Risk-first execution
We don't promise returns.We prove control.
Eleven strategies entered our lab. One survived. HyperCore runs it at 1× with the limits, losses, and kill path visible before a cent moves.
No live track record yet. The mainnet record starts August 2026 and stays public, including every drawdown.
Ending book
$225,932
Worst drawdown
−42.48%
Most strategies lose. Here are ours that did.
A returns promise is the tell. What we can show you instead is the graveyard: every strategy we built, tested and killed, with the number that killed it.
one mark per strategy tested · short and clay = retired
Anyone can show you a winner. The ten failures are the useful part.
Kept · the one we run
CDC ActionZone
daily EMA trend, long-only
- Retired: Mirrorcopy the leader's resting orders−$12.08 / 148 fills
- Retired: Fadefade the leader's skew−$0.18 / 27 fills
- Retired: Makerquote passively for the spread+0.74 bps, noise
- Retired: Taker-copyfollow a market-maker's fillsnegative, 53 of 53
- Retired: Method Ddirection-follow + fee-aware exitnegative, 16 of 16
- Retired: Method Mleader-skewed passive quoting0 of 36 positive
- Retired: Edge compositestack several weak signalsPF 1.06
- Retired: Funding reversalfade funding extremesPF 0.80
- Retired: Cascade fadefade liquidation cascadesPF 0.43
- Retired: Contrarian-macro5-layer macro/sentiment fade7 winners in 33
Several were killed by a rule we wrote before running them: a strategy that only works in one walk-forward block is a market regime, not an edge. The contrarian-macro fade looked profitable until a review found accounting errors in its own harness. We fixed the harness, re-ran, and retired it.
The one that survived, and what it actually did.
CDC ActionZone is a long-only daily trend rule: two moving averages, no leverage, no shorts. It is not our invention and not a secret. We sell consistent execution with the losses published beside the wins—not a hidden signal.
8 coins · 2,398 days · 1× · perpetuals, net of fees and funding
derivedAnnual return
longer is betterWorst drawdown
shorter is better- Sharpe
- 1.56
- Calmar
- 1.85
- Beta to market
- 0.38
- Annualised alpha
- 39.14% · t≈3.2
The point is not the return. It is the shape. Beta 0.38 to an always-long basket and roughly half the drawdown, because the rule sits in cash through bear markets instead of riding them down. It is in the market 44.8% of the time, and a 5,000-dollar book ends the window at $225,932.
Read this before you read the numbers
- 01
This is a backtest, not a track record.
A simulation over historical candles. No money was at risk in it. The live mainnet record starts August 2026 and will be published here whether it flatters us or not.
- 02
A 42% drawdown is a real experience.
On $1,000 that is watching $580 and waiting. Most people sell there. If you would, this is not for you. The strategy only works if you don't.
- 03
Leverage liquidates this strategy.
We tested it: at 3× and above our own sweep was liquidated to zero in the May 2021 crash. That is why the engine runs at 1×.
- 04
We published the worse of two honest numbers.
This run charges perpetual funding. On spot (the venue we intend to use), the same strategy pays none and scores around 92% CAGR. Choosing the flattering configuration is the trick this page exists to argue against.
- 05
Past results do not predict future ones.
Six years is one sample of one regime, and 2020–2026 was a historic bull market, so our number and the benchmark are both inflated by it. The edge is concentrated in sitting out crashes, so this will underperform an always-long basket during a straight-up bull.
How much of the book was in the market, every day.
Height is how many of the eight coins the rule held at the close. Where it touches the floor, everything was in cash. Those stretches make the shape above possible.
- BTC
- 52%
- BNB
- 45%
- ETH
- 42%
- LINK
- 39%
- SOL
- 36%
- XRP
- 34%
- DOGE
- 31%
- AVAX
- 30%
Share of the 2,398 trading days each coin was held · 857 days entirely in cash. Both figures are exact; the shape above is plotted as weekly means. Holding fewer coins is not the same as holding less money. Capital concentrates into whatever is still trending, so average deployed exposure over the window is 44.8% rather than the average height of the shape.
What we actually control.
Not the market. Four gates stand between a signal and the exchange, and every one must clear before an order becomes a trade. Each makes the system do less.
Four checks. One decision.
Every order is checked before it can reach the exchange.
1 / 4
Per-order ceiling
No single order may exceed a fixed notional.
Over the line it is skipped, not shrunk. A mistyped figure should do nothing.
2 / 4
Portfolio gross cap
Total long exposure is bounded across every coin at once.
Eight coins turning bullish must not stack into eight times the size.
3 / 4
Drawdown breaker
A fall from the high-water mark past a set percentage halts new entries.
Position caps bound trade size. None of them notice the account simply losing.
4 / 4
Daily-loss breaker
A fall from the 00:00 UTC equity anchor past a set percentage halts new entries.
A slow bleed and one bad day are different failures.
- The halt latches
- A tripped breaker is written to disk and clears only on an explicit acknowledgement naming the event. The engine restarts on a schedule, and a halt held in memory would vanish within seconds and re-enter the position it just refused.
- Exits always run
- A halt blocks opening, never closing, and the kill switch closes with widening slippage steps until it is flat. A halt has to stop the bleeding, not freeze the patient.
One caveat we owe you: the breakers were built and tested, and the halt-latch behaviour is covered by tests that drive a real drawdown through a fresh engine. But they have not yet stopped a real loss with real money, because there has not been one. Until then they are engineering, not evidence.
A Rust core that can be killed.
Real engineering, not a wrapper. The frontend makes the safety legible; the engine is what stays in control.
What a tripped breaker does, in order
A limit is crossed
New entries stop on the spot. Exits are untouched.
The halt is written to disk
Not held in memory, where a restart would erase it.
The engine restarts
It runs on a schedule. It comes back up still halted.
An operator acknowledges it
By naming the event. Nothing else clears the latch.
Hyperliquid's validators sit in one region. We don't pretend to beat physics. An honest latency beats a fake number.
- Exact-decimal money math
- rust_decimal everywhere. An f64 on a price is a bug, not a rounding error.
- Two independent breakers
- A drawdown limit and a daily-loss limit. Either halts the account alone.
- Halts that survive a restart
- Written to disk; only an explicit operator acknowledgement clears one.
- Testable before a cent moves
- Every behaviour runs against a mock exchange. Testnet before mainnet, always.
- Drilled on testnet
- Kill-9 and flatten drills passed; they are being re-run against the current engine before real money.
HyperCore is infrastructure, not advice.
We do not promise, project, or guarantee returns. Trading loses money, and leveraged trading can lose all of it. Ten of the eleven strategies we tested lost. We publish that list because it is the most useful thing we know. Nothing here is investment advice, and nothing here is a track record yet.
There is nothing to buy. The live record starts August 2026 with the author's own money, at 1×, and it will be published here as it happens, including the drawdowns. If it fails, that will be on this page too.
No signup, no sales call. The second one goes to a person who wrote the engine.