What the data actually says
Every article below is a faithful public retelling of a findings document in our version-controlled research record — same numbers, same t-stats, same caveats. We measure on millions of ticks we recorded ourselves, simulate with honest fills (stops on gap bars fill at the adverse extreme, not at your stop), and we publish negative results as first-class products. If a claim on this site can't be traced to a findings doc, it doesn't belong here.
Crash 500 and Boom 500 are structurally untradeable — 2.6 million ticks of proof
The flagship negative result. The clean grind funds the violent spike to within 0.018% of total motion; spikes are single-tick gaps that turn a "+2.16R" backtest into −0.47R the moment stops are modeled honestly; and spike timing is memoryless — the "every ~500 ticks" rhythm is a mean, not a clock.
Gold has no directional edge — but its volatility keeps a schedule
A year of gold: every directional setup dies under a t-test; volatility clusters and concentrates ~50% harder in the London/NY overlap; and the one construct that survived strictly causal validation (+0.185R, t=1.70) is a trend-regime filter on a breakout — a candidate now standing trial, not an edge.
We tried to break our own strategy on two more markets. Here's the score.
Silver refused to confirm the mechanism (inverted tercile signature, cost-confounded); EURUSD confirmed the mechanism's shape but failed the filter again. Structure 3/3 · mechanism 2/3 · filter 1/3 — the doubt is now localized, and the shadow test is the arbiter.
Methodology constants across all studies: ticks recorded at the source (~1/second), market-hours-aware session handling (returns never measured across a closure), no lookahead anywhere, spike gap-through fill modeling, significance tests over point estimates, pre-registered predictions on replication runs, and a standing no-tuning rule — parameters that look better after the fact are reported, never adopted.