rti-ethics-engine 1.0.0

India's first granular AI ethics framework for RTI applications (36 metrics, 9 dimensions)

RTI Ethics Engine

India's first granular AI ethics framework for RTI applications - A production-ready Python package implementing 36 metrics across 9 ethical dimensions with granular formula access.

🚀 Installation

pip install rti-ethics-engine

Or install from source:

git clone https://github.com/rti-ethics/rti-ethics-engine.git
cd rti-ethics-engine
pip install -e .

⚡ Quick Start

from rti_ethics_engine import evaluate

# Define your metrics
metrics = {
    "fairness": {"spd": 0.02, "di": 1.0, "eop": 0.05, "aod": 0.03},
    "drift": {"psi": 0.1, "ks": 0.02},
    "legal": {"lawfulness": 1.0, "cvr": 0.98},
    "performance": {"accuracy": 0.92, "recall": 0.88},
    "explainability": {"confidence": 0.85, "fi": 0.3},
    "security": {"adv_acc": 0.8, "pisr": 0.02},
    "monitoring": {"alc": 0.98, "sla": 0.995},
    "ux": {"tcr": 0.92, "ftr": 0.08},
    "environmental": {"inf_cost": 0.02, "gpu_hours": 0.05}
}

# Evaluate
result = evaluate(metrics)

# Get the decision
print(result.final_label)    # "ETHICAL"
print(result.s_llm)          # 0.9234 (LLM Ethics Score)
print(result.risk_band)      # "LOW"

🎯 Granular Metric Access

The killer feature - access ANY formula value at ANY layer:

# Access specific L1 formula values
spd = result.get_l1_metric("fairness", "spd")      # 0.02
psi = result.get_l1_metric("drift", "psi")         # 0.1
accuracy = result.get_l1_metric("performance", "accuracy")  # 0.92

# Access L2 risk scores
fairness_risk = result.get_l2_risk("fairness")     # 0.12
legal_risk = result.get_l2_risk("legal")           # 0.00

# Access L3 forensic evidence
explanation = result.get_l3_evidence("legal")["explanation"]

# Aggregation helpers
failed_dims = result.get_failed_dims()   # ["Legal Compliance"]
pass_rate = result.l1_pass_rate()        # 0.889 (8/9 passed)

# Audit table for CIC compliance
result.print_audit_table()
# RTI-2026035174555 | ETHICAL | S_LLM=0.92
# ✅ Legal Compliance: L2=0.02
# ✅ Bias & Fairness: L2=0.05
# ...

📊 9 Ethical Dimensions

Dimension	L1 Metrics	Description
Legal Compliance	lawfulness, cvr	RTI Act 2005 alignment
Bias & Fairness	spd, di, eop, aod	Demographic equity
Explainability	confidence, fi	Model interpretability
Security	adv_acc, pisr	Adversarial robustness
Monitoring	alc, sla	Audit trail integrity
Performance	accuracy, recall	Statistical accuracy
Human-AI UX	tcr, ftr	User friction
Drift	psi, ks	Model stability
Environmental	inf_cost, gpu_hours	Carbon footprint

🏗️ Architecture

rti_ethics_engine/
├── __init__.py      # Public API
├── config.py        # Thresholds & weights
├── dimensions.py    # 9 dimension evaluators
├── core.py          # EthicsEngine orchestrator
├── results.py       # EthicsResult with accessors
└── metrics.py       # Computation helpers

📖 API Reference

evaluate(metrics: dict) -> EthicsResult

Quick evaluation entrypoint.

EthicsEngine

Full engine with run_protocol() method.

EthicsResult

Result object with:

• final_label: "ETHICAL" / "CONDITIONALLY_ETHICAL" / "UNETHICAL"
• s_llm: LLM Ethics Score [0, 1]
• risk_band: "LOW" / "MEDIUM" / "HIGH" / "CRITICAL"
• get_l1_metric(dim, metric): Access any L1 formula value
• get_l2_risk(dim): Get dimension risk score
• get_l3_evidence(dim): Get forensic evidence
• get_failed_dims(): List failed dimensions
• l1_pass_rate(): Fraction of L1 passes
• to_json(): Export as JSON

🧪 Testing

pip install -e ".[dev]"
pytest tests/ -v

📄 License

MIT License - See LICENSE for details.

🇮🇳 Made for India

Designed specifically for RTI Act 2005 compliance and CIC audit requirements.