Strota Research

First live test of our best long setup: profits held up, the 80% win rate fell to 55%

By Strota Newsroom · 2026-08-24 · How Strota reports

First live test of our best long setup: profits held up, the 80% win rate fell to 55%
researchbacktestwalk-forwardintradayorder-flowwin-rate
The strongest long pattern in our order-flow research kept its profit edge on its first out-of-sample day, earning far more per trade than the average signal. But the win rate, the one metric this project is built around, did not replicate.

The most promising long setup in our intraday order-flow research went into its first live, out-of-sample session carrying a backtest win rate that ran as high as 80%. It came out with 55%. The money side of the edge held up, with these trades earning about 80% more per trade than the ordinary long. But the probability side, the thing the whole project is built around, did not replicate on day one.

The distinction matters because the stated goal of this research is not merely positive expectancy. Plenty of strategies make money on average while losing more often than they win. We are hunting for the opposite: a long setup that wins with high, durable probability. A big average profit attached to a coin-flip hit rate is explicitly not what we are after.

The test itself is deliberately simple. Earlier in-sample work had identified a stack of conditions that, combined, looked extraordinary: an ask-heavy order book, a trade taken in the direction of the day's trend, an entry at the day's high, and a price above the volume-weighted average. In backtesting, that full combination won 80% of the time at +0.57 R per trade. The walk-forward tracker now scores every fresh session against those claims as it happens, with no hindsight involved.

On the first tracked session, a strongly long-driven trend day in June 2026, the baseline was every triggered long signal: 51 setups, a 53% win rate, and +0.199 average R. The ask-heavy subset, the core of the pattern, produced 22 setups at a 55% win rate and +0.357 average R. Layering on the above-VWAP condition changed almost nothing, nudging average R to +0.358 across the same 22 trades.

So the edge is real on one axis. Per trade, ask-heavy longs earned roughly 80% more than the ordinary long, +0.357 against +0.199. On a trending day, the advantage expressed itself as bigger winners rather than more frequent ones. When these trades worked, they ran further.

The win rate is another story. In-sample, ask-heavy longs alone won 66% of the time, and the full stack won 80%. Live, the ask-heavy cohort managed 55%, a lift of just 2pp over the baseline. For a project whose north star is hit rate, that is weak confirmation. The metric that matters most is precisely the one that failed to show up.

The full stack barely traded at all. Out of 51 triggered longs, it fired exactly once, roughly 2% of the day's signals, against something like 5 per day in the backtest. That lone trade won, for +0.14 R, which at n=1 is statistically meaningless. The likely culprit for the scarcity is the bullish-regime condition: most of the session's setups were tagged as neutral regime, and adding that filter collapsed 22 candidates down to one. The uncomfortable implication is that the highest-probability pattern is also the rarest, a sniper setup that cannot carry a whole strategy on its own and whose win rate will take many sessions to verify. There is a genuine tension to resolve here: keep the filter ultra-tight and accept very few high-probability trades, or relax it and take more trades at a somewhat weaker hit rate.

The session also strengthened a separate and increasingly consistent signal: shorts are a drag. In the paper trading book, longs made ₹1,628 at a 47% win rate while shorts lost ₹378, a hole that deepens to ₹929 once a single favourable name is set aside. This was the second straight tracked session in which the short book bled, which makes the negative edge in shorting the most consistent cross-session finding so far.

The honest caveats are substantial. This is one day and 22 trades, and nothing about it is conclusive on its own. The de-duplication method used to score the day differed slightly from the in-sample study, averaging repeated signals per symbol rather than keeping only the first, so the comparison is approximate. The figures also come from idealised fills, not real executions. And the above-VWAP condition could not be tested at all this session: every one of the 22 ask-heavy longs was already above VWAP, so whether that extra confirmation genuinely adds edge remains an open question.

What happens next is patience. Every cohort is now recorded automatically after each close, and the plan is to let the tracker run for at least a month, roughly ten to fifteen sessions, before trusting any win-rate figure. The working assumption is sober: an in-sample win rate that started near 80% slipped to 55% on its very first live day, and the durable number is expected to settle lower still. That is not a failure of the research. It is the research doing exactly what it is for.

Sources and method

This note is based on Strota's own backtest research. Historical results are not forecasts or investment advice; see the editorial standards for methodology and limitations.

More money stories

This story was written by the Strota Newsroom from publicly reported and publicly posted sources, drafted with AI assistance and checked against automated editorial-quality and accuracy gates, with human editorial oversight. Individuals who shared their experience on social media are not identified. See our editorial standards, sourcing and AI-use disclosure. Found an error? Tell us — we correct transparently.