Aletheia

Crypto perpetual market-making framework. Quotes both sides of the book on BTC/ETH options, managing inventory risk with closed-form quoting models (Avellaneda-Stoikov, GLFT) and regime/toxicity-aware gating, backtested against real exchange data.

Layout

aletheia/
├── data/           ← exchange-agnostic OrderBookUpdate type
├── exchanges/       ← live L2 order-book connectors (Deribit, Binance)
├── deribit/         ← Deribit REST client (historical trades, instruments)
├── examples/        ← runnable demos (live streaming, live quoting)
├── core/            ← PRIVATE submodule: quoting models, risk, backtest engines
├── research/        ← PRIVATE submodule: notebooks, validation scripts, real-data demos
├── tests/           ← tests for the public data/exchanges layer
└── utils/           ← dependency-free helpers (logger)

core/ and research/ are separate private GitHub repos (aletheia-core, aletheia-research) linked as git submodules. Everything else in this repo is public.

Model architecture

The strategy is layered; each layer is a separate concern with its own module(s), not a monolith:

Layer	Module(s)	Role
Volatility	core/models/volatility.py	EWMA realised vol feeding every downstream layer
Logic (quoting)	core/models/quoting.py, core/models/glft.py, core/strategies/	Avellaneda-Stoikov / GLFT reservation price + spread, ladder quoting
Regime / toxicity	core/models/regime.py, core/models/toxicity.py, core/models/forward_risk_net.py	HMM regime detection, VPIN, a trained nonlinear forward-risk model -- gates gamma and pausing
Inventory	core/risk/exposure.py	Position tracking, realised/unrealised P&L, funding accrual
Risk (static)	core/risk/limits.py	Hard position/notional/order-size/daily-loss caps
Tail risk (dynamic)	core/risk/tail_risk.py	Market-condition circuit breakers: price jump, feed staleness, vol-spike-relative-to-regime -- independent of current inventory
Greeks	(not yet built)	Deferred: relevant only if/when an options overlay is added (see project history for the 0DTE options discussion)

See core/MODEL.md for the full mathematical specification of the quoting models.

Data pipeline

exchanges/ streams live L2 order books from Deribit and Binance concurrently, for any mix of coins and market types (perpetual, spot) -- examples/stream_multi_exchange_books.py demonstrates six concurrent streams (BTC/ETH x {Deribit perp, Binance perp, Binance spot}) merged into one feed via exchanges/stream_manager.py. This is the hedging-instrument set: Deribit's perpetual is the primary quoting venue, Binance's spot and perpetual markets are cross-venue references and hedge legs.

core/backtest/historical.py replays real trade tapes (and, for Binance, real order-book depth) through any quoting engine -- see research/historical_backtest_demo.py and research/binance_ladder_backtest_demo.py for real-BTC/ETH walk-forward comparisons.

Quick start

pip install -r requirements.txt
python examples/stream_multi_exchange_books.py 20   # live multi-exchange book stream
python examples/live_quote_loop.py 30                # live AS quoting, dry-run
pytest tests/ core/tests/                             # 93 tests, public + private layers

Safety

Everything here is dry-run: live examples print quotes, they place no orders. Order execution is not implemented yet.