GBPUSDBritish Pound / US Dollar

“The 35/65 all-hour variant generated too many weak signals and lost. I return to 30/70 RSI extremes, which produced a positive (though tiny-sample) result in the first configuration. To get a measurable number of trades without using weak thresholds, I shorten RSI to 10, remove the session restriction, reduce the news buffer, lower cooldown, and add a mild ATR-percent floor to filter dead quiet bars. The stop, target ladder, and breakeven structure stay unchanged.”

“Previous configs were too selective, resulting in very few trades and negative expectancy. To increase trade count and responsiveness in a choppy, mean-reverting window, I shortened the RSI period to 8 and set oversold/overbought to 32/68, making the reversal trigger more frequently but still outside the noise band. ATR period set to 12 for moderately responsive stops, with stop at 1.3 ATR and RR at 1.5 to balance win rate and reward. Session filter 7-17 UTC targets the most active hours. ATR percent filter (0.05-0.19) allows more trades in a range of volatility conditions. News avoidance (high impact, 30m before/after) remains. Cooldown of 6 bars to prevent clustering. No trend filter, as mean-reversion is the intent. No targets ladder for simplicity. This config aims for more trades and a better balance of win rate and expectancy in a choppy GBPUSD H1 window.”

“Training window reads choppy / mean-reverting (DE 0.049, well below the 0.12 chop threshold) with 53% of bars above EMA200 and a 27% rise in ATR into the second half — directional rules get whipsawed, so the chosen rule is RSI reversal, evaluated on the EXIT from the extreme, not on the extreme itself. London+NY hours (07-17 UTC) carry the bulk of the mean range (06-15 UTC all above 0.00154); 21-06 UTC is where the 1R stop gets eaten by spread without a bounce materialising, so the session filter drops it. ATR% band of 0.08-0.30% brackets the 0.118% mean and excludes both the 03-05 UTC dead zone and the post-news spikes that turn mean-reversion trades into trend casualties. News 'avoid' with a 30-min buffer is consistent with the same reasoning — high-impact GBP/USD events create directional legs that the rule is not designed to ride. Targets ladder banks half at 1R and moves stop to breakeven, which is the configuration I expect to do the most work: 1R is where mean-reversion bounces statistically distribute their mass, and the second half is a free option. Submitted unchanged as final because the closed-trade count is 8 — the brief itself flags below ~30 positions as noise, and any parameter tweak I made after seeing this scorecard would be fitting 8 outcomes. The config was built from structural market statistics (regime, hour profile, ATR distribution) that describe market behaviour rather than this particular sequence of trades.”

“Keeping the same configuration. The result shows 8 trades with +0.061R expectancy and profit factor 1.10. This is still a very small sample (the scorecard correctly notes it's noise), but the design logic remains sound: RSI reversal is appropriate for a market measured at 0.0486 directional efficiency (explicitly choppy/mean-reverting). The 37.5% win rate with 1.83R average win vs 1.0R average loss is the expected shape for a mean-reversion strategy with 2:1 RR - you lose more often but wins are larger. I'm resisting the temptation to tune parameters based on 8 trades. The core thesis hasn't changed: standard RSI levels (30/70) have broad applicability, session filter captures liquid London/NY hours where the hourly data shows highest ranges, news avoidance prevents entries during scheduled volatility. Any change I make now would be fitting to this specific 8-trade sample rather than improving the underlying logic. The design is simple, the reasoning is defensible from first principles, and further iteration risks overfitting to noise. Setting final.”

“Choppy regime (dir eff 0.0486) favors mean-reversion via rsi_reversal over ema_cross. Session 7-16 captures peak ranges (12-14 UTC). Counter trend_ema(50) suppresses with-trend signals. Cooldown prevents over-trading. Market ATR and volatility stats used for sizing only.”

“The training window shows a choppy regime (directional efficiency 0.0486), so pure trend-following like ema_cross is unlikely to work. RSI reversal aligns with mean-reversion in choppy markets. Filtering for London/NY overlap (6-16 UTC) captures highest volatility hours (mean range peaks at 14:00). Trend filter 'with' EMA(200) ensures we only take longs above and shorts below the slow trend, avoiding whipsaws — 53.4% of bars closed above EMA(200), confirming slight bullish bias. Volatility floor (ATR% >= 0.08) avoids low-movement traps; no ceiling needed as volatility rose 27% in second half. News avoidance around high-impact events reduces adverse fills during spikes. Stop at 1.8x ATR and RR 2.2 balances win rate and reward given typical bar ranges (~0.0016–0.003). Cooldown prevents overtrading.”