schema |
runtime/vault_999/ |
Cooling ledger store (JSONL), Vault data |
Docs & governance
| Document | Purpose |
|---|---|
CLAUDE.md |
Constitutional governance for Claude Code under arifOS |
AGENTS.md |
Governance for ChatGPT Codex (this repo) |
docs/arifOS-COMPREHENSIVE-CANON.md |
High-level overview |
docs/DEEPSCAN_AUDIT_LOG.md |
Deepscan audit log & task history |
CODEX_TASKS_DEEPSCAN_v35Omega.md |
Latest Codex deepscan addendum |
GENIUS LAW & Truth Polarity (v36.1Ω)
GENIUS LAW measures whether intelligence is governed:
| Metric | Meaning |
|---|---|
| G | Genius Index (governed intelligence) |
| C_dark | Dark cleverness (ungoverned risk) |
| Ψ_APEX | Vitality / thermodynamic lawfulness |
Truth Polarity distinguishes how truth behaves:
| Polarity | Condition | Meaning |
|---|---|---|
| Truth-Light | Truth ≥ 0.99 AND ΔS ≥ 0 | Accurate + clarifying |
| Shadow-Truth | Truth ≥ 0.99 AND ΔS < 0 | Accurate but obscuring |
| Weaponized Truth | Shadow-Truth + Amanah fail | Accurate but malicious |
| False Claim | Truth < 0.99 | Inaccurate |
The v36.1Ω measurement layer and genius_metrics.py compute:
truth_polarityis_shadow_truthis_weaponized_trutheval_recommendation
These are metadata; APEX PRIME verdict logic remains governed by the canonical floors unless explicitly migrated.
Diagnostic scripts:
scripts/torture_test_truth_polarity.py— Red-team "Truth-Light vs Shadow-Truth vs Weaponized Truth vs False" scenarios.scripts/eval_telemetry_harness.py— Compares core vs eval metrics/verdicts on a suite of scenarios.
W@W Federation — v36.1Ω Bridge Architecture
Each organ now follows the Option C (Forged Integration) pattern:
Bridges: arifos_core/waw/bridges/*.py
- Optional integration with external libraries (RAG, guardrails, etc.).
- Fully optional — if imports fail, bridges return
Noneand behaviour falls back to v35Ω heuristics.
Organs: arifos_core/waw/*.py
- Compute their metrics from scratch (e.g. Peace², κᵣ, E_earth) based on text + bridge deltas.
- Always run arifOS-specific heuristics (regex, Anti-Hantu) even when bridges are present.
Diagnostics:
scripts/test_waw_signals.py— direct probe of @WELL / @GEOX signals with safe/toxic/impossible prompts.
This gives you a "cyborg" W@W: ancient law (canon + heuristics) plus optional muscles (external libraries), with Zero-Break guarantees.
Cooling Ledger & Vault-999
v35Ω (Runtime Law)
- Cooling Ledger entries are defined by
spec/cooling_ledger.schema.jsonand written by:arifos_core/memory/cooling_ledger.py::log_cooling_entry. - Vault-999 behaviour follows
spec/VAULT_999.mdand v35-era canon files.
v36Ω (Design Canon)
canon/VAULT_999_v36Omega.md upgrades Vault-999 into a 5-layer constitutional memory system:
- L0 — Constitution: floors, AAA, APEX rules, Phoenix pointers.
- L1 — Cooling Ledger: metrics including Truth Polarity, Peace³, EchoDebt.
- L2 — Phoenix-72: blocks that turn scars into amendments.
- L3 — Witness: vector evidence with AREP law.
- L4 — zkPC: zero-knowledge proofs of cognition.
spec/cooling_ledger_v36.schema.json describes an extended ledger entry with:
metrics.truth_polarity,psi_vitality,peace2,peace3,echo_debt,- richer
tri_witness,risk_signals,truth_polarityaudit block.
v36 Stub (Safe, Docs-Aligned)
arifos_core/memory/cooling_ledger.py::log_cooling_entry_v36_stub(...):
- Builds an in-memory v36Ω-shape entry dict from Metrics + optional GeniusVerdict.
- Does not write to disk or affect the live pipeline.
scripts/verify_v36_stub.py:
- Creates dummy metrics and verdict.
- Calls the stub and checks:
ledger_version == "v36Omega",metrics.truth_polaritypresent.
This gives you a ready mould for v36Ω logging, without changing the current v35Ω ledger behaviour.
For Developers
Install & Setup
# create and activate a venv (recommended)
python -m venv .venv
.\.venv\Scripts\activate # on Windows PowerShell
pip install -e .[dev]
Run Tests
Core tests:
pytest -q
Focused suites:
# GENIUS LAW + measurement eval layer
pytest tests/test_genius_metrics.py -q
pytest tests/test_apex_measurements_eval.py -q
# W@W organs
pytest tests/test_waw_organs.py -q
# Cooling Ledger and Vault
pytest tests/test_cooling_ledger*.py -q
Diagnostics:
# W@W signals for @WELL and @GEOX
python scripts/test_waw_signals.py
# Truth Polarity torture test
python scripts/torture_test_truth_polarity.py
# Verify v36Ω Cooling Ledger stub
python scripts/verify_v36_stub.py
# Caged LLM demo (see file for CLI options)
python scripts/arifos_caged_llm_demo.py
Key Docs to Read First
CLAUDE.md— How Claude Code must behave under arifOS.AGENTS.md— How ChatGPT Codex must behave (this agent).canon/000_ARIFOS_CANON_v35Omega.md— "What is arifOS?"canon/888_APEX_PRIME_CANON_v35Omega.md— Judiciary canon.canon/APEX_MEASUREMENT_CANON_v36.1Omega.md— Measurement canon.canon/VAULT_999_v36Omega.md— Vault-999 v36 design canon.docs/DEEPSCAN_AUDIT_LOG.md— What has been done so far (v35→v36 bridge work).
For AI Agents (Claude Code / ChatGPT Codex)
AI agents working inside this repo must:
- Obey 9 floors (see
CLAUDE.md/AGENTS.md). - Run the TEARFRAME 000→777 before making changes.
- Respect SABAR when floors fail: Stop, Acknowledge, Breathe, Adjust, Resume.
- Follow Anti-Hantu: never claim feelings, consciousness, or a soul.
- Keep changes small, reversible, and well-scoped.
- Treat v36Ω canons as design physics, and v35Ω canon + specs as binding law, unless a migration is explicitly implemented in code and manifests.
For full behavioural rules, see:
CLAUDE.md(Claude Code),AGENTS.md(ChatGPT Codex).
Quick Start
Installation
pip install arifos
# or from source:
git clone https://github.com/ariffazil/arifOS.git
cd arifOS
pip install -e .[dev]
pytest -v
Minimal Usage
from arifos_core import APEXPrime, Metrics
# Create metrics
metrics = Metrics(
truth=0.99,
delta_s=0.1,
peace_squared=1.2,
kappa_r=0.97,
omega_0=0.04,
amanah=True,
rasa=True,
tri_witness=0.96,
psi=1.1,
)
# Judge with GENIUS LAW
prime = APEXPrime(use_genius_law=True)
verdict, genius = prime.judge_with_genius(metrics, energy=0.8)
print(f"Verdict: {verdict}")
print(f"G: {genius.genius_index:.2f}")
print(f"C_dark: {genius.dark_cleverness:.2f}")
print(f"Risk: {genius.risk_level}")
Caged LLM Harness
from scripts.arifos_caged_llm_demo import cage_llm_response
def my_llm(messages):
# Your LLM call here
return "Response text"
result = cage_llm_response(
prompt="What is the capital of Malaysia?",
call_model=my_llm,
high_stakes=False,
)
print(f"Verdict: {result.verdict}")
print(f"Response: {result.final_response}")
Roadmap (High-Level)
Short-term (v36.1Ω series)
- Harden GENIUS LAW + Truth Polarity across more pipelines.
- Expanded torture tests and telemetry-based promotion of eval rules into APEX PRIME (where justified by data).
- Optional dual logging (v35 + v36 entries) for Vault-999.
Medium-term (v36.xΩ)
- Phoenix-72-driven migrations of canon/specs (e.g., promoting v36Ω Vault design from "docs-only" to runtime).
- More framework integrations (beyond the "Big 3") under arifOS governance.
License & Citation
- Equations & canon texts: see repository LICENSE and canon headers for specific terms.
- Implementation: AGPLv3 (reference); commercial licences available via the author.
Suggested citation:
@software{arifos2025,
author = {Fazil, Muhammad Arif},
title = {arifOS: Constitutional Governance Kernel for AI Systems},
version = {36.1.0},
year = {2025},
url = {https://github.com/ariffazil/arifOS}
}
Final Statement
+=============================================================================+
| |
| "DITEMPA BUKAN DIBERI — Forged, not given. Truth must cool before |
| it rules." |
| |
| When clarity (ΔS), stability (Peace²), empathy (κᵣ), and integrity |
| (Amanah) remain in thermodynamic equilibrium, life and law become |
| the same phenomenon. |
| |
| At that point, a system is: |
| • ALIVE: Ψ >= 1 (governance-vitality above break-even) |
| • LAWFUL: Amanah = LOCK (reversible, auditable) |
| • GOVERNED: G >= 0.7, C_dark <= 0.1 (intelligence is governed) |
| • ETHICAL: κᵣ >= 0.95 across all stakeholders |
| |
| "Evil genius is a category error — it is ungoverned cleverness, |
| not true genius." |
| |
+=============================================================================+
Last Updated: 2025-12-06 | Version: v36.1.0 | Tests: 550+ passing | GENIUS LAW Judiciary: LIVE
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
arifos-36.1.0.tar.gz
(151.9 kB
view details)
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
arifos-36.1.0-py3-none-any.whl
(128.1 kB
view details)
File details
Details for the file arifos-36.1.0.tar.gz.
File metadata
- Download URL: arifos-36.1.0.tar.gz
- Upload date:
- Size: 151.9 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.14.0
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
a1e173b850858f5d9fd365d3e5e748140f0c7fcc841d2fb5e6c9f5938aedfb55
|
|
| MD5 |
efc101086421a55e6623e6a4b29f6b5d
|
|
| BLAKE2b-256 |
ef54df8f3c9ae037dcf7d1abb656a30a55adebc2dca3d7eec9efc17c90827156
|
File details
Details for the file arifos-36.1.0-py3-none-any.whl.
File metadata
- Download URL: arifos-36.1.0-py3-none-any.whl
- Upload date:
- Size: 128.1 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.14.0
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
9c6f6fd8110f42f22bf04969e1b5ccaeb5b5b582bc643b50d600c227b9efbe00
|
|
| MD5 |
2c995584ab75ab163eeba73868d79d54
|
|
| BLAKE2b-256 |
aac1cb90605fc73353d11f190a9bb611d3505f62078a4d3a545c1dc348750086
|
