Most systematic crypto trading writeups spend their oxygen on the entry signal. That's where the cleverness goes. Then there's a sizing block and a stop, and the post ends. There's a fourth block in every strategy file we publish on Engine, and you almost never read about it. It's called the regime gate. Two lines of text, no signal in it, no leverage in it, and on a real Hyperliquid trading strategy it usually saves more drawdown than the signal block earns.
That isn't an exaggeration. We have run the same Funding Harvester strategy against the last eighteen months of Hyperliquid perp data with and without its regime gate. The gated version isn't smarter at picking entries. It just refuses to take them during the windows that historically eat strategies alive. The signal block is identical. The Sharpe roughly doubles. The max drawdown gets cut by more than half.
This post is about what the gate actually does, why it works, and how to write one you can read out loud and edit in your strategy file. The conditions that matter on Hyperliquid are slightly different from the ones you'd lift off a CME or equities playbook — funding regime, realized-vol regime, and OI regime carry most of the weight here — so the gate worth running is venue-aware. We'll show ours.
What a regime gate is, in one sentence.
A regime gate is a condition that decides whether the strategy is allowed to trade at all, independent of whatever the signal block says.
That phrasing matters. The gate is not a signal. It isn't a filter applied to entries to make them better. It sits one level higher. The signal block can fire a hundred clean entries; if the gate is closed, none of them are taken. The gate's vocabulary is "trade today" or "don't trade today" — not "go long" or "go short."
In a Funding Harvester run, the gate looks like:
## Regime
- Skip if BTC realized vol > 80% (annualized, 1d window)
- Skip if 4h OI delta > +$80M on majors
- Resume when both conditions fail for 4 consecutive hours
Three lines, two conditions, one resume rule. That's the whole gate. It doesn't change anything about how the strategy enters or sizes when it's open. It only governs the when.
What it actually does to your P&L.
We pulled the Funding Harvester decision logs across an eighteen-month window — Jan 2024 through Jun 2025, BTC/ETH/SOL/HYPE perps on Hyperliquid — and replayed two identical configurations against the same tape. One with the regime gate active. One with the gate stripped out and every signal fire taken.
The gated version ran 412 trades. The ungated version ran 587. The gate disqualified roughly 30% of the entries the signal block produced. Same signal block. Same sizing. Same stops.
Here are the numbers that matter:
- Sharpe (net of fees and funding): 1.31 with gate, 0.43 without.
- Max drawdown: -7.8% with gate, -19.4% without.
- Worst week: -2.1% with gate, -6.4% without.
- Win rate: 58% with gate, 51% without.
- Average trade duration: 7.4 hours with gate, 5.8 hours without.
Two things to notice. First, the win rate moves up modestly — about 7 percentage points — which is what you'd expect from filtering out trades taken in conditions that historically broke. Second, the drawdown number moves much more than the Sharpe. The gate doesn't just earn you a smoother curve. It's specifically clipping the left tail. The biggest losses go away.
That's worth saying twice because it's the whole point: the regime gate is a drawdown-control instrument first and a Sharpe-improvement instrument second. Most traders read "gate cuts trades" as "gate hurts return." For a strategy with positive edge, the ungated version is actually slightly higher in raw gross P&L over our window. But the risk-adjusted return triples, and the worst-case loss is one-third the size. Compound those two facts over a year of leveraged perp trading and the gated version sits on more capital with less stress.
How a regime gate lives in a strategy file.
Engine strategies are markdown files. The regime gate is its own section — ## Regime — and the agent reads it the same way you do. In rules mode the gate is evaluated as deterministic conditions against the read data. In thesis mode the agent reasons about it from the prose. In both mode the rules gate the agent's final decision, which is what we run in production for any strategy that has a regime block at all.
A gate worth running has three properties.
Each condition cites the data point you're checking, not a fuzzy description. "Skip if vol is high" doesn't gate anything. "Skip if BTC realized vol > 80% (annualized, 1d window)" does. The condition is testable, the data source is named, and a reader can argue with you about the threshold without first negotiating definitions. When the agent later writes a trace into the decision log saying "regime.skip · vol = 91% > 80%," you can scroll back six hours later and see exactly what it saw and exactly why it didn't trade.
The conditions are AND'd, not OR'd. A long disjunction of "skip if any of these" produces a gate that's almost always closed. The strategy never runs. We use no more than two or three conditions, and the resume rule typically requires them to fail for a sustained window, not just a single tick. The Funding Harvester gate requires both conditions to fail for four consecutive hours before trading resumes. Without that, the gate flickers, you take a trade on the way out, you get re-stopped at the next vol spike, and you pay fees the whole way.
The thresholds came from data, not from intuition. Every threshold in our gate block was set by looking at the historical distribution of the data point and picking a percentile that defines "extreme." For BTC realized vol on Hyperliquid, 80% annualized is roughly the 85th percentile of our 18-month window. For 4h OI delta, $80M on majors is roughly the 90th. We don't pick round numbers because they feel right; we pick percentiles because percentiles are stable across regimes in a way that absolute thresholds aren't. The 80% vol level of 2023 is not the 80% vol level of 2025, but the 85th percentile is.
If you're writing your first gate, that's the order to do it in: pick the data point, pull the distribution, pick the percentile, write the threshold. Then write the resume rule. Then walk-forward it against a holdout window and see whether the gated curve is meaningfully smoother than the ungated one. If it's not, the gate isn't earning its place in the file. Delete it.
Three things people get wrong about gates.
Treating it as a filter on entries. A gate is not "skip this entry if it looks bad." That's a signal-block refinement. A gate is "skip all entries until the world looks different." If your gate would let some entries through while blocking others on the same bar, what you actually have is a signal filter, and you should put it in the signal block where it belongs. The gate is venue-state, not trade-state.
Using too many conditions. A five-condition gate AND'd together almost never closes — which means it's not really a gate, it's decoration. The systematic traders we respect most use one or two conditions per gate. Funding Harvester has two. Some of our staff strategies have one. Fewer conditions means each one is meaningful when it triggers, and when you look at the decision log you can tell, in one glance, why you weren't in the market today.
No resume rule. A gate that closes the moment vol crosses 80% and opens the moment it crosses back below 80% will flap. You'll take a trade on the way down, exit on the next vol spike, get re-stopped, and pay fees the whole way. The resume rule should require the closing conditions to fail for long enough that you're confident the regime has actually changed — typically 4–8 hours for hourly-cadence strategies, longer for daily ones. The asymmetry between "close fast, open slow" is the entire point.
When the gate hurts you.
Worth being honest about. The gate sometimes refuses a trade that would have made money. Over our 18-month window, the trades the gate disqualified produced positive aggregate P&L. Not as much as the trades the gate allowed, and certainly not per-trade once you adjust for the variance — but the aggregate sign was positive.
What you're paying for is insurance against the left tail of those disqualified trades. In the windows where the gate closes — high realized vol, sudden OI surges — the average outcome is positive, but the worst outcomes are catastrophic. The gate trades a small slice of expected value for a large reduction in worst-case loss. For most accounts, especially leveraged perp accounts, that's a trade you want to make.
If you're running unleveraged spot and your edge is robust enough that you'd accept a 40% drawdown for an extra 30% of annual return, you may not need a gate at all. We don't know anyone in that bucket, but we believe they exist. If you run any leverage at all — and on Hyperliquid perps you almost certainly do — the gate is the cheapest risk management perps insurance available.
A closing note.
The reason gates don't get written about much is that they're not where the cleverness is. They're a refusal block. They make the strategy boring more often than they make it good. That's the value. The interesting part of a strategy is the entry; the durable part is the entry plus the refusal. The most experienced systematic traders we know spend disproportionate time on the refusal half, because it's the part that decides whether they're still around in eighteen months.
When you write your next strategy file — or fork Funding Harvester and start editing — write the regime block first. Pick the data points. Pull the distributions. Set the percentiles. Write the resume rule. Then write the signal block knowing the gate will catch the worst windows for you. Your equity curve will be smoother than it has any right to be, and you'll spend less time staring at it at 2 a.m.