A public log of
what didn't work.

The retail trading industry runs on survivorship bias. Take a hundred random strategies, show the world only the one whose equity curve happened to go up, and you have a marketing asset. Nothing about that process is evidence — it is selection dressed as skill. The uncomfortable truth is that a positive backtest is the easiest thing in this field to manufacture and the least reliable thing to trust.

So we invert the incentive. Our credibility does not come from a winning strategy we are trying to sell you — we are not selling one. It comes from showing our work, including every result that embarrassed us. A site that documents its own failures has nothing to hide behind. That is the whole moat: you cannot fake a paper trail of honest losses.

Every idea here was tested with walk-forward validation. We march through time in folds — train on a past window, test on the next window the model has never seen, step forward, repeat — so a strategy is only ever judged on a future it could not have memorized. No shuffling, no peeking, no look-ahead.

A passing walk-forward result is where our scrutiny starts, not where it ends. Before anything earns belief it has to survive four attacks at once:

• Real costs — spread and commission on every simulated trade. Gross-only edges do not exist.
• Out-of-sample discipline — chronological splits, trained on the past, judged on the unseen future.
• Honest labels — one shared feature-and-labeling implementation across training, verification, backtest, and live serving. Never a second copy that quietly drifts into train/serve skew.
• Concentration & regime check — the question that kills most strategies: does the profit come from the whole period, or from one lucky week or one unusual regime?

We never tune a strategy until it turns profitable. Tuning until the number looks good is just overfitting a signal that was never there.

Between a theoretical edge in a price series and a solo retail trader stands a wall of spread, commission, and market efficiency — the retail cost wall. We ran into it eight times. Here is every attempt, its hypothesis, and precisely what killed it.

Adding exogenous data (the dollar index, real yields) generally made things worse. Only the Commitment of Traders series carried genuine information — and even that meaningfully helped in just one bearish stretch.

Experiment E-07 deserves its own section, because it is the exact trap that turns retail traders into believers. The H1 swing strategy returned a genuine +$8.9k across a multi-year backtest. Every top-line metric looked like a winner. Had we been trying to sell something, this is the screenshot we would have framed.

Then the concentration check asked one question: where did the money actually come from? The answer was brutal — 86% of the entire profit came from a single week. Strip that one week out and the “edge” collapses into noise. The strategy had not found a repeatable process; it had been in the right place for one extraordinary move and spent the rest of the time treading water.

When we then bolted on disciplined fixed-risk stops to “stabilize” it (E-08), the account fell roughly 90% — the fixed stops turned out to be the width of ordinary market noise. Two honest deaths from what looked, at a glance, like the best idea we had.

One model cleared the bar. Not a direction predictor — a volatility-expansion classifier that estimates the probability of a large move over a defined forward horizon. It beat a naive baseline in every out-of-sample fold and across bull, bear, and range regimes. It powers the live reading on our homepage.

We are careful about what that means. Predictable volatility is not the same as profit. Knowing a storm is likely is useful context; it is not a trade, and the model says nothing about direction. That is why we publish it as a measurement and refuse to dress it up as a signal.

Taken together, these results are a careful, expensive-to-obtain measurement of just how efficient the gold and FX markets are for a solo participant. We consider that a finding — not an embarrassment. It is also a standing invitation: if you believe one of these verdicts is wrong, or a method is flawed, tell us and we will investigate and publish a correction.

Everything in this journal is educational research. It is not financial advice, a recommendation, or a solicitation to trade anything. For exactly how our live numbers are built, read the methodology; for the current positioning picture, see the weekly COT report.