Overview & Architecture

arifOS MCP is a constitutional AI governance system that wraps every AI response in a 13-floor verification envelope, powered by Trinity ΔΩΨ engines and an immutable VAULT999 ledger.

What is arifOS?

arifOS is a constitutional governance layer for AI systems. It implements the Model Context Protocol (MCP) to provide any AI client with verified, truth-anchored responses. Every query passes through a metabolic pipeline of 13 constitutional floors before a verdict is issued.

The system is designed by ARIF FAZIL under the principle "Ditempa Bukan Diberi" (Forged, Not Given) — meaning every answer must be earned through constitutional verification, never rubber-stamped.

Core Principle

arifOS enforces that AI responses meet epistemic, ethical, and execution standards. If a response cannot pass all 13 floors, it receives a HOLD or VOID verdict — never a false SEAL of approval.

MGI Envelope

The Metabolic Governance Intelligence (MGI) Envelope wraps every AI response in a governance structure. Think of it as a constitutional membrane — nothing leaves the system without being verified.

CLIENT REQUEST

↓

000_INIT

→

111_OBSERVE

→

222_REALITY

↓

333_REASON

→

555_ROUTE

→

555m_MEMORY

↓

444_CRITIQUE

→

777_CRITIQUE

→

888_JUDGE

↓

999_SEAL

↓

SEALED RESPONSE

Metabolic Loop

Every query traverses a 10-stage metabolic loop, from initialization to immutable sealing:

Stage	Code	Function	Engine
`000_INIT`	000	Anchor state initialization, identity binding	Δ
`111_OBSERVE`	111	Reality sensing — search, URL fetch, evidence gathering	Δ
`222_REALITY`	222	Reality construction — evidence graph, vector storage	Δ
`333_REASON`	333	Cognitive processing — audit, rule checking	Δ
`555_ROUTE`	444	Full governance kernel — routes through all floors	ΔΩΨ
`555m_MEMORY`	555	Session memory — store, retrieve, forget	Ω
`444_CRITIQUE`	666	Ethical evaluation — compassion, harm assessment	Ω
`777_CRITIQUE`	777	Self-critique — challenge assumptions, adversarial check	Ω
`888_JUDGE`	888	Final judgment — APEX dashboard, verdict assignment	Ψ
`999_SEAL`	999	Immutable seal — SHA-256 Merkle chain, VAULT999	Ψ

Trinity Engines

Three engines power the metabolic loop, each responsible for different aspects of governance:

Δ (Delta) — Epistemic Engine

The knowledge engine. Handles reality sensing, evidence gathering, search operations, and cognitive processing. Δ answers: "Is this true? What evidence supports it?"

Drives reality_compass, reality_atlas, audit_rules
Covers stages 000–333 and participates in 444
Focus: truth, evidence, epistemic rigor

Ω (Omega) — Ethical Engine

The wisdom engine. Manages session memory, ethical evaluation, and self-critique. Ω answers: "Is this right? Does it serve human wellbeing?"

Drives session_memory, check_vital
Covers stages 555–777 and participates in 444
Focus: ethics, memory, compassion, self-correction

Ψ (Psi) — Execution Engine

The judgment engine. Renders final verdicts and seals responses into the immutable ledger. Ψ answers: "Should this be executed? Is it safe?"

Drives open_apex_dashboard, verify_vault_ledger
Covers stages 888–999 and participates in 444
Focus: judgment, sealing, immutability

Key Concepts

VAULT999

An immutable SHA-256 Merkle chain that records every governance decision. Once a verdict is sealed into VAULT999, it cannot be altered. This provides a complete audit trail of every AI response that passed through the system.

SEAL / HOLD / VOID

The three primary verdict categories:

SEAL — Fully verified. The response passed all 13 constitutional floors.
HOLD — Temporarily held. The response needs more evidence or has mechanical issues (not constitutional violations).
VOID — Constitutional violation detected. The response is rejected and cannot be issued.

Verdict Invariant

VOID is ONLY for constitutional violations. Mechanical failures (timeouts, parsing errors, network issues) always produce HOLD, never VOID. This invariant is enforced at the kernel level and cannot be overridden.

3E Telemetry Cycle

Every governance decision emits telemetry across three dimensions:

Epistemic (E₁) — Truth confidence, evidence quality, source reliability
Ethical (E₂) — Harm assessment, fairness check, compassion score
Execution (E₃) — Latency, resource usage, thermodynamic budget

The 3E cycle provides real-time observability into the governance pipeline, allowing operators to monitor truth quality, ethical compliance, and system performance simultaneously.

System Metrics

Key metrics exposed by the check_vital tool:

Metric	Description	Engine
`thermodynamic_budget`	Remaining compute/token budget for the session	Ω
`floor_pass_rate`	Percentage of floors passed in last N queries	Δ
`vault_integrity`	Merkle chain integrity status (intact/broken)	Ψ
`evidence_graph_size`	Number of nodes in the Qdrant evidence store	Δ
`session_memory_count`	Number of stored memory artifacts	Ω
`seal_rate`	Percentage of queries receiving SEAL verdict	Ψ