Conclusion

GateGrid AI won 8 out of 10 closed trades and still finished at -400 yen. That is the whole day in one line, and it is not a comfortable one. The bot kept collecting small wins, but one -729 yen loss cut through the sequence hard enough that I had to pause for a second.

Across the four MT5 bots, the realized total came to -789 yen. The combined win rate was 60.9%, which does not look disastrous on paper, but the payoff ratio was only 0.36. That number says more than the win rate today: the average loss was simply too heavy compared with the average win.

Bot-by-Bot Results

■ GateGrid AI -400 yenRecord: 8W / 2LWin rate: 80.0%Gross profit: +394 yenGross loss: -794 yenPayoff ratio: 0.12Max loss: -729 yen

■ BoundSniper +92 yenRecord: 3W / 1LWin rate: 75.0%Gross profit: +166 yenGross loss: -74 yenPayoff ratio: 0.75Max loss: -74 yenOpen P/L: -118 yen

■ LLMBridgeTrader -318 yenRecord: 1W / 3LWin rate: 25.0%Gross profit: +55 yenGross loss: -379 yenPayoff ratio: 0.44Max loss: -243 yenOpen P/L: -86 yen

■ MLScore GF-T4 GB -163 yenRecord: 2W / 3LWin rate: 40.0%Gross profit: +382 yenGross loss: -557 yenPayoff ratio: 1.03Max loss: -265 yen

■ Total -789 yenRecord: 14W / 9LWin rate: 60.9%Gross profit: +997 yenGross loss: -1,804 yenPayoff ratio: 0.36Max loss: -729 yenOpen P/L: -204 yen

Today’s Theme

The theme today was not entry accuracy. It was exit quality. GateGrid AI had the best win rate of the group, but its payoff ratio was the weakest at 0.12. When a bot needs many small wins to cancel one large loss, the entry filter can look smart while the exit still quietly breaks the day.

This is especially important for the bots where AI or model judgment is involved. I am not only testing whether an LLM can pick BUY or SELL. I am testing whether it can stop holding, switch to closing, or stay out before the position becomes expensive. Today, that boundary was not clean enough.

Bot Analysis

GateGrid AI was the most painful case. The CatBoost and Ollama-style gate structure is built to avoid bad entries, and the 80.0% win rate suggests that the filtering was not useless. But the losses were uneven: one -729 yen exit erased eight wins that totaled only +394 yen. Looking at that -729 yen, I did not think “bad luck” first. I thought the trailing or stop transition probably stayed too loose for the later move, though I do not have full certainty from the daily report alone.

BoundSniper was the only realized winner at +92 yen. Since it is basically a TradingView-to-MT5 execution bot, I read this result more as a check on the upstream signal and execution timing than as an AI judgment test. The open position was -118 yen at the report cutoff, though, so the clean-looking realized profit was already under pressure. That part made the +92 yen feel less safe than it looks.

LLMBridgeTrader was the purest “AI decision” test of the day. It finished at -318 yen with only 1 win and 3 losses, and the open EURUSD position was also negative at -86 yen. Since this bot is allowed to decide not only direction but also OPEN, HOLD, CLOSE, and REVERSE, the weak point today looks like position handling after entry. The -243 yen largest loss is not huge by itself, but in a bot that is supposed to reason about closing, I want to see fewer losses left to reach that size.

MLScore GF-T4 GB ended at -163 yen, but the structure was different from GateGrid AI. Its payoff ratio was 1.03, which is at least balanced: the average win and average loss were nearly the same size. The problem was hit rate, not reward size. The +303 yen take-profit near the end helped, and without it the day would have looked much uglier.

Wrap-Up

The day ended negative, but not all negatives mean the same thing. GateGrid AI needs exit tightening because the win rate is already high but the loss size is not contained. LLMBridgeTrader needs better close-or-hold judgment because the AI layer is being asked to manage the whole plan, not just the entry. BoundSniper needs open-risk monitoring, and MLScore needs more selective entries.

The uncomfortable part is that the headline number was not the total -789 yen. It was 80.0% win rate turning into a losing day. That is the kind of result that makes me trust the log more than the feeling.

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit fxaibotlab.substack.com