The graveyard
The strategies we tested that failed.
Every entry below was tested on real data and lost — and it's published, with its sample size. This is the one page a signal seller can never show you, because publishing it would end their business.
Ask whoever's selling you signals to show you their failure list. That silence is what you've been paying for.
Fading every round-number touch, any trend. The location is real — gold's round numbers show a measured ~+10pt reaction over base rate — but the lazy rule at it loses money. The level was real; the rule was the lie. Location ≠ rule.
Entering blindly on a liquidity sweep, as a system. Loses net across 9,214 trades. (Our confirmation-gated, higher-timeframe-located sweep-reclaim is a different, selective thing — a few times a month, not this.)
We searched 419 mechanical configurations net of cost. Only 3.3% beat buy-and-hold out-of-sample — exactly what you'd expect from chance under multiple testing. Nothing mechanical robustly beat simply holding.
Six popular intraday gold systems, each tested across 17 years of 1-minute data. Every one loses once real costs are applied. A statistic is not a system.
Seductive in-sample. The sign flips out-of-sample — the hallmark of a curve-fit, not an edge. We publish it so you don't pay to relearn it.
Two crowd favourites. Neither beats its own base rate once measured honestly. Popularity is not evidence.
The portfolio looked like a robust momentum edge. Drop the single best contributor (crypto) and it vanishes. Always test whether one asset is quietly carrying the entire result.
We ran the same honest engine on these markets and found no mechanical edge. We show the negatives too — “no edge on the S&P” is on the desk, not hidden. Gold is the flagship because that's where the 17-year evidence actually is.
Education & analytics only — not signals, not advice, not FCA-regulated. Past results, including tested failures, do not indicate future performance. Figures are measured from real historical data with the sample size shown; where a result is qualitative we say so rather than invent a number.