Aetherius Risk Intelligence

Deterministic, backtest-validated downside-risk screening on primary-source financial news and regulatory filings.

---

What is Aetherius?

Aetherius is an open-source risk-screening engine built for concentrated public-equity books, search funds, litigation-finance analysts, and credit desks.

Instead of relying on black-box LLM prompts that hallucinate or non-deterministic sentiment models, Aetherius uses a purely deterministic, auditable pipeline:

1. Primary-Source Ingestion — Scrapes real-time SEC EDGAR filings (8-K, 10-Q, 10-K, NT 10-K) and global news wires via GDELT DOC 2.0.
2. Word-Boundary Entity Resolution — Strict regex matching with acronym stoplists to prevent false positives (e.g. distinguishing the ticker AI from words like CHAIN or SAID).
3. Deterministic 7-Factor Severity Scoring — Computes risk and urgency based on fixed compile-time taxonomies. Every single score is transparent, verifiable, and explainable in court or an investment committee.
4. Automated Stress-Test Reports — Assembles 8–15 page per-target diligence decks detailing counterparty exposure, filing timelines, and macro sensitivities.

---

The Backtest Benchmark

Aetherius was validated by replaying 161 real historical news observations across three major modern crises on frozen windows using the GDELT archive:

Event	Mechanism	Watchlist	Recall	Median Lead Time	False Positives on Controls
SVB-2023	Regional Bank Contagion (Mar 6–12)	5 affected + MSFT (control)	5 / 5 (100%)	2.34 days	0 / 1
Wirecard-2020	Accounting Fraud & Insolvency (Jun 15–30)	1 affected + DTE (control)	1 / 1 (100%)	6.60 days	0 / 1
FTX-2022	Counterparty Contagion (Nov 2–14)	3 affected + MSFT (control)	3 / 3 (100%)	7.12 days	0 / 1

All fixtures, watchlists, ground-truth records, and replay harnesses are committed and bit-identical across runs. Read the full working paper: docs/working_paper/detection_timing_backtest_2026-07.md.

---

Architecture

┌──────────────────────────────────────────────────────────┐
│                   Primary Ingestion                      │
│   • EDGAR Adapter (SEC 8-K, 10-Q, 10-K, NT filings)       │
│   • GDELT DOC 2.0 Adapter (Financial Whitelist Filter)   │
└────────────────────────────┬─────────────────────────────┘
                             │
                             ▼
┌──────────────────────────────────────────────────────────┐
│                Entity Mapping & Filtering                │
│   • Word-boundary regex matching                         │
│   • Multi-class acronym stoplist (CEO, AI, FED, etc.)    │
│   • Declared alias & counterparty relationship links     │
└────────────────────────────┬─────────────────────────────┘
                             │
                             ▼
┌──────────────────────────────────────────────────────────┐
│              Deterministic Severity Scoring              │
│   • 7-factor weighted scoring formula (compile-time)     │
│   • Adverse-language taxonomy gate (30+ downside terms)  │
│   • Zero LLM non-determinism in flagging logic          │
└────────────────────────────┬─────────────────────────────┘
                             │
                             ▼
┌──────────────────────────────────────────────────────────┐
│                  Delivery & Reporting                    │
│   • 8-15 Page Target Stress-Test Deck generation (HTML)  │
│   • Delivery quality gates (banned language / disclaimers│
└──────────────────────────────────────────────────────────┘

---

Quick Start

1. Install Dependencies

git clone https://github.com/zariffromlatif/Aetherius.git
cd Aetherius
pip install -r requirements.txt

2. Run the Test Suite

pytest aetherius/tests -v

3. Replay a Historical Crisis Backtest

Replay the SVB regional-banking crisis through the production scoring and mapping engine:

python simulations/backtest/run_backtest.py --event svb-2023

(You can also run --event wirecard-2020 or --event ftx-2022).

4. Generate a Stress-Test Diligence Deck (Offline Fixture Mode)

python scripts/build_deck.py \
  --ticker SIVB \
  --name "SVB Financial Group" \
  --sector "Regional Banks" \
  --aliases "Silicon Valley Bank,SVB" \
  --thesis "Concentrated regional-bank position with rate-sensitive HTM book." \
  --counterparty "FRC:First Republic Bank:peer:0.6:First Republic" \
  --window 2023-03-06:2023-03-12 \
  --fixture-jsonl simulations/backtest/events/svb-2023/observations.jsonl \
  --out sivb-deck.html

Open sivb-deck.html in your browser and select Print → Save as PDF for a publication-ready diligence brief.

---

Project Structure

aetherius/
  app/
    services/
      ingestion/         # SEC EDGAR and GDELT API adapters
      entity_mapping/    # Regex and acronym disambiguation
      scoring/           # Deterministic risk & urgency formulas
      signals/           # Downside signal taxonomy
      delivery/          # PDF rendering & regulatory quality gates
  tests/                 # 76-test unit & integration test suite
simulations/
  backtest/
    events/              # Frozen crisis fixtures (SVB, Wirecard, FTX)
    run_backtest.py      # Backtest replay harness
    build_fixture.py     # GDELT fixture builder
scripts/
  build_deck.py          # CLI to compile Target Stress-Test Decks
docs/
  working_paper/         # Full academic write-up and methodology

---

Contributing

We welcome contributions! Please check CONTRIBUTING.md for workflow details. All PRs must:

• Maintain deterministic scoring behavior.
• Include unit/integration tests (pytest passing 100%).
• Ensure historical crisis fixtures remain bit-identical.

---

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

---

Disclaimer

Aetherius Risk Intelligence is an open-source research and decision-support tool. It does not provide investment advice, fiduciary services, or guaranteed return forecasts. All analyses are strictly for informational and quantitative research purposes.