Internal validation · Results published in full
RTS Pattern Profi investigates whether historical price-pattern analogues can forecast market direction. Our tests on BTCUSD and EURUSD did not confirm a robust predictive or trading advantage. We are publishing that result clearly, including the failures.
Public conclusion: in the tested configuration, RTS probabilities were worse calibrated than a constant base-rate forecast and the apparent trading edge did not survive an independence check and realistic costs.
These are walk-forward observations from H4 data using a fixed 20-bar forecast horizon. The non-overlapping subset retains one observation per horizon to reduce dependence between neighbouring forecasts.
13 Feb 2020 – 17 Jun 2026
The gross R proxy was slightly above one, but direction accuracy remained random and probabilities were materially miscalibrated.
13 Feb 2020 – 16 Jun 2026
The full-sample margin was only +0.0091 R per trade before costs. The non-overlapping sample fell to −0.0817 R and PF 0.886.
On EURUSD, realised upside stayed close to 50% across almost every forecast bucket. A reported 84% upside score was followed by an actual rise only 45% of the time.
The panel labels P(up) and P(down) describe the weighted outcome share among selected historical analogues.
They did not behave like calibrated probabilities on the tested data.
Until a revised model passes new out-of-sample tests, these values should be called analogue outcome scores, not expected win rates or probabilities of future market direction.
This is a central limitation, not a minor disclaimer. It changes how every directional panel result must be interpreted.
The research EA recorded every eligible closed-bar forecast and its later realised outcome. Analysis was performed outside the indicator so that reporting rules could not change after seeing individual trades.
Each forecast uses only information available at its reference time. Future H4 bars are used solely to measure the outcome.
Brier score and reliability buckets test whether the reported confidence corresponds to observed frequency.
A one-per-horizon subset reduces the artificial confidence created by overlapping 20-bar forecasts.
RTS is compared with the unconditional base rate and a basic momentum direction, not only with its own past versions.
Mean R, TP-first, SL-first and a gross profit-factor proxy are reported. Costs are explicitly separated.
Weak years, negative subsets and scenario categories that underperform remain part of the published conclusion.
A durable research claim should remain directionally consistent through time. The observed proxy profit factor did not.
| Year | Records | Accuracy | Brier | Mean R | Gross PF |
|---|---|---|---|---|---|
| 2020 | 1,375 | 49.75% | 0.308 | +0.008 | 1.011 |
| 2021 | 1,559 | 48.49% | 0.311 | −0.011 | 0.984 |
| 2022 | 1,554 | 51.80% | 0.287 | +0.099 | 1.139 |
| 2023 | 1,554 | 51.35% | 0.290 | +0.077 | 1.119 |
| 2024 | 1,560 | 45.58% | 0.310 | −0.121 | 0.828 |
| 2025 | 1,554 | 50.58% | 0.293 | +0.014 | 1.020 |
| 2026* | 705 | 49.50% | 0.301 | +0.001 | 1.001 |
* 2026 is partial through 16 June. All PF values are research proxies before spread, commission and slippage.
We found no basis for claiming calibrated directional probabilities, reliable signal accuracy, a robust standalone trading edge, guaranteed performance or suitability for automatic execution.
RTS can visualise similar historical paths, dispersion, support/resistance context and temporal metadata. These are exploratory research aids, not validated trade recommendations.
Current published evidence covers BTCUSD and EURUSD on H4, primarily with a 24-bar pattern and 20-bar horizon. Results may not generalise, but alternative settings require fresh pre-specified tests—not marketing extrapolation.
Gross proxy PF excludes spread, commission, slippage, execution latency and financing. On EURUSD, the full-sample margin was so small that approximately 0.0091 R of cost per trade would remove it.
This project is educational and experimental. It does not provide personal investment advice. Do not risk capital on these outputs. Leveraged trading can cause rapid and substantial losses.
The screenshot documents the tested v23.11 instrument. Labels shown inside the historical capture are preserved, but the published findings above supersede any implication that its percentages are calibrated forecasts.
All commercial plans and paid upgrade paths have been removed. Future work is organised as research: hypotheses are written first, tests are reproducible, and negative results remain public.
Access, if distributed, is for education, review and reproducibility—not as a paid signal product.
Updates will not erase earlier failures. New models will be reported as separate versions and tests.
Calibration, independence, costs and time stability must all pass before any stronger claim is considered.
The published BTCUSD and EURUSD evidence does not support that claim. It should not be used as a standalone trading system.
Not on the tested configuration. They are weighted historical analogue outcome shares and were not calibrated to realised market frequency.
Because a negative result protects users, prevents unsupported marketing and provides a valid foundation for better research.
No paid plans are offered on this page. The project is presented as non-commercial research.
Possibly, but it would require a new, pre-specified model and fresh out-of-sample validation. It cannot inherit credibility from the current results.
This is an internal research validation by the project author, not an independent third-party audit. That distinction is part of the disclosure.
RTS Pattern Profi did not meet the required validation gates. The project therefore continues, if at all, as an open investigation of why historical similarity fails and what a genuinely testable forecasting model would require.