v007 is the smallest failure in the record, and that is what makes it instructive. It bought and sold 15-minute RSI extremes when the four-hour trend agreed, on BTC, ETH and SOL futures. The unseen test was negative. The live paper record agrees. The strategy is failed and kept on paper for observation only. It sends no signals.
The rule and what it was trying to measure
The rule was specific:
- Signal: a 15-minute RSI extreme.
- Filter: the four-hour trend agrees.
- Universe: BTC, ETH and SOL futures.
- Warmup: 1-day worth of 15-minute bars.
- Stop: 1.2 times the 15-minute ATR from entry.
The intent was to catch an intraday rebound in the direction of a larger trend. The holding period was short. The stop was close. The universe was 3 symbols.
That design creates a narrow measurement. The signal fires on 15-minute bars. The trend filter is four-hour. The stop is scaled to 15-minute ATR. The edge, if it existed, would have to survive futures fees and slippage. The unseen test already charges those costs. So the results are net.
The unseen test: 32 trades and little breadth
The unseen test produced 32 trades at −0.15R net per trade. Profit factor was 0.79. That is a negative net result. It is not a large sample. More importantly, it is not a broad sample. Only 3 symbols were available. A 15-minute signal on 3 coins cannot produce enough independent observations to support a conclusion. The trades are not independent in the way a researcher needs.
Crypto assets often move together. A trend filter on the four-hour chart may select the same regime across all 3 symbols. A 15-minute RSI extreme may cluster in time. So 32 trades can look like more evidence than they are.
The problem is not only the count. It is the structure. The test had little breadth. Breadth is what lets a researcher separate a signal from a coincidence. With 3 symbols, a symbol's behavior can dominate. With a short holding period, the outcome depends on what happens immediately after entry. With futures costs, the strategy must clear a cost hurdle on every trade. The test result says it did not. The negative net result is the measurement. The cause is a thin universe and a fast signal.
The live paper record repeats the result in miniature
The live engine entered BTC and ETH shorts at 02:30 UTC on 19 October 2026. Both were stopped out within 30 minutes. BTC was −1.35R and ETH was −1.22R on those trades. This is paper. It is not live capital. It is not proof. But it is a forward observation that agrees with the unseen test.
The test said the edge was negative net. The paper book then produced the same sign on the entries it took. The stop did what the rule said it would do. The 1.2 times 15-minute ATR stop was hit quickly.
This matters because the live paper record is a separate clock. It is not a re-run of the backtest. It is the engine running forward with no signals sent. The result is small. It is only the BTC and ETH trades. But it is consistent with the test's −0.15R net per trade and profit factor 0.79. The failure did not need a long live record to show its shape. The shape appeared immediately.
Why the failure was structural, not incidental
The strategy was the smallest in the record for a reason. It had 3 symbols. It had a 15-minute signal. It had a short holding period. It had futures costs. Each of those facts limits what the test can prove. Together they make a negative result more likely and a positive result harder to trust.
A 15-minute RSI extreme is a fast event. A four-hour trend filter is slow. The mismatch means the signal can be correct about the trend and still be stopped by 15-minute noise. The 1.2 times 15-minute ATR stop is close enough that normal intraday movement can end the trade before the trend has time to matter.
The costs make this worse. The test already charges fees and slippage on futures. So every trade starts behind. A short holding period gives little time to overcome that. A 3-symbol universe gives little room to diversify the cost drag. The result is a net loss per trade. The profit factor was 0.79, which is a losing net result.
What the record supports and what it does not
The record supports a clear statement. v007 failed its unseen test. The unseen test was negative at −0.15R net per trade with profit factor 0.79. The live paper record agrees, with BTC at −1.35R and ETH at −1.22R after entries at 02:30 UTC on 19 October 2026. The strategy is kept on paper for observation only. It sends no signals.
The record does not support a claim that the strategy works on other symbols, other timeframes, or other market conditions. It does not support a claim that the failure is permanent for all similar rules. It only supports the conclusion that this rule, on 3 symbols, with a 15-minute signal, a four-hour trend filter, a 1.2 times 15-minute ATR stop, and a 1-day warmup, did not produce a positive net result.
The engine still runs on paper so its forward behavior can be compared with the test. That comparison is for observation. It is not a prediction. It is not advice. The status is failed and unproven for any other use.