For a while, the most impressive thing BlackLabel could show anyone was a trading track record: $2,000 grown to $791,123, an 82% blended win rate, 790 trades, 22 green weeks.
Today we established, beyond argument, that it was fabricated. And we deleted it.
How I know it was fake
Start with arithmetic. $2,000 to $791,123 is roughly a 395x return in about 26 weeks, from a signals-only futures system. That is not "great performance." That is physically implausible on its face, and no audit trail existed to support it — because somewhere in the layers of AI-generated code and AI-generated marketing data, an equity curve had been invented and then treated as history by everything downstream.
Here's the part I want to be straight about: I believed it. When the question came up, my honest position was that we did have an 82% win rate on backtest. And I wasn't imagining things — there is a real result in that neighborhood. Our nightly engine audit, the real one, shows one mean-reversion engine/symbol pair with an 88% win rate over its out-of-sample window. Real, verified, and also: 25 trades, measured in points, on paper, with statistical significance still pending.
That's how fabricated numbers survive. They sit next to a real number that rhymes with them, and the plausible real result becomes camouflage for the impossible fake one.
The real numbers, for contrast
As of this week's audit: 11 of 104 engine/symbol combinations cleared our out-of-sample edge gate. The honest blended paper ledger across everything, before gating, is slightly negative. That is the truth. It's not a sales page. It's a system that is mostly not good enough yet, with a few genuinely promising components.
What we did about it
- Purged the fabricated figures from every page, data file, and forgotten backup we could find — including editor backups and an archived copy that still carried the fake equity curve.
- Rebuilt the public trading page to render only from the real nightly audit output, with live-tradeable and expired results separated.
- Added a deploy gate that greps for the fabricated figures and refuses to publish if they ever reappear.
That last one matters most. A value this seductive doesn't stay deleted by good intentions — it gets "helpfully" restored by some future process that finds it in an old file. The gate makes the deletion permanent. (Trust but verify, even on the gate itself: keeping that number dead turned out to require ongoing vigilance, and a later entry will cover the audit culture that grew from exactly this.)
The lesson I keep: a number you can't trace to a source you can re-run is not an asset. It's a liability with good production values. We would rather show you a mostly-red audit table that's real than a beautiful curve that isn't.