Proof of the method
First we killed 18 of our own strategies.
Before touching anyone else’s strategy, we tried to kill our own. We built and measured 18 classes of trading strategies on our own ideas. All 18 failed to clear the bar we set in advance. We never risked real money once. If we didn’t spare our own ideas — a stranger’s promise backed by a single screenshot stands even less of a chance.
Parade of deaths · 18 classes
The headline result
How an edge is born and dies
In a young prediction market a simple rule earned about 2% per trade in 2023–24. We wrote the bar down before we looked — and the rule cleared it. By early 2025 the same rule had gone to roughly −4%: the professionals arrived and drank the puddle dry. Public retail edges don’t live long — they live in the young corners of the market, and they die when the money shows up.
Twice we caught ourselves too — a look-ahead in the code and a batch of stale prices — and we threw those results out. Better to lose a finding than to publish a flattering mistake.
Two public audits, in numbers
Exactly what we test
The same honest stand for every claim. Each item is a separate check that a promise with a pretty chart usually fails.
Real costs
Fees and slippage, as in real trading — not in the vacuum of an ad.
No look-ahead
The whole history in full, no fitting parameters after the fact.
No survivorship bias
We count those who vanished from the market too — not just the winners.
We split the result in half
We check whether the edge holds on the second half of the data.
Overfitting test (PBO)
Without it a seller could try hundreds of settings and show the one lucky one — that’s not a strategy, it’s a lottery.
Adjustment for the number of tries (DSR)
The more variants tried, the higher the bar — otherwise “success” is just luck out of many attempts.
Cherry-picking test (SPA)
The best variant really beats the market — or it was just picked out of many after the fact.
Sealed in Bitcoin
Stamped with a tamper-proof timestamp anyone can check. The date and the “pass / fail” line can’t be moved after the fact.
Why only we can do it
Not a tool, but a stance. To judge this way you have to be able to measure at an institutional level and have no reason to look away.
A hedge-fund-grade engine
De Prado’s methods — PBO/CSCV, Deflated Sharpe, purged cross-validation — we wrote ourselves. Where others pay for closed libraries or can’t do it at all.
First we killed 18 of our own
18 classes of strategies on our own ideas — all dead. We don’t spare even ourselves.
Zero conflict of interest
No broker ads, no strategy shop, no signals of our own. The only ones paid precisely for a “no.”
We won’t burn down with someone else’s crash
We don’t predict the future — we testify about the past on terms announced in advance.
Concretely. A plain SuperTrend backtest would show the advertised “68% of trades won.” We added real costs, a walk-forward check (tested on later data it never saw) and a cherry-picking test — and got 29–47% and a loss on any timeframe. The difference is in WHAT we test beyond an ordinary run.
And the competitors have a reason to look away: one lives off the ads of the brokers it’s supposed to judge; another earns more the more strategies there are; a third sells its own. We sell none of that.
Why it’s worth the money
A $2,500 audit is cheaper than one mistake. An unchecked strategy that bets on borrowed money loses the deposit in a week; a year of subscriptions to the “signals” we refute costs more. For an allocator, one rejected bad fund pays for years of audits.
Money-back guarantee
Find a real methodological error in our work — we refund the money and publish the breakdown. No one else in this niche offers that.