What happened to the signals
This platform began as a signal service: SWIFT, PULSE, and ANCHOR — patterns that faded retail crowd positioning at statistical extremes. The directional edge was real for years. Then it stopped being real, and when we rebuilt the record with honest data, we published the result instead of the pitch. This page is that record.
Every number below comes from a full replay of the production signal logic — 9,133 events across 28 pairs from 2014 through 2026, priced on real bid/ask spreads with 1-minute resolution, and no forward-looking data. It is the strictest accounting we know how to do.
The edge, year by year
Share of signals that resolved favorably, by entry year. Bars grow up from the coin-flip line when the fade worked, down when it didn't.
View the data as a table
| Year | Events | Favorable rate |
|---|---|---|
| 2014 | 466 | 59.0% |
| 2015 | 655 | 54.2% |
| 2016 | 940 | 58.8% |
| 2017 | 783 | 55.8% |
| 2018 | 794 | 57.7% |
| 2019 | 789 | 56.8% |
| 2020 | 726 | 57.0% |
| 2021 | 684 | 52.9% |
| 2022 | 779 | 48.9% |
| 2023 | 762 | 47.1% |
| 2024 | 713 | 47.8% |
| 2025 | 806 | 41.9% |
| 2026 (partial) | 236 | 47.0% |
The edge was real — then it left
From 2014 through 2020 the crowd-fade resolved favorably 57.1% of the time after real spread costs — modest, but consistent, and it survived walk-forward testing. Since 2021 the same logic, on the same data, runs 47.5%: below a coin flip once you pay the spread. 2025 was the worst year on record at 41.9%.
The decay spared nothing. London AM — the strongest session in the record, 62–70% favorable every year through 2020 — has hovered near 46–51% since 2023. The edge also rotated as it faded: USD and AUD pairs led 2014–2020, EURCAD led 2021–2023, and what remains today is concentrated in CHF pairs. Parameter-sensitivity testing says the signal logic itself is robust, not overfit — the environment changed, not the arithmetic.
We do not claim to know whether the directional edge returns. We publish the record as it stands and keep replaying it.
The correction that mattered
Earlier versions of this platform reported win rates above 80%. Those numbers were wrong. They came from a backtest with forward-looking bias — tier assignments computed on data that included the outcomes being measured — and from mid-price fills that ignored the spread. When we rebuilt the record point-in-time on bid/ask data, the true all-period figure came out at 52.9%. We retracted the old numbers, retired the tier system built on them, and preserved the corrected studies in the research library — the corrections are the point.
Anything on this site that still shows pattern-era language or archives of those reports links back here. This page is the standing disclaimer.
What survived
One finding held in every era, every era boundary, and every robustness check: 92.8% of positioning extremes were followed by a move of 15+ pips — in one direction or the other. The crowd reaching an extreme reliably marks that something is about to move; it just doesn't say which way. Our published study of 75,864 extremes against a random baseline (RP-004) reached the same conclusion: positioning extremes are a volatility detector, not a direction detector.
That is why this platform pivoted. Instead of selling a direction, it measures the situation — where price sits, how stretched positioning is, what the options market is doing, and what has typically followed moments like this one. Context, not signals.
Methodology: v2 production replay of the exact SWIFT/PULSE/ANCHOR engine logic, 9,133 events, 28 FX pairs, 2014–2026. Entries and exits priced on real bid/ask (average cost 2.7 pips per event), excursions measured on 1-minute data, no forward-looking inputs. “Favorable” means the faded direction was ahead at the pattern's fixed resolution point. 2026 is a partial year. Full write-up: research library, report RN-014 (how we report our numbers).
See what the platform measures now — the situation, not a signal.