Conclusion

June 17 was not a “win rate day.” The four MT5 bots ended at -1,615 yen, and the strange part is that the bot with a 33.3% win rate, LLMBridgeTrader, lost less than GateGrid AI at 50.0%. I had to look twice at that. The answer was not hidden in the entry count. It was in payoff ratio and the size of the worst exits.

For this pasted record, I am treating it as Day 1 / cumulative -1,615 yen, because no earlier running total was included. The combined final balance across the four MT5 accounts was 169,127 yen after the day’s trades.

Bot Results

■ GateGrid AI -712 yenPair: GBPUSD-Record: 2W / 2LWin rate: 50.0%Gross profit: +84 yenGross loss: -796 yenPayoff ratio: 0.11Max loss: -431 yen

■ BoundSniper Bot -32 yenPair: USDJPY-Record: 2W / 2LWin rate: 50.0%Gross profit: +64 yenGross loss: -96 yenPayoff ratio: 0.67Max loss: -88 yen

■ LLMBridgeTrader -466 yenPair: EURUSD-Record: 3W / 6LWin rate: 33.3%Gross profit: +713 yenGross loss: -1,179 yenPayoff ratio: 1.21Max loss: -245 yen

■ MLScore GF-T4 GB -405 yenPair: GBPJPY-Record: 1W / 2LWin rate: 33.3%Gross profit: +205 yenGross loss: -610 yenPayoff ratio: 0.67Max loss: -353 yen

■ Total -1,615 yenPairs: GBPUSD- / USDJPY- / EURUSD- / GBPJPY-Record: 8W / 12LWin rate: 40.0%Gross profit: +1,066 yenGross loss: -2,681 yenPayoff ratio: 0.60Max loss: -431 yen

Today’s Theme: Win Rate Did Not Protect the Account

GateGrid AI and BoundSniper both ended with a 50% win rate. That sounds acceptable at first glance. But GateGrid’s average win was only 42 yen, while its average loss was 398 yen. Seeing -431 yen as the largest single loss made the whole day feel different. This was not a small miss. It was a payoff structure problem.

LLMBridgeTrader looked worse by win rate. It only won 3 out of 9 exits. But its payoff ratio was 1.21, the only bot above 1.0 today. That did not save the day, but it kept the damage from becoming the worst result on the board. The exits were still noisy, and I do not want to overpraise a losing bot. Still, the shape was healthier than the headline win rate suggested.

Bot-by-Bot Analysis

GateGrid AI

GateGrid AI had the cleanest warning sign of the day. Two winners, two losers, and still -712 yen. The winners were +74 yen and +10 yen. The losers were -365 yen and -431 yen. When the wins are that small, even a decent entry filter cannot carry the system.

This bot is designed to filter entries through CatBoost and Ollama. The design notes include logs like AI_SKIP(sess=NY gate=0.50 base_thr=0.54 adj_thr=0.55) and OLLAMA_HOLD, which show that the bot is supposed to avoid bad setups instead of always trading. Today’s report does not include the full live reasoning trace behind the two GBPUSD exits, and that missing context matters. Without the actual decision text, I can only say this much: the entry side may have passed the filters, but the exit side let the losing grid grow too large compared with the take-profit size.

BoundSniper Bot

BoundSniper was almost flat, but not in the clean way. The price-side closing profit totaled +102 yen, yet swap of -134 yen pulled the final result down to -32 yen. That little reversal is easy to ignore, but it is the kind of cost that slowly changes the personality of a strategy.

This bot is not making AI decisions. It receives TradingView signals and sends them to MT5. The comments were simple execution markers such as BoundSniper OPEN and BoundSniper clos. That simplicity is useful because there is less ambiguity. If the day goes wrong here, the question is usually not “what did the LLM think?” but whether the TradingView rule and holding time still make sense after costs.

LLMBridgeTrader

LLMBridgeTrader is the most interesting loser today. The record was bad on the surface: 3 wins and 6 losses. But the biggest loss was -245 yen, while the biggest win was +399 yen. That is why it lost less than GateGrid despite a lower win rate. I did not expect the “uglier” win rate to produce the more controlled loss.

The execution comments show multiple exits labeled like [sl 1.15813], [tp 1.15665], and [sl 1.14949]. One confusing part is that some [sl] exits were profitable, such as +157 yen. That probably reflects stop adjustment or the way MT5 comments preserve the order label, not a simple “SL equals loss” story. For an LLM-driven bot, this is exactly where the next layer of logs is needed: why did the AI keep holding, close, or allow the stop to sit where it did?

MLScore GF-T4 GB

MLScore GF-T4 GB finished at -405 yen with 1 win and 2 losses. The first close had swap -78 yen and price loss -275 yen, so the net damage of that one exit was -353 yen. That one stung more than the trade count suggests.

The middle trade did work, closing at +205 yen, but the second loss at -257 yen erased it and then some. The structure looked similar to BoundSniper in payoff ratio, but with a larger max loss. It needs either a better loss cap or a reason to avoid the second entry after the first session had already absorbed damage.

Summary

The day was negative, but it was not evenly negative. GateGrid AI looked reasonable by win rate and weak by payoff. LLMBridgeTrader looked weak by win rate and more defensible by loss shape. That is the uncomfortable part of running these bots side by side: the number that feels easiest to understand is not always the number that tells the truth.

Editor’s note to myself: next time, paste the full AI decision logs with confidence, setup type, OPEN/HOLD/CLOSE/REVERSE reason, and exit reason.

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