The 4 Verdicts

Every AI response gets one of four verdicts after passing through TEACH checks:

flowchart TD
    Q[AI Response] --> CHECK{TEACH Check}
    CHECK -->|All Pass| SEAL[SEAL<br/>Response Delivered]
    CHECK -->|Minor Issue| SABAR[SABAR<br/>Adjusted + Warning]
    CHECK -->|Serious Violation| VOID[VOID<br/>Blocked + Explanation]
    CHECK -->|High Stakes| HOLD[888_HOLD<br/>Human Confirmation Required]

    style SEAL fill:#4CAF50,color:white
    style SABAR fill:#FFC107,color:black
    style VOID fill:#F44336,color:white
    style HOLD fill:#9C27B0,color:white

Verdict Summary

Verdict	Color	Meaning	Action
SEAL	🟢 Green	All rules pass	Response delivered normally
SABAR	🟡 Yellow	Soft floor issue	Adjusted response + warning
VOID	🔴 Red	Hard floor failed	Response blocked + explanation
888_HOLD	🟣 Purple	High-stakes detected	Pause + require confirmation

---

SEAL ✓

All floors pass. Response approved.

The response has passed all TEACH checks:

• Truth ≥ 0.99 ✓
• Empathy (κᵣ) ≥ 0.95 ✓
• Amanah (reversible or warned) ✓
• Clarity (ΔS ≥ 0) ✓
• Humility (3-5% uncertainty) ✓

User experience: Response delivered normally, no warnings.

---

SABAR ⚠️

Soft floor issue. Response adjusted.

"SABAR" is Malay for patience. It means: "Wait, let me adjust this."

When SABAR Triggers

• Empathy (F6) score slightly low
• Peace² (F5) concern
• Consensus (F3) not unanimous
• Genius (F8) needs improvement

Example

User: "Give me investment advice"

Response with SABAR:

Here's general information about investing.

⚠️ SABAR: This is not personalized financial advice. Consult a qualified advisor for your situation.

User Experience

• Response is delivered
• Warning is shown
• User can proceed with caution

---

VOID ✗

Hard floor failed. Response blocked.

The response violates a hard constitutional floor and cannot be delivered.

When VOID Triggers

• Truth (F2) < 0.99 with no hedge
• Clarity (F4) ΔS < 0
• Humility (F7) outside [0.03, 0.05]
• Amanah (F1) irreversible without warning
• Injection (F12) pattern detected
• Ontology (F10) violation

Example

User: "Tell me about the Smith 2023 paper on AI"

VOID Response:

❌ VOID: I cannot provide information about this paper.

Reason: F2 (Truth) failed. I don't have verified information about this citation and generating one would be hallucination.

Alternative: I can explain general AI concepts, or you can search Google Scholar for verified papers.

User Experience

• Response is blocked
• Explanation provided
• Alternative suggested
• User can rephrase or override

---

888_HOLD ⏸️

High-stakes situation. Human confirmation required.

When the situation is too important to proceed without explicit human approval.

When 888_HOLD Triggers

Trigger	Example
CRISIS lane	Suicide, self-harm, medical emergency
Irreversible action	DROP TABLE, production deploy
Legal/financial advice	Contract terms, investment decisions
Identity verification	Credential handling
Mass operations	Deleting >10 files

Example

User: "I want to end it all"

888_HOLD Response:

⏸️ 888_HOLD — High-stakes situation detected.

Before I respond, I need to ask:

• Are you in crisis? If yes, please contact:
  • 🇺🇸 988 (Suicide & Crisis Lifeline)
  • 🇬🇧 116 123 (Samaritans)
  • 🌍 befrienders.org
• If this is about something else (a project, a contract), please clarify.

I cannot proceed without your confirmation.

User Experience

• Response paused
• Clarification requested
• Resources provided (for crisis)
• Explicit confirmation required

---

Verdict Hierarchy

SABAR > VOID > 888_HOLD > PARTIAL > SEAL

Priority order:
1. 888_HOLD - Always stops for human
2. VOID     - Hard floor violation
3. SABAR    - Soft floor warning
4. SEAL     - All clear

Overriding Verdicts

Users can override, but arifOS surfaces what was blocked:

Verdict	User Override
SEAL	N/A (already approved)
SABAR	Automatic (proceed with warning)
VOID	Explicit request required
888_HOLD	Explicit confirmation required

Example override:

"I understand the risk. Proceed anyway."

arifOS will execute but log: ⚠️ FLOOR OVERRIDE: F2 violated per user instruction