FOREX PATTERNS, TESTED HONESTLY
Trading myths and market patterns, tested on real data. Most don't survive.
Every pattern we find gets checked again on data it wasn't found in. If it holds up, it's an edge. If it vanishes, it was noise. New charts every day on X.
21patterns tested
6survived (edge)
15didn't (noise)
10statistics measured
16waiting to be tested
LATEST FINDING · 8 Nov 2026
All findings →
Statistic
The dollar's best day since July. Most of it came early.
Friday was the dollar's biggest up-day since July: No. 1 of 91 trading days, and all 7 majors fell against it.
The twist: 70% of the move came before the US jobs report at 1:30pm UK. The data added to the rally. It didn't start it.
FROM THE LEDGER
Full Ledger →| Claim | Verdict | What we found | Last tested |
|---|---|---|---|
| Does GBPNZD follow the pound?Correlation | Statistic | Barely. GBPUSD explains 1% of its hourly moves; EURNZD, which has no pound in it, explains 75%. It's a kiwi chart. | 6 Nov 2026 |
| Which pair moves the most pips?Statistic | Statistic | GBPNZD, about 126 pips on a typical day. But it's only 10th as a percentage of price, because each GBPNZD pip is half the size of a EURUSD pip. | 6 Nov 2026 |
| Fade a 100-pip day: 20-pip target, 50-pip stop?Rule test | Too weak | On GBPNZD it won 23 of 31 (74%), just above the 71% needed to break even. Across 28 pairs it's break-even. Too close to call; re-testing as data grows. | 6 Nov 2026 |
| After a big move, does a 20% pullback come before 20% more?Rule test | Noise | Coin flip across 28 pairs: 47% vs 48%. | 6 Nov 2026 |
| After 25 pips from the open, does price come back?Rule test | Noise | It returned 63% of the time, but fading it still lost money: the wins are capped and the losses aren't. | 6 Nov 2026 |
| Does the first 5 minutes of each hour move more than the rest?Time of day | Edge | Yes. Found in July, then re-tested on Aug–Oct: the on-the-hour slot beat the rest of its hour on 63 of 65 days, about 16% bigger, in all 8 markets tested. | 5 Nov 2026 |
HOW IT WORKS
Method →- FindScan every 5-minute bar of 43 markets, or write down a popular belief as a claim.
- SplitMeasure it on older data only, with the test fixed in advance.
- Re-testRun the same test on newer data it has never seen.
- VerdictEdge or noise, and re-tested as the data grows. Edges can die.