One edge avoids price. The other trades it — carefully.
Onyx-Algo runs two engines with two different edges. The carry engine captures a structural funding premium without forecasting price. The directional bot doestake a side — but only on confirmed trends, with strict entries and modeled costs: a measured edge we’re still proving in shadow, not a structural law. Here’s the case for each, and exactly what would break it.
Funding, in one paragraph.
A perpetual future never expires, so exchanges use a funding rate to keep its price glued to spot. When the perp trades above spot — too many crowding the long side — longs pay shorts a fee every few hours. Crypto skews long far more often than not, so that fee flows, on balance, to whoever holds the short. Onyx-Algo holds the short.
Three structural reasons.
Speculators want leveraged long exposure and will pay a recurring fee to hold it.
Little capital wants the other side, so the premium isn’t competed away.
Perps are the easiest leverage there is. A risk-transfer premium — not free money.
The price cancels. The funding doesn’t.
Hold one unit of spot and short one unit of the perp. If price rises, the spot gain offsets the perp loss; if it falls, the reverse. The price cancels out — what’s left is the funding. We don’t forecast direction; we collect the spread. The short leg runs on leverage and margin, sized with liquidation buffers — disclosed, not hidden.
This is the part the last cycle got wrong. The return is funding carry from real, hedged positions on your own account — not lendingyour coins to a counterparty, not rehypothecation, not a pooled “earn” product. Nobody borrows your assets. There is no platform balance sheet between you and your money.
Why it lasts — and when it doesn’t.
the demand for leverage is structural and durable. New entrants keep arriving, keep crowding the long side, and keep paying to be there.
markets go flat or bearish (funding falls, even turns negative), too much capital crowds the trade, or a deleveraging shock hits. The Negative book harvests negative regimes; capacity is capped on purpose.
What would break it.
How we manage each → Method.
The trade, in one paragraph.
The directional bot is an event-driven momentum system. It doesn’t forecast from a calendar or a hunch — it waits for a trend to already be established, then presses it: shorting coins that are clearly falling, buying ones that are clearly rising. Every candidate must clear a strict checklist of trend, momentum and volume conditions before it trades, and every simulated fill is charged modeled spread and slippage. Where carry avoids price entirely, this engine takes a deliberate, rules-bound side.
Measured, not assumed.
Carry rests on a structural fact: crypto is durably long, so funding durably flows to the short. The directional edge is different in kind — we claim no law of markets. If there is an edge here, it comes from discipline, not prediction: entering only trends that are already established, refusing everything that fails the checklist, charging every fill a realistic cost, and standing aside when the market gives no clear trend. That is a signal we are measuring in shadow and testing against the ways it could fail — not a premium we assert. The caution that follows carries as much weight as the case.
It takes a side — but only when both locks open.
This engine is directional: it profits or loses with the direction it picks, so selectivity is the whole safeguard — and there are two locks, not one.
First, a strategy isn’t even allowed to trade a coin until that coin’s own results clear a Bayesian edge test. Rather than ask “has it traded enough times?”, the gate asks “how likely is a real edge here?” — it builds a probability distribution over the coin’s true win-rate from its actual wins and losses, weighs that against the break-even implied by the strategy’s own payoff (how large its wins run versus its losses), and admits the coin only when the probability of a genuine positive-expectancy edge clears a deliberately high bar. Sample size alone earns nothing: a coin that traded often but never actually won is refused, and as fresh results arrive the test re-decides.
Second, on any given bar the entry must still pass a checklist of as many as eleven conditions — trend, momentum and volume all lining up on the same completed candle. Fail either lock and the bot stands aside; when nothing clears both, it simply doesn’t trade. It presses established trends and never tries to call a turn.
Two things stand between shadow and real money.
A signal that looks good in shadow still has to answer two questions before it earns real money — and neither is settled yet. We treat both as open; they matter as much as the case for the edge.
The trades that work are in small coins that don’t change hands much — where even modest buying or selling nudges the price. Today the bot trades in sizes small enough to barely leave a mark. Make those trades much bigger and it turns on itself: its own orders push the price, so it buys a little higher and sells a little lower than it meant to — and that cost climbs faster than the trade grows. Push far enough and it’s effectively trading against itself. No one has found where that ceiling sits — how much money the edge can hold before its own footprint eats the profit. Of the two unknowns, this is the more dangerous.
Trend-following works best when the market picks a direction and holds it. It has a harder time in choppy, back-and-forth markets — where a move looks real, pulls the bot in, then reverses on it: a trap. The bot is built to dodge the worst of this by acting only on trends that are already established and sitting out everything else. But the stretch we’ve watched it in happened to suit it. “It sidesteps the traps and waits out a long, sideways grind” is what we’d expect from the way it’s built — not something a lucky window has actually tested. Only a full market cycle, including the kind that works against it, will show whether that holds.
And nothing here trades real money: the directional book is paper, in shadow, until the FCA gate.
What could break it — and the check on each.
The full validated record publishes here once we complete FCA authorisation. In the paper phase we don’t show performance figures publicly.
Returns wait until the shadow record matures — but the desk is already transparent. Members watch the bot decide in real time: every coin it’s watching and how close each is to firing, and for each signal whether it entered or which gate stopped it, and why. The checklist behind it is parity-locked to the live engine — verified to match its decisions on every bar tested — so what members see is exactly what the bot did, not a reconstruction.
Capital at risk. The thesis describes a strategy, not a promise — funding premia can compress, turn negative, or be overwhelmed by venue, liquidation or regulatory risk. Onyx-Algo is a non-custodial automation service for certified sophisticated & professional investors; not advice, not a deposit, not FSCS-protected. Past simulated performance is not a guide to future results.