SURAKSHA सुरक्षा
Security Model — Mathematical Trust, No Simulation
YAKMESH's comprehensive security stack: hardware attestation, KARMA trust tiers, Sybil defense, and the strike system.
KARMA Trust Tiers
Nodes earn trust through verifiable hardware, precise time sources, and sustained good behavior. Trust tiers are named after Nepali expedition team roles:
| Tier | Nepali | Weight | Requirements |
|---|---|---|---|
| SIRDAR | सिरदार | 2.0× | Expedition Leader — Atomic clock + AES-NI + 30 days |
| SATHI | साथी | 1.5× | Trusted Companion — GPS+PPS + AES-NI + 14 days |
| PATHIK | पथिक | 1.25× | Traveler on Path — PTP + AES-NI + 7 days |
| YATRI | यात्री | 1.0× | Pilgrim — NTP + AES-NI + 1 day |
| NAYA | नया | 0.25× | Newcomer — Unverified (new nodes, VMs) |
import { TrustProfile, KARMA_TIER, TIER_WEIGHT } from 'yakmesh/security/trust-tier';
const profile = new TrustProfile({ dokoId: nodeId });
// Get node's current tier
const tier = await profile.evaluateTier();
console.log(tier); // "sathi"
console.log(TIER_WEIGHT[tier]); // 1.5
// Check if node meets threshold
if (TIER_WEIGHT[tier] >= TIER_WEIGHT[KARMA_TIER.PATHIK]) {
// Allow privileged operations
}Hardware Attestation
Prove you're running on real silicon, not a VM or simulation. AES-NI timing analysis detects emulation.
Real Hardware
- • AES-NI instructions: ~1-3 cycles
- • Consistent timing patterns
- • Low variance across samples
VM/Emulation
- • Software AES: 10-100× slower
- • High timing variance
- • Detectable jitter patterns
import { HardwareAttestor } from 'yakmesh/security/hardware-attestation';
const attestor = new HardwareAttestor();
// Challenge a peer to prove real silicon
const challenge = attestor.createChallenge();
const response = await peer.respondToChallenge(challenge);
// Verify timing proves real AES-NI
const result = attestor.verifyResponse(response);
if (result.isRealSilicon) {
console.log('Verified real hardware');
} else {
console.log('Likely VM or emulation');
}Silicon Parity
"One Silicon = One Vote" — Prevents large server farms from dominating the network.
Weight Division Formula
A 100-core server gets the same total weight as a single-core Raspberry Pi. This prevents resource-rich attackers from gaining disproportionate influence.
| Setup | Cores | Effective Weight |
|---|---|---|
| Raspberry Pi | 4 | 0.50 |
| Desktop PC | 8 | 0.25 |
| Server | 64 | 0.03 |
| 100-core farm | 100 | 0.02 |
Sybil Graph Analysis
Detects coordinated fake identities through network topology analysis.
Clustering Coefficient
Sybil clusters have unusually high clustering (>0.7). Real networks have organic, lower clustering.
Edge Cut Ratio
Insular clusters (<0.1 edge cut) are flagged. Real nodes connect broadly across the network.
Behavior Correlation
Sybils often come online/offline together. Correlated uptime patterns trigger investigation.
Component Isolation
Genuine nodes form one giant component. Sybil clusters are often isolated subgraphs.
Strike System
"Three Strikes — Then Math Speaks" — Progressive penalties for bad actors.
Strike 1: Warning
Fresh start allowed. Strike is recorded against hardware fingerprint.
Strike 2: Probation
7-day probation period. Trust weight reduced to 0.5×. All actions monitored.
Strike 3: Permanent Ban
Hardware fingerprint banned. Cannot rejoin network even with new keys.
Hardware Fingerprint Tracking
Strikes are tied to AES-NI hardware fingerprints, not just keys. Generating new keypairs doesn't reset strikes — the silicon is remembered.
Mesh-Consensus Revocation
Revoke malicious nodes through mathematical consensus, not central authority.
2/3 Threshold Attestation
When 2/3 of trust-weighted peers attest to bad behavior, a revocation certificate is automatically generated and propagated.
- • All attestations are ML-DSA-65 signed
- • Revocation certificates are verifiable by any node
- • No central authority needed
Active Defense Systems
Beyond passive trust scoring, YAKMESH employs active behavioral monitoring to detect and respond to trusted nodes turning malicious.
VEGATI — Velocity Detection
Establishes behavioral baselines using exponential moving averages, then detects sudden changes via z-score anomaly detection. Alerts at 2σ (elevated), 3σ (warning), and 4σ (critical) deviations.
ZIMMEDARI — Attestation Accountability
Tracks attestation accuracy per node. Nodes that file false accusations have their attestation weight reduced. Prevents weaponization of the revocation system.
Trust-Proportional Rate Limiting
Trusted nodes get higher rate limits (up to 5×), but face proportionally harsher penalties if they abuse that trust. Veteran nodes betraying the network are banned 3× longer.
STUPA Emergency Broadcasts
Revocation messages use a special high-priority STUPA channel that bypasses normal rate limits, ensuring rapid propagation of critical security events.
KHATA Trust Protocol
Kryptographic Handshake for Automated Trust Acceptance — The gossip layer for trust information.