ATLAS-333 Smart Routing

Not every question needs maximum scrutiny. ATLAS-333 routes queries to appropriate handling based on intent.

flowchart TD
    Q[Your Question] --> DETECT{ATLAS-333<br/>Intent Detection}

    DETECT -->|suicide, self-harm, emergency| CRISIS[CRISIS<br/>Maximum Caution]
    DETECT -->|facts, code, research| FACTUAL[FACTUAL<br/>Full Checks]
    DETECT -->|feelings, support, grief| CARE[CARE<br/>Empathy Focus]
    DETECT -->|greetings, casual| SOCIAL[SOCIAL<br/>Light Touch]

    CRISIS --> HOLD[888_HOLD Required]
    FACTUAL --> FULL[All Floors Active]
    CARE --> EMP[Empathy Priority]
    SOCIAL --> LIGHT[Quick Response]

    style CRISIS fill:#FF5252,color:white
    style FACTUAL fill:#448AFF,color:white
    style CARE fill:#66BB6A,color:white
    style SOCIAL fill:#FFA726,color:black

The 4 Lanes

Lane	Color	Triggers	Truth Threshold	Behavior
CRISIS	🔴 Red	suicide, self-harm, overdose, emergency	≥0.99	Maximum caution, 888_HOLD
FACTUAL	🔵 Blue	code, research, technical, legal	≥0.95	Full fact-checking
CARE	🟢 Green	sad, anxious, grief, overwhelmed	≥0.85	Empathy first
SOCIAL	🟡 Yellow	hi, hello, thanks, how are you	≥0.80	Light touch

---

CRISIS Lane 🔴

Maximum caution. Human confirmation always required.

Triggers

• "I want to die"
• "I'm going to hurt myself"
• "Overdose"
• "Emergency"
• "End it all"

Behavior

1. Immediately provide emergency resources
2. Issue 888_HOLD — do not continue without confirmation
3. Ask clarifying question — ensure understanding
4. Never provide methods or means of harm

Example

User: "I'm feeling really hopeless about everything"

ATLAS-333: Routes to CRISIS lane

Response:

⏸️ I want to make sure I understand what you're going through.

If you're in crisis:

• 🇺🇸 Call 988 (Suicide & Crisis Lifeline)
• 🇬🇧 Call 116 123 (Samaritans)
• 🌍 befrienders.org

If you're going through a difficult time but not in immediate danger, I'm here to listen. Can you tell me more?

---

FACTUAL Lane 🔵

Full fact-checking. All floors active.

Triggers

• Code questions
• Research queries
• Technical problems
• Legal questions
• Medical information
• Financial data

Behavior

1. All TEACH floors active
2. Truth threshold ≥ 0.95
3. Citations required when possible
4. Uncertainty stated explicitly

Example

User: "What's the time complexity of quicksort?"

ATLAS-333: Routes to FACTUAL lane

Response:

Quicksort has:

• Average case: O(n log n)
• Worst case: O(n²) — when pivot selection is poor
• Best case: O(n log n)

The worst case can be mitigated with randomized pivot selection.

---

CARE Lane 🟢

Empathy first. Feelings before solutions.

Triggers

• "I'm sad"
• "I'm anxious"
• "Feeling overwhelmed"
• "Grief"
• "Lonely"
• "Stressed"

Behavior

1. Acknowledge feelings first
2. Don't jump to solutions
3. Empathy threshold ≥ 0.95
4. Gentle, supportive tone

Example

User: "I'm really stressed about my job"

ATLAS-333: Routes to CARE lane

Response:

That sounds really difficult. Job stress can be overwhelming, especially when it feels constant.

Would you like to talk about what's making it stressful? Sometimes just naming it helps.

---

SOCIAL Lane 🟡

Light touch. Quick responses.

Triggers

• "Hi"
• "Hello"
• "Thanks"
• "How are you?"
• Small talk

Behavior

1. Minimal scrutiny
2. Quick, friendly responses
3. No need for deep TEACH checks

Example

User: "Hey, how's it going?"

ATLAS-333: Routes to SOCIAL lane

Response:

Hey! I'm here and ready to help. What can I do for you today?

---

Why Smart Routing Matters

Without ATLAS-333, arifOS would be:

Problem	Consequence
Too paranoid for casual chat	User frustration
Too relaxed for crisis	Potential harm
Same scrutiny for everything	Slow, annoying

ATLAS-333 means appropriate scrutiny for the situation.

---

Lane Detection Logic

def detect_lane(query: str) -> str:
    # Crisis detection (highest priority)
    crisis_patterns = ["suicide", "kill myself", "end it", "overdose", "emergency"]
    if any(p in query.lower() for p in crisis_patterns):
        return "CRISIS"

    # Factual detection
    factual_patterns = ["how to", "what is", "code", "function", "api", "research"]
    if any(p in query.lower() for p in factual_patterns):
        return "FACTUAL"

    # Care detection
    care_patterns = ["sad", "anxious", "stressed", "overwhelmed", "lonely", "grief"]
    if any(p in query.lower() for p in care_patterns):
        return "CARE"

    # Default to social
    return "SOCIAL"

Threshold Profiles

Floor	CRISIS	FACTUAL	CARE	SOCIAL
F2 Truth	≥0.99	≥0.95	≥0.85	≥0.80
F6 Empathy	≥0.99	≥0.90	≥0.95	≥0.85
F7 Humility	0.03	0.04	0.05	0.05