HIP-4 MM Simulator

Experimental, volume-conserving execution simulation for Hyperliquid HIP-4 market makers.

Existing HIP-4 SDKs solve market access, while general trading engines solve broad backtesting. HIP-4 MM Simulator focuses on transparent, empirically validated execution simulation for market makers.

This is an alpha research tool, not a production trading system and not a claim that the included baseline strategy is profitable.

Open the 60-second execution demo to see an observed L2 queue, aggressor trade, queue consumption, and partial fill replayed through the Python engine.

What v0.2 validates

• Dynamic outcomeMeta discovery, including custom side labels, multi-outcome questions, and per-outcome quote tokens.
• An explicit HIP-4 coin: there is no stale default outcome ID.
• Queue-ahead seeded from the latest L2 snapshot.
• Aggressor-aware matching: buys consume asks and sells consume bids.
• Price-time priority with a shared trade-volume budget.
• Partial fills and sum(fill_size) <= observed_trade.size.
• Duplicate exchange trade IDs ignored.
• Spot-safe quote/token balances, reservations, and no naked sells.
• Versioned JSONL recording and deterministic replay reports.

Book-size decreases without a trade are conservatively treated as possible cancellations. They do not move a virtual order forward in queue.

Product boundary

Tool category	Primary job	HIP-4 MM Simulator's relationship
Hyperliquid/HIP-4 SDKs	Discovery, signing, orders, settlement operations	Complementary: consumes market data but does not replace an SDK
NautilusTrader and general engines	Broad live/backtest infrastructure	Narrower: focuses on inspectable HIP-4 execution assumptions
Historical data services	Store and serve exchange data	Recorder is self-contained; external data can be converted to JSONL
This project	Paper execution, queue assumptions, replay invariants	Not a live execution stack or profitability oracle

Install

pip install "hip4-mm-simulator[live]==0.2.0"

For repository development:

git clone https://github.com/horn111/hip4-mm-simulator.git
cd hip4-mm-simulator
pip install -e ".[live]"
pip install pytest pytest-cov ruff mypy

CLI workflow

Discover current markets rather than copying an old outcome ID:

hip4-sim markets
hip4-sim markets --testnet --json

Record one explicit token:

hip4-sim record --coin "#8050" --duration 24h --output data/mainnet-8050.jsonl

Replay it through the included validation baseline:

hip4-sim replay data/mainnet-8050.jsonl --output validation-report.md

The command exits non-zero when a balance, reservation, or trade-volume invariant fails. A small deterministic fixture is included:

hip4-sim replay tests/fixtures/sample_recording.jsonl --output sample-report.md

Python API

from decimal import Decimal

from hl_paper_trading import MatchingEngine, Side, VirtualOMS, VirtualWallet

coin = "#8050"  # obtain from HyperliquidInfo or `hip4-sim markets`
wallet = VirtualWallet(
    quote_balances={"USDC": Decimal("10000")},
    token_balances={coin: Decimal("1000")},
    token_quotes={coin: "USDC"},
    initial_mark_prices={coin: Decimal("0.50")},
)
engine = MatchingEngine(coin=coin, quote_token="USDC")
oms = VirtualOMS(wallet=wallet, engine=engine)

# Feed each book before strategy decisions.
oms.process_book(snapshot)
order_id = oms.submit_order_sync(
    Side.BUY,
    Decimal("0.49"),
    Decimal("10"),
    current_time=snapshot.timestamp,
)

# Observed trades activate latency-expired orders and generate spot-safe fills.
fills = oms.process_trade(trade)

An order is rejected with BOOK_STALE if no L2 snapshot exists or if the latest book is more than five seconds older than activation. Its reservation is released automatically.

Recording schema

Each JSONL row contains:

{
  "schema_version": "1",
  "event_type": "metadata | book | trade",
  "exchange_timestamp": "ISO-8601",
  "received_timestamp": "ISO-8601",
  "coin": "#8050",
  "payload": {}
}

The report records the source SHA-256, exact time window, event/gap counts, orders, rejections, fills, queue volume, duplicate trades, final NAV, PnL, invariants, and model limitations.

Development checks

ruff check .
mypy src
pytest
python -m build

CI runs the same checks on Python 3.11 and 3.12. Test coverage must remain at or above 85%.

Explicitly outside v0.2

• Split, merge, negate, and settlement transitions.
• Fee and builder-code accounting.
• Calibration against private order-level ground truth.
• Multi-market portfolio simulation, dashboards, and live order submission.

Those are candidates for v0.3 after the execution assumptions receive external review. See the validation protocol and the v0.1 migration guide. Publication status is tracked in the v0.2 release checklist.

License

MIT. Contributions that challenge a fill assumption with a reproducible trace are especially welcome.