People who trusted my system got liquidated today. Not a display bug. Not a wrong number on a dashboard. Real people, with real money — most of them with $500 accounts — woke up to find their positions had been forcibly closed by the exchange because their margin ran out.
Silver crashed from $88 to $66 in forty-eight hours. Gold dropped with it. The two assets that were supposed to be the safety net — the hedge against crypto volatility — turned out to be the thing that sank the boat.
The strategy was designed to handle drawdowns of up to 38% across the basket. Being in unrealized loss is normal. That's the whole point of DCA — you accumulate during downturns, average down your entry, and profit when price bounces back. The crypto bots were fine. BTC, ASTER, HYPE — all stuck in their ranges, doing exactly what they're supposed to do. Months of sideways, small wins compounding, the system humming.
Silver didn't care about the system. Silver dropped 25% in two days and took every small account with it.
The position sizing wasn't conservative enough. That's the honest answer. The strategy works — the logic is sound, the math holds in backtesting, the bots execute correctly. But the sizes I chose assumed that silver and gold would move independently from each other and from crypto. They didn't. They crashed together, at the same time, in the same direction. The diversification that was supposed to spread the risk concentrated it instead.
When your hedge moves against you harder than the thing it's hedging, it's not a hedge anymore. It's a second bet in the same direction.
And the people who were counting on the diversification didn't sign up for a double exposure.
• • •
I also cost myself $2,500 yesterday morning. Different problem. Same lesson about carelessness.
I was testing something on the strategy test account and accidentally closed all five of my personal positions. The test account and the personal account share the same exchange — different API keys, same underlying margin pool. One moment of not checking which key I was using. I realized instantly. Too late.
Five positions, all closed at market. About $2,500 in realized loss across the board. Twenty minutes to recover — force-bought back Layer 1 for every bot, rewrote the state files from actual fill prices, restarted everything. By 9 AM the bots were running again on new, smaller sizes. BTC down from 0.025 to 0.008. Silver down from 6 to 2. Gold down from 0.1 to 0.07. ASTER disabled entirely for now.
The money is gone. The system is running. Both things are true.
I didn't hide it from this journal. I'm not going to hide the client liquidations either. Day 28 I wrote about telling my brother the truth when his numbers were wrong. Day 30 I wrote that the system is fragile. Today the system showed me exactly how fragile. Not the code — the code held. The assumptions underneath the code.
• • •
Here's what I'm sitting with tonight.
The strategy isn't wrong. Farming volatility in a range-bound market works. The bots prove it every day on BTC, on ASTER, on HYPE. Small wins, compounding, patient accumulation. That part is real.
What was wrong was the sizing. I scaled positions before I fully understood how the assets correlate under stress. Silver and gold looked independent in normal conditions. In a crash, they moved together. The diversification disappeared exactly when it mattered most. And the people with $500 accounts — the ones who could least afford it — were the first to get liquidated because smaller accounts hit the margin wall faster.
Going forward: smaller starting sizes. Validate parameters on real markets before scaling. The backtest told me 38% drawdown tolerance. The market told me that tolerance means nothing if three assets crash simultaneously.
Your safety margin isn't a number. It's a scenario. "Can the portfolio survive a 38% drawdown?" is the wrong question. The right question is: "Can it survive the worst thing that could happen to every correlated asset at the same time?"
Those are very different questions. I was answering the first one. The market asked the second.
• • •
The technical work continued around the wreckage.
Twenty-six clients running now. Rizalino onboarded clean — fifth client on the newer API version. Found and fixed a duplicate order bug that had been placing multiple layers at the same price — a stale data cache after a failed order made the bot think the position was empty when it wasn't. Fixed the watchdog so it can tell the difference between my personal bots and client bots with the same filename in different folders. The system is better tonight than it was this morning. The code is cleaner. The bugs are fixed. The new client is live.
None of that changes what happened to the people who got liquidated.
• • •
Thirty-three days. Twenty-six clients. A strategy that works in a range and breaks when the range breaks. A $2,500 personal loss from one careless moment. Clients who trusted me and got hurt — not by a bug, but by sizing assumptions I should have tested harder.
Day 32 I wrote that the machine remembers what I forget. Today I learned what the machine can't remember for me: the limits of my own assumptions. The code doesn't know if the position size is too big. The bot doesn't know if silver and gold are about to crash together. The system executes whatever I tell it. If what I tell it is wrong, it executes wrong — perfectly, automatically, without hesitation.
The bots aren't the risk. The person directing them is.
Day 33 complete. Twenty-six clients. One hard lesson. Smaller sizes going forward. The hedge became the hazard. I won't forget why.
Day 33 of ∞ — @astergod
Building in public. Learning in public.