Fifty days, 1,809 closed trades, cumulative P&L of 20,293 notional dollars with a maximum drawdown of 5.4 percent. Below is how we calculated it, how we double-checked it, and what we still cannot claim.
This is not a backtest. The lab runs live: it captures the Binance market in real time and keeps score continuously, 50 days straight, hour by hour. Each trade was opened when the signal appeared, and its outcome was calculated later, on data that didn't exist at the time of entry. You can't fit such a result retroactively. That said, execution is virtual: no orders were sent to the exchange, no real money was traded on these lines, and fees, funding, and slippage were modeled on real quotes and rates. Position size is fixed at 1,000 notional dollars, with no reinvestment: the total is the arithmetic sum of 1,809 individual trades, not deposit growth from zero to twenty thousand. We don't sell signals, we don't manage other people's money, and we make no promises about returns.
In a niche where every other person shows a ninety percent win rate, the only credible argument for trust is showing how we cut down our own ideas.
BigWay is not a bot; it's our internal derivatives lab. It continuously captures market anomalies: funding rate skew, open interest spikes, long/short positioning skew, abnormal volume. Each hypothesis then goes into its own execution engine, where trades are calculated with honest costs: fees, slippage, funding for holding.
We set a high acceptance bar: a result only counts if the hypothesis survives all checks at once.
During the lab's operation, more than twelve hypotheses went through this pipeline. Almost all were rejected and remain in the rejected log. Below is an analysis of what survived and what we still cannot claim.
We don't publish what exactly was changed. We publish how strongly it changes the result.
The experiment is set up like this: the input stream is the same, the engine is the same, the costs are the same. Only the rule configuration differs, and which one it is stays inside the lab. Eight parallel lines were calculated simultaneously in real time on the same input stream, each keeping its own trade log.
We don't disclose what's behind letters A and B. That's the only part of the work that can't be reconstructed from the outside, and it's the part that's worth money. Everything else on this page is open and verifiable: period, trade count, drawdown, win rate, average trade sizes, and the entire control methodology. A client gets the configuration with the project; a case reader only needs to know it exists and that its contribution was measured.
The difference between the best and worst line was 42,154 notional dollars on completely identical inputs. That leads to the conclusion this whole analysis was made for: a track record by itself proves nothing. The same trades fit into a range forty-two thousand wide, and until the entire setup is fixed, a pretty curve means absolutely nothing.
The last two tiles matter more than they seem. The average losing trade is larger than the average winning one: 71 dollars vs 49. The line holds not on the size of wins but on their frequency: there are twice as many wins. That's also its main vulnerability, and we watch frequency, not the total.
We don't publish a breakdown of how positions were closed: it would reveal the setup. That doesn't affect verifiability. The period, trade count, drawdown, win rate, and average trade sizes are fully listed above and haven't changed since publication, and any of these numbers can be recalculated with the same methodology on your own data.
A number you can't double-check counts as zero
This section is usually cut out. We write it because without it, everything above is worthless.
Separate from these experiments, there's the Cloud Diver line: cumulative volume delta divergence plus an ML filter. It passed walk forward and is released as a separate public product with open verification.
We don't hand over a strategy; we hand over a verification engine.
Strategies come and go, but the question is always the same: does this work, or am I being scammed? The same pipeline we used to kill our own ideas is applied to someone else's: your idea, a bought bot, a signal channel with a pretty screenshot.
Send an idea, a bot, or trade logs. We'll run it through the same pipeline and tell you what we saw. If we see it doesn't work, we'll say so.