If you wanted to lose money in this dataset, the easiest way was to trade Bitcoin. If you wanted to make money, the easiest way was to trade the German DAX or US30, both of which the trader almost completely abandoned after the first eight weeks of the dataset. The hardest way to make money, oddly, was to trade gold, which the trader used for 67% of every single click across two years. The strategy was the same. The trader was the same. The instrument is the variable that explains the most.
This is the ninth instalment in our Inside 1,797 Trades series. Posts 1 to 8 were about when the trader was profitable, how long they held, which days they thrived, and which streaks killed them. Post #9 is about what they traded. The picture is the cleanest single-variable explanation in the entire forensic series. Some instruments were the engine. Some were the bleed. The trader could not tell them apart.
The argument of this post is uncomfortable for traders who think of themselves as discretionary stylists. The data does not care about your style. It cares about the expected value of your strategy on the specific instrument you applied it to. Two instruments. Same trader. Same week. Vastly different outcomes. The chart pattern looks similar. The math does not. The principle behind it is covered in our broader risk management framework: edge is a property of the interaction between strategy and instrument, not of either alone.
A Note on This Analysis
Every finding in this series is drawn from a single trader’s 1,797 trades across 12 prop firm accounts. The patterns we describe are real for Trader A, but they are not universal laws. A different trader, with a different strategy, different sleep, different diet, different life circumstances, different time zone, different instruments, or different psychological wiring may produce completely different data. Use these findings as a forensic case study, not a prescription. The most useful application is the method, not the conclusions: pull your own data, run the same splits, and see what your own patterns reveal.
The Full Asset Breakdown
Across all 12 accounts, the trader touched 11 different instruments. The breakdown by trade count and P&L tells the entire story before we add any nuance.
| Asset | Trades | % of all clicks | Win Rate | Total P&L | $ per trade |
|---|---|---|---|---|---|
| Gold (XAUUSD) | 1,204 | 67.0% | 58.5% | −$3,774 | −$3.13 |
| Bitcoin (BTCUSD) | 249 | 13.9% | 40.6% | −$893 | −$3.59 |
| Nasdaq (NDX100) | 78 | 4.3% | 43.6% | −$352 | −$4.52 |
| S&P 500 (SPX500) | 50 | 2.8% | 44.0% | −$21 | −$0.43 |
| DAX (GER30) | 44 | 2.4% | 65.9% | +$333 | +$7.56 |
| Silver (XAGUSD) | 34 | 1.9% | 44.1% | +$1,014 | +$29.82 |
| GBP/JPY | 32 | 1.8% | 46.9% | −$389 | −$12.16 |
| US30 (Dow) | 32 | 1.8% | 87.5% | +$638 | +$19.94 |
| Nikkei (JP225) | 29 | 1.6% | 51.7% | +$12 | +$0.41 |
| Oil (USOUSD) | 26 | 1.4% | 50.0% | +$592 | +$22.79 |
| Ethereum (ETHUSD) | 10 | 0.6% | 50.0% | −$231 | −$23.10 |
Five assets were profitable. Six were not. The trader spent 81% of their clicks on the two assets that lost the most money in absolute terms (gold and Bitcoin). The four most profitable instruments by dollar contribution (silver, US30, oil, DAX) combined accounted for 7.5% of total trade volume.
This is not a strategy problem. The setups that worked on the DAX, silver, US30 and oil were the same setups that failed on Bitcoin, GBP/JPY, and Ethereum. The trader was applying one approach to many different instruments and the instruments were responding differently to that approach. The dataset is the auditor’s note: some instruments suit the strategy, others do not.
The Stacked Counterfactual
What happens if we apply the simplest possible filter to the 1,797 trades? Keep only the four profitable non-gold assets (silver, US30, DAX, oil), and keep gold only during core hours (08:00 to 16:59 server time, the rule from Post #1).
| View | Trades | Win Rate | Total P&L |
|---|---|---|---|
| Actual career (all assets, all hours) | 1,797 | 51.3% | −$3,103 |
| Filtered: profitable assets + gold core hours only | 860 | 62.0% | +$3,945 |
Same trader. Same setups. Same risk per trade. The change is which instruments are on the watchlist and which window is open for gold. The total swing is $7,048, from a $3,103 loss to a $3,945 profit. The filtered trader is taking less than half the trades and is making money on every account they touch.
That number, $7,048, is the dollar cost of asset selection in this dataset. It is larger than the cost of any single behavioural pattern we have measured in this series. Larger than the cost of revenge trading, larger than the cost of the danger zone hours, larger than the cost of missing stop-losses considered alone. The single most expensive choice Trader A made was choosing which markets to trade.
Why Gold Both Worked and Failed
Gold is the asset that did most of the damage in dollar terms, but it also did most of the heavy lifting on the passed accounts. The reason gold appears in both columns of the breakdown is that gold has two completely different personalities depending on what time of day you trade it.
| Gold session | Trades | P&L |
|---|---|---|
| Core hours (00:00 – 16:59 server time) | 832 | +$1,085 |
| Danger zone (17:00 – 23:59 server time) | 372 | −$4,859 |
| Net gold P&L | 1,204 | −$3,774 |
Gold during the trader’s main analytical window was up $1,085 across 832 trades. Gold during the evening sessions, the post-news drift, and the New York PM was down $4,859 across 372 trades. The instrument is fine. The problem is the trader was using the same setup criteria during hours when the criteria did not apply. Asian session liquidity behaves differently to London-NY overlap liquidity. The ICT and Volume Profile concepts that work cleanly during the day rely on participant volume that drops off sharply after 17:00. This is the same finding mapped from a different angle in the trading hours analysis.
This is the deepest insight on asset selection in the dataset: an instrument is not a single object. Gold at 11:00 server time and gold at 21:00 server time are functionally different markets. The trader who treats them as the same market is the trader who breaches.
Bitcoin: The Bleed That Never Stopped
Bitcoin was the trader’s second-most-traded instrument. 249 clicks, 13.9% of all trading activity, $893 lost. Unlike gold, Bitcoin did not have a profitable window. It was losing during core hours (−$435 across 114 trades) and losing during the danger zone (−$444 across 95 trades). The instrument simply did not work for this trader.
What makes the Bitcoin data especially clean is the progression over time. The trader started using Bitcoin in Q3 2024 and was actually profitable in the first quarter of use. Then the relationship inverted.
| Quarter | BTC trades | Win Rate | P&L |
|---|---|---|---|
| 2024 Q3 | 96 | 58.3% | +$482 |
| 2024 Q4 | 69 | 33.3% | −$531 |
| 2025 Q4 | 59 | 25.4% | −$730 |
| 2026 Q1 | 25 | 28.0% | −$114 |
Bitcoin was the trader’s best instrument in the first 90 days of use. By the next quarter the win rate had collapsed from 58% to 33% and never recovered. Across the following 18 months, the trader continued to take roughly 60 Bitcoin trades per quarter at win rates between 25% and 33%. Every Bitcoin quarter after Q3 2024 was a losing quarter.
The instrument changed character. Bitcoin in late 2024 was a different beast to Bitcoin in 2025. Volatility regimes shifted. The 24-hour nature of the asset interacted poorly with the trader’s session-based setup criteria. None of this is unique to Trader A or unique to Bitcoin. Every instrument has personality phases. The data point that matters is that the trader did not respond to the changing data. They kept the Bitcoin allocation through six losing quarters.
The Abandoned Winners
The most painful section of the asset breakdown is not the assets that lost money. It is the assets that made money and were then abandoned.
US30 (Dow Jones). 32 trades. 87.5% win rate. +$638 total. The highest win rate of any asset in the dataset with more than 25 trades. The trader took every US30 trade between September and October 2024. Then they stopped using it entirely. Not one US30 trade was placed in 2025 or 2026.
DAX (GER30). 44 trades. 65.9% win rate. +$333 total. Spread across 14 months but with most of the activity in the first two months. The last DAX trade was in December 2025 and the asset received fewer than 10 trades in the second half of the dataset.
Oil (USOUSD). 26 trades. 50% win rate. +$592 total. The win rate is not stellar but the average winner was $61 against an average loser of $16, so the expectancy was strong. Oil was traded sporadically across the entire dataset, never as a primary instrument.
Silver (XAGUSD). 34 trades. 44.1% win rate. +$1,014 total. The single most profitable instrument by total dollars in the dataset on a per-trade basis. Average winner $98, average loser $24. Held for an average of 98 minutes, the shortest hold time of any profitable asset. Despite this, silver received less than 2% of total trading attention.
Together these four instruments contributed $2,577 of positive expectancy across just 136 trades. The trader knew the setups worked. The accounts where these instruments were used most heavily were the accounts that passed. And yet the dominant allocation went to the asset that lost money on the breached accounts.
The Gravitational Pull of the Familiar
Why did the trader keep going back to gold when other instruments were objectively performing better? The data does not give us the answer directly but it gives us strong circumstantial evidence.
Gold was the first instrument the trader used. The very first trade in the dataset, in August 2024, was a gold buy. The trader’s mental model of “what a setup looks like” was built on gold charts. The ICT levels, the Volume Profile shelves, the killzone overlays, the order block patterns. All of these concepts were learned in the gold context first, and the trader’s pattern-recognition was sharper on gold than on any other market.
The cost of this familiarity is that the trader interpreted Bitcoin charts as if they were gold charts. The same imbalance areas. The same supply zones. The same liquidity grab patterns. Bitcoin does not respect those patterns to the same degree. Bitcoin has its own personality, dominated by 24-hour participation, derivatives liquidations, and a different microstructure. The setups that read cleanly on gold do not necessarily read cleanly on Bitcoin.
This is not an argument that one instrument is universally better than another. It is an argument that a given strategy will have wildly different expectancies across instruments that look superficially similar on the chart. The trader needed to test which instruments their specific approach worked on. The data did the test for them. They just did not change behaviour in response.
The Passed Accounts: How They Allocated
One feature of the dataset deserves direct attention. Every account, passed or breached, was primarily a gold account. Even the passed accounts allocated the majority of their clicks to XAUUSD. So if gold was the problem, how did six accounts still pass?
| Account | Outcome | Gold trades | Gold P&L |
|---|---|---|---|
| 9358 | Passed | 287 | +$1,910 |
| 4224 | Passed | 110 | +$712 |
| 8032 | Passed | 124 | +$753 |
| 1607 | Passed | 150 | +$630 |
| 6608 | Passed | 70 | +$587 |
| 2716 | Passed | 37 | +$202 |
| 7705 | Breached | 102 | −$3,398 |
| 5372 | Breached | 122 | −$1,251 |
| 7492 | Breached | 47 | −$1,435 |
| 2354 | Breached | 33 | −$1,176 |
| 5608 | Breached | 115 | −$474 |
| 8903 | Breached | 7 | −$834 |
Gold worked on the passed accounts. Gold did not work on the breached accounts. The asset itself did not change. What changed was when it was traded, how long positions were held, whether stops were used, and whether the trader stopped after hitting their profit target. The combined damage on gold across the six breached accounts was $8,568. The combined contribution on gold across the six passed accounts was $4,794.
This is the cleanest version of a finding that has been building across the series. Gold is not the problem. Gold-in-the-wrong-hands is the problem. The same instrument, traded by the same person, in the same week, can produce a $1,910 win or a $3,398 loss depending on what session the trader was in, what state of mind they were in, and whether they followed the rules they themselves had written.
The Asset Selection Rule
The branded heuristic emerging from this analysis is what we will call the positive expectancy filter. It is a simple rule with a sharp edge.
An asset earns a place on your trading watchlist only after at least 50 trades on that asset have produced positive expectancy under your strategy. Until that threshold is met, the asset is being tested at small size with strict logging. After that threshold is met, the asset is graduated to live trading at normal size. If at any point the expectancy turns negative across a rolling 50-trade window, the asset is demoted back to test size or removed from the watchlist.
This is mechanical. It is not about gut feel, not about narrative, not about “what is moving today”. It is about whether your specific approach, applied to this specific instrument, generates positive expectancy. If yes, trade it. If no, do not. If it used to but no longer does, accept that the regime has changed and stop.
Applied retroactively to Trader A’s dataset, this rule would have produced the following actions:
- Gold: Pass the 50-trade test in Q3 2024. Graduate to live. Demote on hours-based segmentation rather than instrument level (the danger zone rule from Post #1).
- Silver: Pass the 50-trade test on contribution per trade. Increase allocation rather than treating it as a side instrument.
- US30: Pass the test by the end of October 2024 with the highest win rate in the dataset. Increase allocation rather than abandoning it.
- DAX: Pass the test in late 2024. Maintain allocation rather than letting it taper to zero.
- Oil: Pass the test on per-trade expectancy. Increase allocation.
- Bitcoin: Pass the test in Q3 2024, then fail it in Q4 2024. Demote. Resume only if a fresh 50-trade test returns positive expectancy. The trader’s actual behaviour was the opposite: maintain allocation through six losing quarters.
- Ethereum: Never pass the test. Stop trading after 10 trades show negative expectancy.
- NDX100, GBPJPY: Same as Ethereum. The data showed early. The trader continued anyway.
The mistake is not that the trader did not know which instruments were working. The data was visible in every monthly account summary. The mistake is that the trader did not respond to the data with action.
How to Apply This to Your Own Trading
Run the per-asset breakdown on your last six months of trading. Export your trade history. Group by symbol. Calculate total trades, win rate, total P&L, and average P&L per trade for each asset you have touched. If you trade fewer than five instruments, the analysis takes 20 minutes. If you trade more, an afternoon. The output is one of the most useful single documents in any trading review.
Sort by per-trade expectancy, not total P&L. Total dollars can mislead because volume distorts the picture. An asset that produced $500 across 200 trades has an expectancy of $2.50 per click, which is barely above transaction cost. An asset that produced $300 across 30 trades has an expectancy of $10 per click, which is a real edge. Allocation should follow expectancy, not familiarity.
Identify your “abandoned winners” and your “stubborn losers”. The abandoned winners are instruments where you have positive expectancy but low trade count, because you stopped using them. The stubborn losers are instruments where you have negative expectancy but high trade count, because you keep coming back. The data corrects the bias in both directions: more of the first category, less of the second.
Test new instruments at reduced size first. A new instrument is by definition an instrument where you have not yet established that your strategy produces positive expectancy. Trade it at 30% to 50% of your normal size for the first 50 trades. The reduced size means the test is cheap. After 50 trades, the data tells you whether to graduate the instrument or remove it.
Re-test occasionally. Markets change personality. An instrument that worked for you in 2024 may not work for you in 2026. A quarterly review of per-asset expectancy is a low-cost discipline. The cost of running the analysis is two hours. The cost of not running it, in this dataset, was $7,048.
The Mind/Method/Money Read
Asset selection is a Method-pillar rule in the Mind · Method · Money framework that has direct Money-pillar consequences. The Mind-pillar component is more subtle: it is the tendency to anchor to the asset you first learned on, and the tendency to interpret negative data on that asset as bad luck rather than as a regime signal.
The Method discipline is mechanical: every instrument earns its allocation by producing positive expectancy under your specific approach. The Money discipline is what happens when the instrument loses that status: reduce position size first, then remove from watchlist. The Mind discipline is the hardest: accepting that an instrument you have spent years getting good at may no longer be the best fit for the way you trade. The data is allowed to override your story about yourself as a trader.
This post is also where the series begins to converge. The trading hours rule (Post #1), the stop-loss rule (Post #2), the no-revenge rule (Post #3), the stop-at-peak rule (Post #4), the math-of-expectancy rule (Post #5), the 60-minute floor (Post #6), the Friday tempo rule (Post #7), and the cool-off rule (Post #8) are all Method interventions designed to constrain Mind failures and bound Money damage. The positive expectancy filter is the meta-rule that sits above them: trade the instruments your data says you should trade.
What’s Next in the Series
Post #10 is the pillar synthesis. We have now mapped nine specific rules across the Mind, Method, and Money framework. The next post stacks every rule on top of the dataset at once and answers the question the series has been building toward: what would Trader A’s career P&L have been if every rule we have identified had been applied from day one?
The answer is not a small number. The full stacked counterfactual will be the headline of Post #10, with the breakdown of which rule contributed how much. Some rules are surprisingly small. Some are surprisingly large. The order of magnitude of the combined effect is the punchline of the entire series.
Frequently Asked Questions
Should I trade gold or Bitcoin if I am new to prop firm challenges?
The data from this single trader’s history suggests gold during specific hours can produce positive expectancy, while Bitcoin produced negative expectancy across 18 of the 24 months in the dataset for this trader’s specific strategy. Your data will be different. The actionable advice is to test any instrument at reduced position size for at least 50 trades before allocating normal capital to it, and to remove any instrument from your watchlist if a rolling 50-trade window shows negative expectancy. The choice of instrument should be a data-driven outcome, not a preference.
Why was gold profitable on some accounts and not others when the trader was the same?
The combined gold P&L across the six passed accounts was +$4,794. The combined gold P&L across the six breached accounts was −$8,568. The instrument is identical. What differs is the timing of trades within the session, the position holding behaviour, the use of stop-losses, and whether the trader continued trading after reaching their profit target. Gold itself is neither good nor bad. Gold-traded-during-the-danger-zone is unprofitable for this trader. Gold-traded-in-core-hours is profitable. The same logic applies to most instruments: the question is not whether the instrument has edge, but whether your interaction with the instrument under your specific rules has edge.
How many trades do I need on a new instrument before I trust the data?
50 trades is a reasonable working threshold for retail-frequency traders. Below 50 trades the win rate is too noisy to be informative. Above 50 trades you can begin to assess whether your expectancy on that instrument is positive or negative. Some instruments will declare themselves much earlier (Bitcoin in this dataset showed clear divergence from Q4 2024). Some will require longer. The 50-trade rule is conservative on the upside and lets you exit losing instruments at modest cost while still allowing strong instruments to demonstrate themselves.
What if my favourite instrument has negative expectancy for me?
This is the uncomfortable question the data forces. Two responses are reasonable. First, you can study why the instrument is not working under your approach: maybe a specific session, a specific setup type, or a specific market regime is producing the losses, and a tightened filter would change the picture. Second, if no filter rescues the expectancy, you accept that this instrument is not for you under your current method and you stop trading it. The brain finds this hard because instruments often carry identity (“I am a gold trader”), but the dataset does not care about identity. It cares about expectancy.
Can a market change personality and stop working for my strategy?
Yes. Bitcoin in this dataset is the clearest example. The first 90 days of Bitcoin trading produced a 58% win rate and a $482 profit. The next 18 months produced a 27% win rate and a $1,375 loss across 153 trades. The instrument itself shifted: derivatives flow changed, dominant participant types changed, volatility regimes changed. The strategy did not adapt. The lesson is that an instrument that earned its place on your watchlist three years ago has not necessarily kept that place. Quarterly review of per-instrument expectancy catches the regime change before it becomes a year of negative trading.
