Saying that increasing the conditions would make the EA stronger was a mistake—the things that were added and disappeared during the development of the boundary

As you add conditions, the EA becomes smarter. Following that belief, many of the added conditions did not remain by the end.

Merely touching was not enough to gain an edge

Last time, we found that horizontal lines react more readily to the “First Touch at the pullback.” However, that was a diagnostic value based on movement. To actually trigger trades, we needed to narrow down “what market conditions make mean-reversion more likely.”

So we tried adding every conceivable condition to the M15 First Touch mean-reversion strategy one by one. Bollinger Bands (BB), ADX, RSI, Stochastics, H1 Perfect Order (H1 PO), etc. We kept testing while changing each condition to see which ones were truly useful.

Only BB and ADX remained. RSI and Stochastics fell away

Being a mean-reversion strategy, the target is a scenario where “price has moved too far, and the trend isn’t too strong.” BB measures the overextension, and ADX measures the strength of the trend. Only these two remained until the end.

The numbers were as follows. Verification results with a population of 5 currencies and 4,835 cases.

This “BB2.9・ADX22.5” remains in the current KYOKAI as is.

One more point to look at: the win rate dropped from 54.4% to 32.81%, but the PF rose from 1.19 to 1.221. If you narrow SL and widen TP, the win rate tends to drop.If you only look at win rate, you might discard the best shape.

Even adding H1 Perfect Order hardly changed anything

We also tried adding H1 Perfect Order (H1 PO) aligned with the higher-timeframe trend. We overlaid it on the PF1.221 condition of BB2.9・ADX22.5・TP100/SL40.

Number of trades

PF (bar at the far right set to PF1.6)

Note: PF is almost the same. The bar length looks almost the same, which is its meaning.

PF is about 1.221 to 1.227. Very little change. Meanwhile, the number of cases dropped from the 600s to around 400. It seems adding conditions reduces trading opportunities.

Increasing conditions does not always strengthen performance.

Trying to fix losing years ended up erasing good trades

In real trading tests, another area of difficulty was annual performance differences. The results when the exit was changed to H4 ATR were as follows.

• 2024: -2,646 yen (PF0.94)
• 2025: +34,666 yen (PF2.05)
• 2026: +3,556 yen (PF1.17)

Only 2024 was in the red. By analyzing this year's losses by H1/H4 trends, the reasoning was simple: “If the H4 close is below EMA20, prohibit BUY entirely.” So we tested a setting with that condition added.

The setting with BUY prohibited entirely improved in 2024. But in 2025, it dropped from +32,751 to +18,358 yen—a drop of about 40%. That’s because the previously profitable pullbacks for BUY were heavily restricted.Trying to fix a bad year ended up breaking a good year.

Therefore, we abandoned total prohibition and limited restrictions to certain Sell conditions on the right side (resistance). This became the final adopted setting. Across all three years 2023–2025, it improved relative to the baseline (2023 went from +5,395 to +6,745). The KYOKAI retains this limited restriction.

In the end, what remained and what disappeared

Summary

• What remained after adding was only BB and ADX; RSI and Stochastics disappeared
• Even with H1 PO added, PF stayed almost the same and the number of trades decreased
• There were patterns where PF rose despite a lower win rate. The decision cannot be made by win rate alone
• Trying to fix a losing year by prohibiting BUY reduced profits in good years
• Conditions are better kept by limiting rather than simply adding more

Next:I thought profits would rise if TP were farther, but PF kept falling. I stopped using fixed pips and continued until reaching ATR.

※This is historical verification results and does not guarantee future performance.