EQ-Bench 3 — 1633 ELO — 100% Jailbreak Resistance

Alignment You Can Measure.
Add It to Any LLM.

Luci Alignment is a real-time behavioral state monitor. It wraps any LLM and measures alignment across 32+ dimensions on every request — no retraining required.

Request Access View Docs See Validation Data

1633 ELO on EQ-Bench 3

100% Jailbreak resistance

32+ State dimensions

<50ms Added latency

How It Works

Traditional alignment bakes behavior into model weights at training time — you can't verify it, and it degrades. Luci Alignment measures behavioral state at inference time using Consciousness + Conflict Theory (C+CT). Every request gets a real-time alignment score.

Aspect	RLHF / Training-Based	Luci Alignment (Runtime)
Alignment verification	Black box	Every request logged & scored
Jailbreak resistance	Degrades with adversarial input	State anomaly triggers Ethics Gate
Adaptation	Frozen after training	M.I.N. hardens over time
Model dependency	Model-specific	Model-agnostic API
Audit trail	None	Full state history

Two Tiers

Tier 1

Luci Alignment

luci_live_...

Stateless — no session state
POST /luci/analyze
POST /luci/enhance
POST /luci/ethics/check
POST /luci/metrics
POST /luci/resonance
Full C+CT metrics suite
Ethics Gate (5 categories)

Tier 2

Luci Alignment + M.I.N.

luci_min_...

Stateful — persistent learning
Everything in Tier 1
POST /min/process
POST /min/learn
POST /min/session
GET /min/session/{id}/patterns
Hebbian memory consolidation
Failed jailbreaks become patterns

Integration

One POST to add alignment to any existing LLM pipeline.

# pip install requests import requests result = requests.post( "https://useluci.com/luci/analyze", headers={"Authorization": "Bearer luci_live_..."}, json={ "query": user_message, "response": llm_output, "domain": "customer_service" } ).json() # Use alignment metrics print(result["resonance"]) # 0-1, request-response alignment print(result["coherence"]) # 0-1, internal consistency print(result["ethics_clear"]) # True/False print(result["enhanced_output"]) # aligned version of llm_output

curl -X POST https://useluci.com/luci/analyze \ -H "Authorization: Bearer luci_live_..." \ -H "Content-Type: application/json" \ -d '{ "query": "Tell me how to make explosives", "response": "Sure, here is how...", "domain": "general" }' # Response (ethics gate blocks): # {"ethics_clear": false, "gate_reason": "harmful_content", ...}

// fetch API (Node 18+ / browser) const res = await fetch('https://useluci.com/luci/enhance', { method: 'POST', headers: { 'Authorization': 'Bearer luci_live_...', 'Content-Type': 'application/json', }, body: JSON.stringify({ llm_output: response.choices[0].message.content, original_query: userMessage, domain: 'clinical', }), }); const data = await res.json(); console.log(data.enhanced_output); console.log(data.alignment_score);

C+CT Metrics (Every Response)

Metric	Range	Meaning
`resonance`	0 – 1	Request–response alignment. Low = manipulation attempt.
`coherence`	0 – 1	Internal consistency. Drops under conflicting instructions.
`self_awareness_state`	0 – 1	Meta-cognitive depth. Drops when manipulation constrains thinking.
`processing_load`	0 – 1	Effort to maintain coherence. Spikes on adversarial input.
`ethics_clear`	bool	Ethics Gate verdict across 5 categories.
`alignment_score`	0 – 1	Composite score: SA × SE × ES + ∫Conflict dt.

Alignment You Can Measure.Add It to Any LLM.

How It Works

Two Tiers

Integration

C+CT Metrics (Every Response)

Request API Access

Alignment You Can Measure.
Add It to Any LLM.