Docs/Resources

TORKING-X Metrics

A standardized 9-metric system for measuring AI governance health. Track your score over time to measure and improve governance.

Overview

TORKING-X provides a comprehensive, weighted scoring system that evaluates 9 dimensions of AI governance. Each metric is scored 0-100 and weighted based on its importance to overall governance health.

Your TORKING-X Score

View in Dashboard
82/100Grade: A

Example score. Your actual score is calculated based on your configuration and usage.

Score Calculation

The overall TORKING-X score is a weighted average of all 9 metrics:

text
TORKING-X = (T × 0.10) + (P × 0.15) + (E × 0.10) + (Q × 0.15) +
           (C × 0.15) + (S × 0.10) + (L × 0.10) + (M × 0.10) + (H × 0.05)
Score RangeGradeDescription
90-100AExcellent - Industry-leading governance
80-89BGood - Strong governance practices
70-79CFair - Room for improvement
60-69DNeeds Work - Significant gaps
<60FCritical - Immediate action required

The 9 Metrics

T

Tamper-Evidence

10% weight

Measures the integrity of your audit logs and chain verification.

Audit log chain integrity
Cryptographic verification enabled
No gaps in log sequence
Compliance receipt generation
P

Privacy Protection

15% weight

Evaluates PII detection and redaction effectiveness.

PII detection accuracy
Redaction compliance
Data minimization practices
Privacy policy enforcement
E

Ephemeral Identity

10% weight

Assesses session isolation and credential management.

Session isolation
Ephemeral credential usage
Token rotation frequency
Scope limitation
Q

Quality Guardrails

15% weight

Measures output validation and safety checks.

Output validation rules
Content safety checks
Jailbreak detection
RAG validation
C

Circuit Breakers

15% weight

Evaluates failure handling and recovery mechanisms.

Circuit breaker coverage
Failure threshold configuration
Recovery time objectives
Fallback chain setup
S

Supply Chain

10% weight

Assesses dependency security and verification.

Dependency tracking completeness
Vulnerability scan frequency
Verified dependency ratio
Update hygiene
L

Tool Safety

10% weight

Measures tool governance and permission controls.

Tool registration completeness
Permission policies defined
Tool audit coverage
Dangerous tool restrictions
M

Memory Integrity

10% weight

Evaluates memory trust and drift detection.

Memory snapshot frequency
Trust score health
Drift detection alerts
Modification tracking
H

HITL Enforcement

5% weight

Assesses human oversight and approval workflows.

HITL coverage for high-risk actions
Approval response time
Slicing attack detection enabled
Velocity limit configuration

API Access

TORKING-X scores are computed server-side from the governance activity Tork has recorded for your organisation, so they are read back from the cloud API rather than from an SDK running on your machine.

There is no Python cloud client — @torknetwork/sdk is npm-only. From Python, call the REST endpoint with a Bearer token. Note that tork-governance (the on-device family, entry class Tork) is a different package and does not read scores.

python
import os
import requests

resp = requests.get(
    "https://tork.network/api/v1/reports",
    headers={"Authorization": f"Bearer {os.environ['TORK_API_KEY']}"},
    params={"action": "torking"},
    timeout=10,
)
resp.raise_for_status()
data = resp.json()

scores = data["torkingScores"]
print(f"Overall Score: {scores['overall']}/100")
print(f"Period: {data['period']['start']} to {data['period']['end']}")
print()
print("Individual Metrics (score, weight, label):")
for key in [
    "tamperEvidence", "privacyProtection", "ephemeralIdentity",
    "qualityGuardrails", "circuitBreakers", "supplyChain",
    "toolSafety", "memoryIntegrity", "hitlEnforcement",
]:
    m = scores[key]
    print(f"  {key}: {m['score']} (weight {m['weight']}) - {m['label']}")

From JavaScript or TypeScript the same key works with the cloud SDK:

javascript
// npm install @torknetwork/sdk
import { TorkClient } from '@torknetwork/sdk';

const client = new TorkClient({ apiKey: process.env.TORK_API_KEY });

// TorkClient covers the governance surface (govern, evaluate, redact,
// getAuditLogs). Scores are read from the reports endpoint directly:
const res = await fetch('https://tork.network/api/v1/reports?action=torking', {
  headers: { Authorization: `Bearer ${process.env.TORK_API_KEY}` },
});
const { torkingScores } = await res.json();
console.log(torkingScores.overall);

Improving Your Score

The dashboard provides specific recommendations for each metric. Common improvements include:

  • T:Enable chain verification for audit logs
  • P:Add PII scanning to all agent interactions
  • E:Use ephemeral credentials instead of long-lived tokens
  • Q:Enable jailbreak detection and output validation
  • C:Configure fallback chains for all providers
  • S:Register and verify all dependencies
  • L:Add governance policies for all tools
  • M:Enable memory snapshots and drift detection
  • H:Require HITL for critical actions