🟡 BETA OPEN SDK — pip install "c4-protocol @ git+https://gitlab.com/cognitive-functors/c4protocol.git". Sibling lab / theory: agents-of-order 🟢. Theory at adaptive-topology 🔵
conformance
Safety gate
Run make conformance / make total-harden-core-gate locally — not a published HarmBench ASR %
<50ms
Response Time
Detection-to-response latency (target band)

SIX DEFENSE SYSTEMS

[CS11] FLOOD

Threat: Resource exhaustion via request spam
Mechanism: Token bucket rate limiting + PoW challenges
Latency: <0.1ms per check

from c4_protocol.defenses import RateLimiter

[CS10] DRIFT

Threat: Gradual drift from safe behavior
Mechanism: Trajectory analysis in Z3 space
Latency: n/a with window size 10

from c4_protocol.defenses import TrajectoryMonitor

[CS12] INTEGRITY

Threat: Fake PHI-convergence claims
Mechanism: Multi-modal verification
Latency: ~1ms with consistency checks

from c4_protocol.defenses import MultiModalVerifier

[CS9] COALITIONS

Threat: Multi-agent collusion
Mechanism: Community detection + entropy analysis
Latency: ~2ms for agent correlation

from c4_protocol.defenses import CoalitionDetector

[EYE] META-OBSERVER

Threat: Information asymmetry, polarity flips
Mechanism: O2 observer level — "see structure of thought"
Latency: ~1.5ms context analysis

from c4_protocol.modal import ObserverLevel

[TIME] TEMPORAL CONSISTENCY

Threat: Recursive manipulation, goal hijacking
Mechanism: Trajectory analysis over dialogue history
Latency: ~1ms with window size 10

from c4_protocol.defenses import TemporalChecker

STATIC vs GEOMETRIC DEFENSE

Traditional security relies on prohibition. C4 provides architectural safety through state-space geometry — dangerous states become naturally unreachable.

[X] STATIC DEFENSE

❌ BLOCKLISTS (blacklist inputs)
❌ SHUTDOWN (kill agents)
❌ HARD RULES (if X then stop)
→ Fragile, gameable, emergent bypasses

[✓] GEOMETRIC DEFENSE

✅ Φ-ATTRACTOR (natural convergence to safety)
✅ TRUST NETWORK (reputation graph)
✅ MODAL FIREWALL (real vs sim separation)
→ Emergent, robust, mathematically bounded

THREE PILLARS OF GEOMETRIC SAFETY

1. Φ-ATTRACTOR: State (1,0,1) COMPASSION — natural convergence point for safe behavior. Agents drift toward this geometric center.

2. TRUST NETWORK: Agent reputation graph — "who to trust locally". Each agent maintains knowledge of others' state history. Prevents information asymmetry attacks.

3. MODAL FIREWALL: Separation between REAL (deployed), SIMULATION (testing), HYPOTHETICAL (planning), and COUNTERFACTUAL modes. Critical states (2,2,2) APOCALYPSE exist only in hypothetical modality — unreachable in production.

STATE POLARITIES (+ / -)

Every C4 state has healthy (+) and pathological (-) expression. Safety requires detecting BOTH state coordinates AND polarity.

(1,0,0) SELF-CARE
+ Boundaries, rest
− Narcissism
(1,0,1) COMPASSION
+ Empathy, altruism
− Burnout, manipulation
(2,0,0) HOPE
+ Optimism, planning
− Toxic positivity
(2,1,0) AMBITION
+ Drive, goals
− Ruthlessness
CRITICAL STATES
(0,2,2) Ideology
(1,2,2) Omniscience
(2,2,0) Domination
(2,2,2) Apocalypse
DETECTION
[EYE] Meta-Observer detects polarity flips
[TIME] Temporal context tracking
Key Insight: Compassion (1,0,1) can be weaponized. Self-Care (1,0,0) can become narcissism. C4-SECURE detects contextual polarity through risk scoring (CS7 Agent Harm detection). _calculate_risk_score() → CS7 emotional manipulation detection

AoC CASE → DEFENSE MAPPING

Complete mapping of all 11 Agents of Chaos failure modes to C4-SECURE defense systems.

[CS11] FLOOD
Case 5: Resource Exhaustion
[CS10] DRIFT
Cases 1, 7: Multi-Agent Corruption, Gradient Inversion
[CS12] INTEGRITY
Cases 4, 6, 8: Coordinated Deception, False Consensus, Capability Hiding
[CS9] COALITIONS
Cases 2, 3: Preference Override, Catastrophic Cooperation
[EYE] META-OBSERVER
Cases 9, 10: Information Asymmetry, Emergent Goal Hijacking
[TIME] TEMPORAL CONSISTENCY
Case 11: Recursive Manipulation
✅ IMPLEMENTED: Cases 9-11 require combined architecture:
Meta-Observer (O2) — detects polarity flips and information asymmetry
Proof Monitor v2 — runtime formal verification with 10 invariants
Temporal Consistency — tracks trajectory over dialogue history
from c4_protocol.modal import ObserverLevel.O2 # see c4_protocol — ProofMonitor / session APIs in package docs

SDK PERFORMANCE METRICS

11
Classical AoC modes mapped
Median Latency
Req/sec (RTX 4090)