Interview noteThis page restates a public interview in new words. Dollar results, win rates, and payouts are claims made by the speaker or the host. They are not this desk's record, and they are not a result to expect.
Gian Luca, whom the host also introduces as John Luca, is 23 and, on the host's account, a two-time winner of the World Cup the caption calls the Robins Cup. The host's scoreboard is first place in the fourth quarter of 2025 at 58 percent, first in the second quarter of 2026 at 104 percent, and, at the time of the interview, first in the third quarter of 2026 at 254 percent. Luca's own line is that he was unprofitable for more than two years, became profitable after he stopped trying to explain every move, and then won the championship twice, with a third still open. The hour is the process he says produced that, not a feeling about candles. Nothing in it is advice. He tells the viewer not to trust the results because he stated them. Test them.
Stop asking why the move happened
He started discretionary. He wanted the reason for every print, he tried to predict the next one, and he lost. The first switch was that a profitable trader does not need to understand the market. The move has already happened. Studying why it happened spends the time and does not pay the account. He went looking for properties he could test and exploit, and that search made him mechanical. The second switch made him statistical. Math, not a story, is what he used to beat a benchmark and then a championship. A buy or a sell that comes from a feeling is, in his words, a gambler with a charting subscription. Asked whether a struggling retail trader could be taught the process, he says yes. A few friends, he says, became profitable once the method changed, because a known positive expectancy that is skipped is money left on the table. The image is a coin that lands tails 65 percent of the time at one-to-one. You bet tails. Five tails-failures in a row do not scare you, because the edge is known and it is not a belief. He calls that the easier way to trade, and the thing he credits for the championships.
In sample, out of sample, and a 50 percent drop
Most people, he says, try to repair a strategy by adding a concept. The backtest gets prettier. Live, it is a different system. That is overfitting: enough rules that the past fits perfectly, and a perfect backtest is not an account. The work is to keep what the data say works and to drop what the trader is attached to. Validation, for him, is a ten out of ten. Fifty or a hundred trades can be luck. Ten thousand cannot.
Build the rules on one period. He uses 2020 through 2025 as the example and calls it the in-sample set, the place the strategy is constructed. Then run the same rules, unchanged, on data the build never saw. His example is 2026. He slips once and describes the build period as unseen. The host corrects it, and the rest of the hour uses the ordinary split: in sample is where you design, out of sample is the test. What has to survive is not the win rate. It is the Sharpe ratio and the expectancy. An edge is never stable. A move of about 20 to 25 percent in those two metrics is normal to him. A move of more than 50 percent means the edge is probably not solid. Sharpe is the return reported against the drawdown. Two traders can both make 50 percent. One path is a 10 percent gain and a 5 percent drawdown. The other is plus 10, minus 20, plus 30. Only the first looks like a process. A Sharpe above 1.5 is still good. So is 1.4. Below 1 is not very good. A scalper with 10,000 trades in a year can still have a variance of about 100 trades, which means the first hundred can be luck and the edge, if it exists, shows up after.
The probability that a backtest is overfit rises with the number of variants. Ten triggers and ten versions of market structure will produce something with a handsome Sharpe. That does not make it real. Stability of expectancy and Sharpe in the out-of-sample window is the check he likes. He also mentions chopping a backtest into something like ten chunks and asking whether the metrics hold in each. If the out-of-sample numbers are unrelated to the in-sample numbers, there was no strategy. There was a dream, and the rules have to be rebuilt. If they are similar, the rules stay. The validation window is not the place to add a new filter because "2026 feels different."
Degradation has a cutoff he tested rather than inherited. If a metric falls by more than 50 percent from in sample to out of sample, the strategy is not the same one. An expectancy of 0.2 that loses half of itself out of sample fails. A drop of about 30 percent is a warning: lower the risk, or run what he calls the failure test, which is the end of the protocol. The metric can also improve. He will not guess which. He looks.
Win rate is the wrong score, and the regime is a percentile
An 80 percent win rate, by itself, tells him nothing about whether a strategy makes money. He will not analyze a track record with it. Expectancy is the money returned per dollar risked. An expectancy of 0.2 is that average. The spoken illustration next to it is garbled, a "$120" set against a one-dollar example and a $1,000 risk that never gets a clean second number, so the 0.2 is the figure, and the dollar story is not. The same expectancy is also not one number for every market. It depends on the regime. He codes four: low, medium, high, and extreme. A strategy can be negative in one, positive in another, extremely positive in a third, and extremely negative in a fourth. The only version that matters is the one in force when the chart is opened. Knowing it is mandatory.
Raw ATR will not do it. An ATR of 30 is a volatility per session with no rank. He uses the ATR percentile. The settings he tested and kept are an ATR length of 14 and a percentile length of 200, on a free TradingView study under that name. Up to 25 is low. From 26 to 50 is medium. From 51 to 75 is high. From 76 to 100 is extreme. For the strategy in this hour, high volatility is the best regime. That label is an input to the rating system he calls a meta-label. It is not a direction.
Autocorrelation is a property, and he sizes the trade that follows a flip
A coin has no memory. The market can. He separates three cases. Zero autocorrelation means wins and losses are basically random. The strategy can still be profitable. It just has nothing to say about the next trade. Positive autocorrelation means wins cluster and losses cluster. Negative autocorrelation means they tend to invert, which he sees more in mean-reversion systems than in trend systems. In his career he has had one strategy at a negative reading of 4. Most of what he runs is positive.
His scale, which he treats as a range and not a law of statistics, puts 0 to 2 at zero, 3 to 5 at basic, 6 to 8 at strong, and 9 to 10 at a reading so rare he would suspect overfitting, or a holy grail if it were real. The same buckets run in the negative direction. He will not derive the formula on the whiteboard. The comparison is an unconditional win rate against a conditional one. The unconditional number is the ordinary win rate. The conditional number, in the example, uses only the trades that came after a loss. A 40 percent win rate, and two of six after-loss trades, is the kind of pair he compares until it lands on one score. An online calculator or the platform later in the hour will do the arithmetic. What the score is for is the risk.
On a strongly positive system, a win tends to be followed by a win, and a loss by a loss. The trade he wants largest is the one after a loss cluster finally prints a win. The host's reading, which Luca accepts, is that the second trade, the one that comes after that first win, is the big trade. He will still size a version of it when the autocorrelation is only in the basic range, and he will risk less. The trade he calls the worst, and still takes, is the one where the cluster has just flipped the other way and some advantage remains. The trade he never takes is the one where, on his data, there is no advantage left. If he wants a smaller sample made only of the best trades, he skips the one that follows a loser and keeps the one that follows a winner. He is explicit that the worst of the trades he still takes can remain profitable. The probability language around that sentence is messy. The operational split is not: one setup is the best, one is the worst he will still trade, and one has no edge and is passed.
The reason a trend strategy clusters is the regime of trend itself. A win says the market still likes the condition the strategy was built for, so the next risk can be squeezed. When that same strategy starts losing, price is often ranging, and the positive autocorrelation then predicts more losses, because the condition is gone. He will not take the trade after that switch. He has searched backtests for this property on purpose, and he will give up entry variants to keep it. Gainability, once the property is there, is what gets exploited. The property is not the edge. If a reading of 8 degrades by half, to 4, it is still autocorrelation and the risk comes down. If it falls to zero, the strategy may still have an edge and he can no longer use the property until about a hundred more trades revalidate it. A return to 5 means the property is back.
Freshness, and a candle that moves more than it trades
Regime is not only which bucket. It is how old the bucket is. He uses Bayesian online change detection, available as a TradingView study, or calculable by hand if someone wants the pain of it. It plots a regime age from 0 to 100. His own bins are 0 to 30 fresh, 31 to 60 mature, and 61 to 100 old. For the championship strategy, fresh is better. Another strategy can prefer old or mature, and he rechecks every hundred trades. Freshness is one more point in the rating. A regime that is not in the strategy's best age does not veto the trade. It lowers the size.
The trigger is price efficiency, result against effort. He does not care that a candle is an engulfing in the textbook sense. He cares how much volume it took. A 20-period moving average of volume is the line. The condition he wants is a higher price on lower volume: low effort, big result. A second version is volume still above that average but falling while price rises. He names them so the meta-label can ask which volume condition is the best one for the strategy. If the live candle is that condition, it scores a point. Both are objective. The result is a close through the highest wick of the previous candle, not a larger body and not a classic bullish engulfing. The effort is the volume, not the height of the candle. The entry is the close. The stop is the low of that engulfing candle. The target that won his tests, against one-to-one, one-to-three, one-to-four, and one-to-five, is one-to-two.
Five questions, then a fraction of Kelly
The meta-label is five yes-or-no questions. Is the setup in the strategy's best volatility regime? Is it the best trade in the autocorrelation sequence? Is the regime at the freshness that strategy prefers? Is the volume condition the one that raises expectancy? Is the session the one with the higher expectancy? For this strategy the answers he wants are high volatility, the trade after the loss cluster has flipped, a fresh regime, the low-effort or falling-volume case, and New York rather than London. Five yeses is the best setup he can print, and it gets the most risk.
The size comes from the Kelly number, the mathematical full bet for a stated edge. Full Kelly assumes the strategy never changes, and he calls it absurd. He will not hand out one risk percentage for a prop account, a championship, and a personal account. What he used in the championship is a ladder. Five of five is half Kelly. Four is a quarter Kelly. Below that the caption steps through a fifth and a sixth of Kelly, and the exact rung for a three and a two is not cleanly spoken. He still takes a two-of-five, and he still takes a one-of-five. There is an advantage, so the trade is on. The risk is just small.
Previous value, a valid candle, and six Nasdaq ticks
The strategy is momentum, on one timeframe. He analyzes and enters on the 15-minute chart. Changing timeframe changes the object. A trend is a session volume profile whose value area is developing higher than the previous session's. The profile can be the session study on TradingView or the same idea on an order-flow platform. The setup becomes interesting when price returns to the previous session's value area, and only while the structural low that defines the long is still intact.
That low is a valid candle, and valid is measured. A candle is valid when its body, in ticks, is larger than its longest wick. He uses TradingView's measuring tool. A body of four against a wick of three is valid. A body of three against a wick of six is not. On the 15-minute chart he looks at the retracement before the breakout and takes the valid candle closest to that break. If the nearest candle fails the test, he walks further back until one passes. The low of that candle is the structure low. The same rule, inverted, marks highs. Several valid candles can exist. The closest to the break is the one that counts. A break of an invalid candle's low means nothing. On the Nasdaq example the first red candle has a body of 16 and a wick of 56, so it fails. A later candle has a body of 54 and a longest wick around 24, so it stands, and he takes the low of that candle. He does not keep searching once that retracement has one valid candle. A different retracement gets its own search. If price trades through the structure low, the long is dead. A trigger that appears after that break has, on his test, a negative expectancy. He would rather change instrument than take it.
While the low holds, the entry is a break above the previous session's value-area high, with the efficiency condition. The close also has to clear the value-area high by a minimum distance on the same measuring tool. He states 0.6 for forex. For the Nasdaq the number he uses is 6. A close of 3, or of 5, is not a break yet. A close of 24 is. If the first poke is too small, he waits for another candle. If price comes back and then closes far enough, the trade is eligible. Missing the entire run because the distance was never met is acceptable. That distance, together with the volume condition, is the validation. He does not trade the trigger through news of the NFP or CPI kind. He tested the strategy on heavy-news days. It does not perform, so those triggers are skipped. A messy chop that stays above the structure low and eventually meets the rules is still a positive-expectancy trade on the last three years of his data.
The first target is one-to-two. If that hits and there is still room up to the structure high, he will take the same condition again: another close through the prior wick, volume falling or below the average. Continuations are allowed only if the first trade, the one that broke the previous value-area high, was actually taken. No first trade, no add, however far the market runs. The last target is the structure high even when the remaining reward is less than one-to-two. He will not hold through that high. Then he waits for a new session to build a value area and for price to come back to it.
The Nasdaq walk-through is the rules without a pretty candle. Regime age is about 60 bars, mature or old. Price tags the prior value and even closes through a level, and the volume is rising and not below the average, so there is no trade. Taking it would have been a stop. An engulfing measured at 24, with volume falling, is still early because the level itself has not broken. Taking that would have stopped too. A later close of 3, then 5, is under 6. When the break, the engulfing beyond 6, and volume that is rising but still under the average finally arrive together, that is low effort and a big result, and he enters. The target is the structure high. The first one-to-two pays. The next candles have rising volume and no qualifying engulfing, so he waits. A later engulfing of about 25, with volume under the average, is another trade, and it pays in one candle. He cannot take a third. Volume is rising, the print is high effort for the result, and the structure high is about to be reached. Past that high the conditions can be true and the trade is still off.
147R, a 7.9 memory, and a simulation that is not the championship
On his platform the strategy shows 147R over two years, with slippage, costs, and swap approximated in. The rolling Sharpe of the whole record is 2.2. Over the last 60 trades it is 2.05, a decrease of about 7 percent. Expectancy is 0.2, and the 60-trade average is 0.17. The autocorrelation section reads a memory strength of 7.9. After a win, the next trade wins 52.5 percent of the time. Wins follow wins and losses follow losses, which is the property the risk model uses. A Monte Carlo of 10,000 shuffles loses money in 41 of them. He also describes a worst-case slice of 15 that finish negative, inside that same test. He did not walk the shuffle in detail. The point he wants is that the bad paths are countable.
He then runs the same trades through a prop simulator, and he is clear that a championship and a prop account are different games. In the prop game, how often the strategy trades matters more than the expectancy, because a higher frequency with a smaller edge can still pass and get paid, and a championship system dropped into that game leaves money on the table. On 200 evaluations of the host's firm, the core program rather than the rapid one, the pass rate he reads is 40 percent. The best of those paths passes 92 of 200 and fails 108, spending almost $10,000 to acquire $184,000 of drawdown. The average path spends the same and acquires $162,000. The worst passes 71 of 200 and acquires $142,000. The phrase on the screen is drawdown acquired. It is his simulator's output, not a personal brokerage statement.
A separate run assumes 200 funded accounts and one payout, and it ignores the challenge. The spend in the worst of those paths is spoken as 36 and as $35,000, so the dollar figure is not clean. The gain he states, if every payout follows the rules, is $600,000. The run assumes the trader does everything correctly. He trades it manually, and he says manual execution will make errors. The number is what the rules produce, not a promise that a person will.
The software that drew the report is not free. A MetaTrader export or a broker statement can be uploaded, or the same metrics can be built in a journal. He thinks the minimum reproducible version is the whiteboard plus a real backtest. Gain is possible with less. A high percentage in every quarter, which is what a championship scores, requires knowing how hard the edge can be squeezed, including the Monte Carlo. For someone with no programming background he points at two places: statistical-analysis books, one of which he thinks is titled the Eden Mark of model and whose exact name he cannot remember, and his own free videos. The last instruction is the one he wants kept. Do not trust his word. Do not risk money on confidence. Find one objective model, repeat it, and optimize it. Adding a thousand more concepts reduces the result. Mechanical trading, he says, made him profitable. The statistics are what made him a champion.