Skip to content

README.md

Latest commit

 

History

History
154 lines (118 loc) · 6.39 KB

README.md

File metadata and controls

154 lines (118 loc) · 6.39 KB

ARES-E Genesis — v0.2.0

Agentic Resilience & Evaluation System for Essential-Infrastructure DOE Genesis Mission — Phase I Prototype — DE-FOA-0003612

Overview

ARES-E is a deterministic, physics-constrained evaluation harness for scientific AI agents operating within the American Science Cloud (AmSC). It validates agent-proposed actions against first-principles physics, enforces differential-privacy guarantees on infrastructure telemetry, and produces a cryptographically-chained (SHA-256) audit trail exported as STIX/TAXII 2.1 bundles for DOE threat-intelligence compliance.

Topics Addressed

Topic Code-name Engine Description
16 EWIS ewis_grid.py Physics-Informed Neural Network for grid dispatch under Dynamic Line Rating
21 WOIK woik_fluid.py Closed-loop thermal-hydraulic energy conservation via NetworkX graph
20 PHIAK phiak_cyber.py ε-Differential privacy, adversarial/injection detection, SHA-256 fingerprinting

Architecture

app/
├── main.py                        FastAPI application with CORS & OpenAPI
├── api/
│   └── endpoints.py               Prefix-routed REST endpoints (/api/v1/genesis/*)
├── core/
│   ├── amsc_harness.py            GenesisHarness orchestrator — VVUQ + ledger
│   ├── cyber_ledger.py            ZeroTrustLedger — SHA-256 hash chain + STIX export
│   └── interoperability.py        Plug-and-play engine registry for multi-agency dispatch
├── engines/
│   ├── ewis_grid.py               GridPINN + deterministic dispatch + AI-Advantage
│   ├── woik_fluid.py              First-law ΔT evaluation + violation detection
│   └── phiak_cyber.py             Laplace mechanism + adversarial detection
└── schemas/
    └── genesis_payloads.py        Pydantic V2 strict FAIR-compliant models

notebooks/                         Jupyter demonstrations per topic
tests/                             45 pytest unit & integration tests
docker-compose.yml                 Uvicorn + Ray head for local AmSC simulation

Getting Started

git clone https://github.com/DaScient/Genesis.git
cd Genesis
pip install -e .           # or: poetry install
PYTHONPATH=. pytest -v      # 45 tests, zero warnings
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

API Endpoints

Method Path Description
POST /api/v1/genesis/submit_workflow Submit a scientific workflow for evaluation
GET /api/v1/genesis/status/{job_id} Retrieve progress, violations, VVUQ metrics
GET /api/v1/genesis/ledger Immutable STIX/TAXII 2.1 audit bundle
GET /api/v1/genesis/domains List registered interoperability domains
GET /api/v1/genesis/health Readiness probe with ledger integrity check
GET / Root liveness probe

Example Payload

{
  "workflow_id": "wf-001",
  "domain": "EWIS",
  "agent_id": "genesis-agent",
  "timestamp": "2026-04-02T00:00:00Z",
  "metadata": {
    "source": "sensors/grid/powerflow",
    "modality": "timeseries",
    "timestamp": "2026-04-02T00:00:00Z",
    "uncertainty_margin": 0.05
  },
  "parameters": {
    "load_mw": 60.0,
    "capacity_mw": 100.0,
    "weather_anomaly": 0.2
  }
}

Interoperability

The InteroperabilityManager provides a plug-and-play engine registry:

harness.interop.register_engine("NIST", my_nist_engine)
harness.interop.register_engine("NNSA", my_nnsa_engine)
harness.interop.get_registered_domains()  # ['EWIS', 'NNSA', 'NIST', 'PHIAK', 'WOIK']

Any callable with the signature (payload: Dict[str, Any]) -> Dict[str, Any] that returns physics_violations and vvuq_score keys can be registered for cross-agency dispatch.

DOE & LCF Compliance

  • Python 3.11+ strict typing with Pydantic V2 (ConfigDict, model_validator)
  • Deterministic VVUQ — physics validation layer runs independently of AI predictions
  • Zero-Trust audit — SHA-256 hash-chained ledger with verify_chain() integrity verification
  • STIX/TAXII 2.1 — audit trail exportable as JSON bundles for DOE threat-intelligence
  • Air-gapped ready — zero external API calls, no cloud-vendor dependencies
  • FAIR data — every payload requires Metadata with Source, Modality, Timestamp, Uncertainty_Margin

Testing

PYTHONPATH=. pytest -v --tb=short   # 45 tests across 3 test modules

Test coverage spans:

  • Physics engines — dispatch, PINN, baseline, AI-advantage, thermal hydraulics, privacy, adversarial detection
  • Harness orchestration — all three domains, failures, ledger growth, chain verification, interoperability
  • API integration — all endpoints, validation errors, 404 handling

Docker Compose

docker-compose up --build

Starts Uvicorn service on port 8000 with Ray head node for mock HPC compute.

Documentation

Enterprise documentation is organized under docs/ with a central index.

Submission Artifacts

Document Description
Technical Specification Architecture, scope, FOA alignment, capability summary
Milestone Delivery Matrix OT milestones mapped to deliverables with acceptance criteria
VVUQ Framework Verification, Validation & Uncertainty Quantification methodology
Interoperability Specification Engine interface contract, registration lifecycle, dispatch protocol

Governance & Compliance

Document Description
Security & Compliance Posture Threat model, NIST 800-53/171, ODNI ICD 503, Navy CYBERSAFE, DOE Orders
Data Rights & IP Guide Patent rights, data rights, DUA compliance, marking guidance
Responsible AI Plan NIST AI RMF 1.0, model/data cards, bias assessment, human oversight

Training & Operations

Document Description
Onboarding Guide Developer quickstart, API walkthrough, engine extension tutorial
Deployment Runbook Production deployment, monitoring, incident response, maintenance

License

MIT © 2025 DaScient Corporation