You Don't Have Five Positions. You Have One.

On October 10, 2025, traders running 1.5x leverage got liquidated.

Not 50x. Not 20x. One and a half. That number should bother you, because everything you've been told about blowing up an account involves reckless size, and 1.5x isn't reckless by any definition.

What killed them wasn't leverage. It was where their collateral lived.

Here's the mechanic. In cross margin, your account balance is one shared pool backing every open position. Lose on one, and the loss comes out of the same collateral holding up all the others — their liquidation prices slide closer without you touching them.

Isolated margin does the opposite. Each position gets its own walled-off margin, and when that margin is gone, that position dies and nothing else does. You lose more positions this way. You lose the account less often.

Most people run cross margin because it's the default on most venues and because it looks like the smarter choice. It is more capital-efficient. Profit on one position cushions a loss on another, and you get liquidated later than you would in isolated mode. That's true right up until it isn't.

The assumption baked into cross margin is that your positions won't all lose at once. In crypto, that assumption is wrong on exactly the days it matters.

October 10 is the cleanest example we have. Bitcoin fell 14.5% from its Friday high. Ethereum dropped 12.2%. Solana briefly lost more than 40% in the same window. Nothing about holding all three was diversification — they fell together, in the same hour, for the same reason.

Run five altcoin longs in cross margin and you don't have five positions. You have one position with five entry prices, five funding payments, and a single point of failure.

Over 1.6 million accounts were liquidated that day, roughly $19 billion closed by force, with about 70% of the damage compressed into forty minutes. CoinGecko's post-mortem names the amplifier directly: cross margin. Each losing position drained the shared pool, which pulled every remaining position closer to its liquidation price, which triggered more forced selling.

The part that should bother you most is what happened to the professionals.

Market makers running delta-neutral books — long spot, short perp, no directional exposure by design — got taken out too. Auto-deleveraging closed their profitable short legs to cover other traders' losses. Their long legs stayed open, suddenly unhedged, in a market where $3.2 billion in positions closed inside a single minute. A position built specifically to have no opinion about price ended up with a very expensive one.

These weren't gamblers. They had risk systems, hedges, and a correct understanding of their own strategy. What they didn't have was a margin structure that survives a day when everything correlates.

Most retail traders never make this choice consciously. You open a position, the mode field says "Cross," you leave it. I did that for a long time and told myself it was fine because my per-trade risk was small.

My per-trade risk was small. My per-account risk was 100%, and I had never once done the arithmetic.

So here's the question worth asking, and it isn't "how much am I risking on this trade." You probably know that number. Ask instead: if every open position moved 20% against me inside the same hour, what's left in the account?

In isolated margin you can actually answer that. It's the sum of the margins you allocated, and it's bounded. In cross margin the honest answer is usually the whole balance, and no amount of position sizing makes that number smaller.

Switch to isolated for anything speculative. Keep cross margin only for genuinely offsetting positions, and only if you've accepted that the offset can vanish in a cascade. Then cap how many correlated positions you hold at once, because in crypto "correlated" means very nearly everything.

Every strategy we run at v33 sizes and isolates risk at the position level, because the alternative is a system whose worst case gets defined by a day nobody modelled. Full backtest data and methodology are at v33systematic.com.