From discretionary to systematic on Hyperliquid

Why systematize

The case for going systematic on Hyperliquid is not that discretion is wrong. The case is that discretion does not scale past a single screen and a single brain. You can hold three or four positions in your head with intent, you can react to a funding flip on BTC and ETH in real time, you can square up before a CPI print. You cannot do that for thirty perps across four overlapping signals at four time horizons while sleeping.

The cognitive load of perp tracking is the real ceiling. Discretionary traders who hit a plateau on HL almost always hit it for the same reason: their edge is real but their attention is the bottleneck. Systematizing is how you take an edge that currently runs at 2x leverage on three names and turn it into an edge that runs at the same risk on a thirty-name book with no vigilance tax. Consistency at scale, not magic.

The lifecycle

A systematic strategy moves through seven stages on Keel. Each stage is a surface you actually use:

• Research. State the edge in one sentence. Pick the components that encode it. Evaluate the signal with IC and decay diagnostics before you commit to a backtest.
• Backtest. Run the Pipeline against the HL universe with real fees, real 1-hour funding, and the same portfolio simulator the live worker uses.
• Validate. Look at the funding-decomposed equity curve. Check turnover. Sanity-check the worst drawdown and the regime where it happened. Decide whether the result is robust or curve-fit.
• Paper. Run the live executor with paper-mode fills against the production HL state. The signal computes live; the orders log without firing. This is where you catch staleness, latency, and timing bugs the backtest cannot show you.
• Live. Flip the executor to real. Same pipeline, same buffered rebalancer, same vol target. Start with smaller size than your backtest assumes — degradation from research to live is real.
• Monitor. Dashboard equity curve next to the backtest equity curve. PnL attribution by signal and regime. Alerts on drawdown breach or turnover anomaly.
• Iterate. Edit the pipeline. Re-backtest under the new config. Promote when validation passes. Edge does not stay still on crypto — neither should the strategy.

Backtest-to-live parity

The single most-important design property: the Pipeline object you backtest is the Pipeline object that runs live. Same data loaders, same signal components, same aggregators, same vol target, same buffered rebalancer with the same drift band. The Keel backtest engine hydrates the pipeline from cached data; live execution hydrates the same pipeline from live HL state and routes the resulting target weights through the broker.

The buffered rebalancer matters specifically because it is where most platforms quietly diverge. A backtester that rebalances every bar but a live executor that batches and throttles will show a backtest result you cannot reproduce. On Keel the drift band you choose in the backtest is the drift band the broker enforces; the only delta between research and live is the realized fill price versus the modeled slippage. You can audit the delta from the dashboard.

What changes from discretionary

Going systematic is not just “the bot does it instead.” The things you are responsible for change:

• Sizing is rules, not feel. You set a target portfolio vol and a per-asset cap once. The aggregator and vol-targeting overlay handle the rest. You stop deciding position size per trade and start deciding it per regime.
• Regime detection is explicit. Discretionary traders know “funding is rich right now” in their gut. Systematic strategies encode that in a regime detector — FundingLevelRegime, FundingDispersionRegime, vol regime — and the carry signal scales by it.
• Execution is buffered. The rebalancer fires only when actual drift exceeds your band, rather than reacting to intra-bar order-book noise. Turnover drops, fees drop, slippage drops.
• Manual overrides break parity. The strategy runs the strategy. You can pause it before a known event; you cannot quietly resize a position mid-run without diverging from the backtest the audit trail is measured against. The discipline is the point.

First strategy templates

Three patterns cover most first systematic strategies on HL. Each composes from shipped components — no custom signal code required to start:

• Funding-carry. Long the highest-funding payers, short the lowest. Gated by FundingDispersionRegime so the book only deploys when the cross-sectional spread is wide. See the worked example at /strategies/funding-carry in the lab.
• Cross-sectional momentum. Rank the universe by trailing 30-day return, long the top quintile, short the bottom quintile, rebalance daily with a 10% buffer. Add the vol-targeting overlay to hold portfolio vol constant across regimes.
• Vol-targeted basket. Equal-weight or inverse-vol the top-N HL perps by volume. Apply a vol target of 20% annualized. Simple, hard to overfit, and a useful benchmark to size every other strategy against.

All three live as compositions of the same Pipeline primitive. Start with one, get to live, then layer the next as an additional signal in the same pipeline. The aggregator handles combining multiple forecasts into a single set of target weights.