SD1H4H — Strategy Priority Ranking

← Markets Hub · Best Validated Version → · Recommendations → · All Analyses → · every market/engine/strategy combo, ranked

Strategy Priority Ranking

Ranks every market × engine × strategy combo on the default 1H/4H timeframe pair (the most-tested pair -- the 15min/1H pair isn't included here since it hasn't been through the same depth of walk-forward scrutiny yet), PLUS one additional entry for every validated mechanics/entry/sizing finding on top of a base combo (currently 5, all gold, marked with a blue Variant strategy badge) -- each is its own ranked entry, layered on top of but NOT replacing its base combo's own baseline entry; see FINDINGS.md for the full validation writeup of each. Uses the project's full Jan 2022-present history. This range was extended from the original Aug-2023-only baseline specifically to get a genuine out-of-sample check -- and it caught a real data-quality bug (a bulk historical pull had silently left gaps in 5 of 6 markets' original data, up to 8% of trading days missing for Nasdaq) plus at least one previously-published finding (gold's h4_bias_age_bars filter) that did not survive the extra history. Every number below reflects the corrected, extended dataset. Ranking score = (Monte Carlo probability of profit − risk of ruin), scaled down for samples under 200 trades, with any strategy that doesn't survive real costs (Vantage FX / Bybit, per the Trading Costs page) automatically ranked below every strategy that does. This is a heuristic for prioritizing attention, not a statistical guarantee -- read the per-entry notes, especially the sample-size caveats in Tier 2.

Tier 1 — Priority (survives costs, 200+ trades, high Monte Carlo confidence)

#1 Gold (XAU/USD) — Original — Strategy B — shorter stop lookback (ENGINE=lookback_10) Priority Variant strategy
744 trades 23.92% win rate net avg +0.213R/trade net total +158.6R MC probability of profit 92.7% MC risk of ruin 4.6% median drawdown 40.6%
WhyShortening the swing-based stop's lookback window from 30 to 10 bars improves gold Strategy B net of costs: net avg R +0.170 -> +0.213/trade, net total +128.7R -> +158.6R, Monte Carlo probability of profit 90.0% -> 92.7%, risk of ruin 5.4% -> 4.6%. 3/3 walk-forward folds held. A genuine mechanics improvement, not an entry filter -- trade count barely moves (755 -> 744).
WeaknessThe improvement is real but incremental against an already-strong baseline -- this doesn't change the underlying regime-dependence (directional-bias-driven edge) documented on the base Gold B row above, it just cuts losing trades a little sooner.
ImproveDoes not stack with the no_event_bias variant (Finding 4, also Strategy B) -- use one or the other, not both. Not yet tested on other markets or the variant structure engine.
Full breakdown →
#2 Gold (XAU/USD) — Original — Strategy A — Tokyo/Asian session filter Priority Variant strategy
326 trades 32.21% win rate net avg +0.142R/trade net total +46.4R MC probability of profit 85.3% MC risk of ruin 0.3% median drawdown 22.8%
WhyRestricting gold Strategy A entries to the Tokyo/Asian session (00:00-07:00 UTC) turns the market's thinnest edge into its strongest: 326 of the base combo's 1,125 trades, net avg R +0.142/trade (vs. +0.033 unfiltered), 85.3% Monte Carlo probability of profit vs. 64.0%, risk of ruin down to 0.35% from 13.85%. 3/3 walk-forward folds held, net of costs.
WeaknessFewer trades (326 vs. 1,125) means less total opportunity even though each one is better -- net total R is +46.4R vs. the baseline's +36.7R, a real but not dramatic improvement in absolute terms alongside the much better risk profile.
ImproveLayering conviction-scaled sizing on top of this same signal (size up in the Tokyo window instead of excluding everything else) pushes the risk profile further -- see the conviction-scaled-sizing row below, built directly on this one.
TestedThis is a pre-registered result (from the project's own by_session categorical check), not a discovery from the multi-market sweep below. Specific to gold Strategy A -- Strategy B's Asian-session result on the same market is close (2/3 folds) but not robust, and rolling the identical check out across all 6 markets × both strategies × 5 sessions (60 tests) found 7 "robust" hits, statistically indistinguishable from the ~7.5 expected by pure chance at that many comparisons -- only this gold/A result is legitimate.
Full breakdown →
#3 Gold (XAU/USD) — Original — Strategy B Priority
755 trades 24.37% win rate net avg +0.170R/trade net total +128.7R MC probability of profit 90.0% MC risk of ruin 5.4% median drawdown 40.4%
WhyStill the flagship result, on more and cleaner data: 755 trades (2022-2026), 90.0% Monte Carlo probability of profit, 5.4% risk of ruin, net +128.7R (+0.170R/trade) that comfortably clears its own cost (0.050R/trade).
WeaknessThe edge is NOT evenly distributed across the full history -- the earlier out-of-sample check (run before this extension) found gold's edge concentrated in the Aug 2023+ period specifically, with a flat-to-negative 2022-mid 2023 stretch. The blended net avg R above (+0.170R) already reflects that weaker earlier period pulling the number down from the Aug-2023-only figure (+0.450R) -- this is a real, current-data view, not a stale one, but it means the edge is regime-dependent, not uniform across 4.5 years.
ImproveThe previously-touted h4_bias_age_bars filter (3/3 robust on the shorter Aug-2023-only sample, the project's headline finding at the time) did NOT survive walk-forward on the full 2022-2026 history -- it no longer even passes the single-split screen for this strategy. Treat that as retracted. Nothing has replaced it yet for Strategy B specifically -- see the walk-forward table for what IS currently tested and surviving on this market (mostly Strategy A's filters, below).
Why does this vary by period?Investigated directly: is the period-dependence explained by generic market "trendiness"? Built two market-level trend proxies (rolling 20-day Kaufman efficiency ratio, 60-day return/volatility ratio) and correlated them against quarterly strategy performance -- no significant relationship (p=0.07-0.95, sign even inconsistent between strategies). The real driver is directional alignment: splitting trades by direction and correlating against that quarter's ACTUAL gold return shows longs win in quarters gold rallies and shorts win in quarters it falls, strongly and significantly (long rho=+0.88 p<0.0001, short rho=-0.78 p=0.0001, n=19 quarters). This strategy is fundamentally a directional-structure-follower with no edge independent of getting the multi-month bias right -- gold's 2023-2025 sustained bull run is why the edge looked so strong in the previously-published Aug-2023+ window. Checked whether this is forecastable (prior-quarter return predicting next-quarter performance): no (p=0.28-0.96) -- gold's quarter-to-quarter direction isn't persistent enough to filter on with a simple trailing-momentum signal. This is the risk profile of the strategy, not a fixable modeling gap.
Full breakdown →
#4 Gold (XAU/USD) — Original — Strategy B — use_event_bias=False (ENGINE=no_event_bias) Priority Variant strategy
674 trades 23.0% win rate net avg +0.211R/trade net total +142.2R MC probability of profit 89.8% MC risk of ruin 6.8% median drawdown 40.2%
WhyDisabling the "event bias" mechanism in structure detection (a single self-referential bias now feeds both BOS/CHoCH classification and the HTF filter, instead of a separate one) improves gold Strategy B net of costs: net avg R +0.170 -> +0.211/trade, net total +128.7R -> +142.2R. 3/3 walk-forward folds held. Trade count drops (755 -> 674) since the classification itself changes, not just the stop.
WeaknessNot a uniform improvement across every risk metric -- Monte Carlo probability of profit is essentially flat (89.95% -> 89.85%) and risk of ruin ticks UP slightly (5.4% -> 6.8%) even as avg R and total R both improve. An avg-R/magnitude improvement, not a risk-profile one.
ImproveDoes not stack with the lookback_10 variant (Finding 2, also Strategy B) -- use one or the other, not both. The variant (LuxAlgo) structure engine already uses the equivalent of use_event_bias=False by design, so this finding doesn't port there directly -- would need its own consideration.
Full breakdown →
#5 Gold (XAU/USD) — Original — Strategy A — ATR-based stop (ENGINE=atrstop_15) Priority Variant strategy
1156 trades 30.19% win rate net avg +0.094R/trade net total +108.9R MC probability of profit 79.2% MC risk of ruin 12.2% median drawdown 44.0%
WhyReplacing the swing-based stop entirely with a volatility-normalized one (entry ± 1.5×ATR14) improves gold Strategy A net of costs: net avg R +0.033 -> +0.094/trade, net total +36.7R -> +108.9R, Monte Carlo probability of profit 64.0% -> 79.2%. 3/3 walk-forward folds held. Trade count rises slightly (1,125 -> 1,156) since the ATR stop occasionally sits differently than the nearest opposite swing.
WeaknessNot a uniform win on every metric -- median max drawdown ticks up very slightly (43.35% -> 44.02%) even as probability of profit and total return both improve substantially. Worth stating plainly rather than implying a clean sweep.
ImproveNot yet checked in combination with the Tokyo/Asian session filter (Finding 1) on this same combo -- both are gold Strategy A mechanics, unverified whether they stack. Not yet tested on other markets or the variant structure engine.
Full breakdown →
#6 Nasdaq 100 (US Tech 100) — Original — Strategy A Priority
1170 trades 34.02% win rate net avg +0.053R/trade net total +61.7R MC probability of profit 74.0% MC risk of ruin 10.2% median drawdown 42.0%
WhyMoved up to #2 with the corrected data: 1,170 trades, net +61.7R (+0.053R/trade), 74.0% probability of profit, 10.3% risk of ruin. This market was the WORST-affected by the data-quality bug (86 of ~1,119 original weekdays were silently missing) -- with clean data plus 2022-2023 history added back in, its edge is not just intact but stronger and better-evidenced than the original (buggy) Aug-2023-only publish showed.
WeaknessMedian max drawdown is high (42.0%) relative to the edge size -- this is a strategy with real variance around a real edge, not a smooth equity curve.
Improvetrail_chop_1h (trailing 1H structure-flip count) is walk-forward robust 3/3 on this exact market/strategy/pair with the extended data -- a genuinely new, previously-undetected finding (it wasn't even tested on the shorter sample). Worth building into an actual entry filter next.
TestedThe 3 gold mechanics findings (ENGINE=lookback_10/atrstop_15/no_event_bias) were rolled out here too (2026-08-26): atrstop_15 (this market's pre-registered pairing for A) net avg R +0.053 -> +0.011, only 1/3 folds. As a secondary (non-pre-registered) check, lookback_10 on A actually hit 3/3 folds (+0.053 -> +0.056) -- but this was one of 3 hits out of a 30-test multiple-comparisons sweep where ~3.75 are expected by chance alone, so it is not treated as a validated finding. See All Analyses for the full rollout writeup.
Full breakdown →
#7 Gold (XAU/USD) — Original — Strategy A — conviction-scaled sizing on the Tokyo signal Priority Variant strategy
1125 trades 32.09% win rate net avg +0.033R/trade net total +36.7R MC probability of profit 73.1% MC risk of ruin 11.9% median drawdown 43.0%
WhyInstead of Finding 1's all-or-nothing entry filter, sizing UP Tokyo-session trades (1.5×) and DOWN non-Tokyo trades (0.75×) on the full 1,125-trade set improves account-level risk outcomes: Monte Carlo probability of profit 61.15% -> 73.1%, risk of ruin 16.65% -> 11.95%, median final balance ($1,000 start) $1,173 -> $1,484. Avg R is unchanged (+0.033) by construction -- this is purely a sizing/compounding effect, not a new trade-level edge.
WeaknessNot genuinely a sixth independent finding -- it's Finding 1's signal, sized instead of filtered, and inherits every one of that finding's scope caveats (gold Strategy A only, doesn't generalize to other markets). The specific multiplier pair (1.5×/0.75×) was picked illustratively, not fit to an optimum.
ImproveThe 5-pair sensitivity check found the effect requires genuine reallocation (boost AND reduce) -- boosting Tokyo trades alone, without reducing others, makes risk of ruin worse than flat sizing (a real trap worth remembering before trying a size-up-only variant).
Full breakdown →
#8 Gold (XAU/USD) — Original — Strategy A Priority
1125 trades 32.09% win rate net avg +0.033R/trade net total +36.7R MC probability of profit 64.0% MC risk of ruin 13.8% median drawdown 43.4%
Why755 -> 1,125 trades with the extension, net +36.7R (+0.033R/trade), 64.0% probability of profit, 13.9% risk of ruin -- still clears costs but the thinnest edge of the top three, same as before the extension.
WeaknessSame regime-concentration issue as Strategy B on this market (weaker/negative pre-2023 period blended into the total), and the highest median drawdown of the Tier 1 group (43.4%) against the smallest edge -- the worst risk-adjusted profile of the four.
Improveby_session has now been tested and validated as an actual entry filter -- see "Tested" below. trail_atr_pct (above-median trailing volatility favored) remains untested as a real filter, worth doing next. by_hour's walk-forward edge group was hour=4 UTC specifically, which sits inside the now-validated Tokyo/Asian session window (00-07 UTC) -- likely the same signal, not an independent one, so don't treat it as a second finding without checking for overlap first.
TestedThe by_session flag was formalized into a real entry filter and tested net of costs with a proper 3-fold walk-forward: restricting entries to the Tokyo/Asian session (00:00-07:00 UTC) only holds up in all 3 folds -- 326 trades (vs. 1,125 unrestricted), net avg R +0.142/trade (vs. +0.033 baseline), net total +46.4R, profit factor 1.24. London, NY, the London/NY overlap, and off-hours sessions all fail the same check. This is currently the single strongest validated filter finding in the whole project. It is specific to gold Strategy A, not a general "trade the Tokyo session" rule -- the identical check rolled out across all 6 markets × both strategies × 5 sessions (60 tests total) produced 7 "robust" hits, statistically indistinguishable from the ~7.5 expected by pure chance alone at that many comparisons. Only this gold/A result came from a single pre-registered test (the by_session flag above, confirmed net of costs) rather than that blind sweep -- it does NOT generalize to other markets or to Strategy B on gold itself (2/3 folds, not robust).
Why does this vary by period?Same investigation and same answer as Strategy B on this market (see its regime note in full): generic market trendiness does not explain the period-dependence, but directional alignment with gold's realized quarterly move does (both strategies share the same entry logic and direction, only the exit differs) -- and that realized direction is not forecastable from simple trailing momentum. Same conclusion applies to A: this is a bet on gold's multi-month bias, not a portable structural edge.
Full breakdown →
#9 Nasdaq 100 (US Tech 100) — Original — Strategy B Priority
823 trades 24.3% win rate net avg +0.007R/trade net total +5.7R MC probability of profit 37.3% MC risk of ruin 43.4% median drawdown 57.8%
WhyBarely survives: 823 trades, net +5.7R (+0.007R/trade) -- technically Tier 1 by trade count, but Monte Carlo probability of profit is only 37.3% and risk of ruin is 43.4%, both meaningfully worse than any other Tier 1 entry. This is the strategy the data extension was LEAST kind to -- its edge nearly vanished once diluted across the full 2022-2026 history.
WeaknessThis is functionally a coin-flip-or-worse strategy now. The avg cost (0.035R) is larger than half the gross edge (0.041R) -- there is almost no margin left to absorb any modeling error before this flips net negative for real.
ImproveNo regime/entry filter is currently walk-forward robust for this exact strategy -- none passed even the single-split screen strongly enough to test. Given how thin the surviving edge is, the honest next step is deciding whether to keep trading this at all rather than looking for a filter to rescue it; Strategy A on the same market is a meaningfully better risk-adjusted alternative.
TestedThe 3 gold mechanics findings were rolled out here too (2026-08-26): lookback_10 (pre-registered for B) net avg R +0.007 -> -0.006 (flips to failing costs), 1/3 folds; no_event_bias (also pre-registered for B) +0.007 -> +0.053, the closest near-miss found in this rollout at 2/3 folds -- still short of the bar. Neither treated as validated. See All Analyses for the full rollout writeup.
Full breakdown →

Tier 2 — Promising but thin sample (survives costs, under 200 trades)

#10 Bitcoin (BTC/USD) — Variant (LuxAlgo) — Strategy A Promising — thin sample
99 trades 22.22% win rate net avg +0.512R/trade net total +50.7R MC probability of profit 80.0% MC risk of ruin 0.1% median drawdown 21.8%
WhyThe variant engine's wider stops (~2% of price vs ~1% for the original engine) make it structurally less exposed to Bybit's fee mechanics that sink the original-engine BTC strategies outright (see Tier 3). 99 trades, net +0.512R/trade, 80.0% probability of profit -- the strongest non-gold, non-Nasdaq result on the page, though still under the 200-trade bar.
WeaknessSample size is the core issue -- see the shared note below.
ImproveAccumulate more history (live-forward, since the backtest window is now the full available Dukascopy range) before sizing any of these up. Variant-engine samples grew with the 2022 extension (18-61 trades before, 34-99 now) but are still under the 200-trade Tier 1 bar and mostly still too thin for a reliable walk-forward check. Treat the headline numbers as "consistent with a real edge," not confirmed. This exact wall was hit again (2026-08-26) trying to roll the gold lookback_10/atrstop_15 stop mechanics out to the variant engine on every market: 20 of 24 walk-forward checks had zero complete folds, so no conclusion (positive or negative) is possible on the variant engine with the current data volume -- see All Analyses.
Full breakdown →
#11 Bitcoin (BTC/USD) — Variant (LuxAlgo) — Strategy B Promising — thin sample
73 trades 17.81% win rate net avg +0.928R/trade net total +67.8R MC probability of profit 88.4% MC risk of ruin 0.0% median drawdown 18.3%
WhySame market, same wider-stop mechanism, even stronger: 73 trades, net +0.928R/trade, the single highest per-trade edge outside the flagship gold/Nasdaq entries. Still a small sample for a number this large.
WeaknessSample size is the core issue -- see the shared note below.
ImproveAccumulate more history (live-forward, since the backtest window is now the full available Dukascopy range) before sizing any of these up. Variant-engine samples grew with the 2022 extension (18-61 trades before, 34-99 now) but are still under the 200-trade Tier 1 bar and mostly still too thin for a reliable walk-forward check. Treat the headline numbers as "consistent with a real edge," not confirmed. This exact wall was hit again (2026-08-26) trying to roll the gold lookback_10/atrstop_15 stop mechanics out to the variant engine on every market: 20 of 24 walk-forward checks had zero complete folds, so no conclusion (positive or negative) is possible on the variant engine with the current data volume -- see All Analyses.
Full breakdown →
#12 Euro / US Dollar (EUR/USD) — Variant (LuxAlgo) — Strategy A Promising — thin sample
85 trades 25.88% win rate net avg +0.107R/trade net total +9.1R MC probability of profit 58.5% MC risk of ruin 0.0% median drawdown 17.6%
WhyNotable mainly because EUR/USD's original engine is a clear structural loser under both strategies (see Tier 3) -- the variant engine flips this specific combo positive. 85 trades.
WeaknessSample size is the core issue -- see the shared note below.
ImproveAccumulate more history (live-forward, since the backtest window is now the full available Dukascopy range) before sizing any of these up. Variant-engine samples grew with the 2022 extension (18-61 trades before, 34-99 now) but are still under the 200-trade Tier 1 bar and mostly still too thin for a reliable walk-forward check. Treat the headline numbers as "consistent with a real edge," not confirmed. This exact wall was hit again (2026-08-26) trying to roll the gold lookback_10/atrstop_15 stop mechanics out to the variant engine on every market: 20 of 24 walk-forward checks had zero complete folds, so no conclusion (positive or negative) is possible on the variant engine with the current data volume -- see All Analyses.
Full breakdown →
#13 Nasdaq 100 (US Tech 100) — Variant (LuxAlgo) — Strategy A Promising — thin sample
87 trades 22.99% win rate net avg +0.079R/trade net total +6.9R MC probability of profit 56.1% MC risk of ruin 0.0% median drawdown 18.9%
WhySame direction as the original engine's (Tier 1) Nasdaq A result -- independent corroboration from a different structure definition on the same market, though only 87 trades here.
WeaknessSample size is the core issue -- see the shared note below.
ImproveAccumulate more history (live-forward, since the backtest window is now the full available Dukascopy range) before sizing any of these up. Variant-engine samples grew with the 2022 extension (18-61 trades before, 34-99 now) but are still under the 200-trade Tier 1 bar and mostly still too thin for a reliable walk-forward check. Treat the headline numbers as "consistent with a real edge," not confirmed. This exact wall was hit again (2026-08-26) trying to roll the gold lookback_10/atrstop_15 stop mechanics out to the variant engine on every market: 20 of 24 walk-forward checks had zero complete folds, so no conclusion (positive or negative) is possible on the variant engine with the current data volume -- see All Analyses.
Full breakdown →
#14 Gold (XAU/USD) — Variant (LuxAlgo) — Strategy A Promising — thin sample
80 trades 22.5% win rate net avg +0.060R/trade net total +4.8R MC probability of profit 51.7% MC risk of ruin 0.0% median drawdown 19.1%
WhySame market as the top two Tier 1 results, different (fractal-pivot) structure definition -- gold looking favorable under both engines is a meaningful cross-check. 80 trades.
WeaknessSample size is the core issue -- see the shared note below.
ImproveAccumulate more history (live-forward, since the backtest window is now the full available Dukascopy range) before sizing any of these up. Variant-engine samples grew with the 2022 extension (18-61 trades before, 34-99 now) but are still under the 200-trade Tier 1 bar and mostly still too thin for a reliable walk-forward check. Treat the headline numbers as "consistent with a real edge," not confirmed. This exact wall was hit again (2026-08-26) trying to roll the gold lookback_10/atrstop_15 stop mechanics out to the variant engine on every market: 20 of 24 walk-forward checks had zero complete folds, so no conclusion (positive or negative) is possible on the variant engine with the current data volume -- see All Analyses.
Full breakdown →
#15 Nasdaq 100 (US Tech 100) — Variant (LuxAlgo) — Strategy B Promising — thin sample
47 trades 19.15% win rate net avg +0.712R/trade net total +33.5R MC probability of profit 86.0% MC risk of ruin 0.0% median drawdown 12.8%
WhyStrong Monte Carlo probability of profit (86.1%) on 47 trades -- promising but still thin.
WeaknessSample size is the core issue -- see the shared note below.
ImproveAccumulate more history (live-forward, since the backtest window is now the full available Dukascopy range) before sizing any of these up. Variant-engine samples grew with the 2022 extension (18-61 trades before, 34-99 now) but are still under the 200-trade Tier 1 bar and mostly still too thin for a reliable walk-forward check. Treat the headline numbers as "consistent with a real edge," not confirmed. This exact wall was hit again (2026-08-26) trying to roll the gold lookback_10/atrstop_15 stop mechanics out to the variant engine on every market: 20 of 24 walk-forward checks had zero complete folds, so no conclusion (positive or negative) is possible on the variant engine with the current data volume -- see All Analyses.
Full breakdown →
#16 Silver (XAG/USD) — Variant (LuxAlgo) — Strategy A Promising — thin sample
81 trades 24.69% win rate net avg +0.035R/trade net total +2.8R MC probability of profit 46.4% MC risk of ruin 0.0% median drawdown 20.7%
WhyNotable mainly because silver's ORIGINAL engine is now the single worst performer on this entire page (-398.5R net, see Tier 3). A sharp divergence between structure definitions on the same market -- worth understanding before trusting either version. Also has the highest avg cost (0.148R) in this tier, silver's low-unit-price cost mechanics apply here too, just not enough to erase the larger gross edge.
WeaknessSample size is the core issue -- see the shared note below.
ImproveAccumulate more history (live-forward, since the backtest window is now the full available Dukascopy range) before sizing any of these up. Variant-engine samples grew with the 2022 extension (18-61 trades before, 34-99 now) but are still under the 200-trade Tier 1 bar and mostly still too thin for a reliable walk-forward check. Treat the headline numbers as "consistent with a real edge," not confirmed. This exact wall was hit again (2026-08-26) trying to roll the gold lookback_10/atrstop_15 stop mechanics out to the variant engine on every market: 20 of 24 walk-forward checks had zero complete folds, so no conclusion (positive or negative) is possible on the variant engine with the current data volume -- see All Analyses.
Full breakdown →
#17 Euro / US Dollar (EUR/USD) — Variant (LuxAlgo) — Strategy B Promising — thin sample
58 trades 17.24% win rate net avg +0.317R/trade net total +18.4R MC probability of profit 62.0% MC risk of ruin 0.0% median drawdown 17.8%
WhyAnother EUR/USD variant-engine flip versus its structurally-losing original engine (Tier 3) -- 58 trades, net +0.317R/trade.
WeaknessSample size is the core issue -- see the shared note below.
ImproveAccumulate more history (live-forward, since the backtest window is now the full available Dukascopy range) before sizing any of these up. Variant-engine samples grew with the 2022 extension (18-61 trades before, 34-99 now) but are still under the 200-trade Tier 1 bar and mostly still too thin for a reliable walk-forward check. Treat the headline numbers as "consistent with a real edge," not confirmed. This exact wall was hit again (2026-08-26) trying to roll the gold lookback_10/atrstop_15 stop mechanics out to the variant engine on every market: 20 of 24 walk-forward checks had zero complete folds, so no conclusion (positive or negative) is possible on the variant engine with the current data volume -- see All Analyses.
Full breakdown →
#18 Gold (XAU/USD) — Variant (LuxAlgo) — Strategy B Promising — thin sample
34 trades 23.53% win rate net avg +1.825R/trade net total +62.0R MC probability of profit 90.1% MC risk of ruin 0.0% median drawdown 10.3%
WhyOnly 34 trades even after the 2022 extension (up from 18) -- still too few for the Monte Carlo simulation to run reliably, though it no longer gets silently skipped like it did before. The average R (+1.825R) is still almost certainly a small number of large trend trades, not a repeatable edge.
WeaknessSample size is the core issue -- see the shared note below.
ImproveAccumulate more history (live-forward, since the backtest window is now the full available Dukascopy range) before sizing any of these up. Variant-engine samples grew with the 2022 extension (18-61 trades before, 34-99 now) but are still under the 200-trade Tier 1 bar and mostly still too thin for a reliable walk-forward check. Treat the headline numbers as "consistent with a real edge," not confirmed. This exact wall was hit again (2026-08-26) trying to roll the gold lookback_10/atrstop_15 stop mechanics out to the variant engine on every market: 20 of 24 walk-forward checks had zero complete folds, so no conclusion (positive or negative) is possible on the variant engine with the current data volume -- see All Analyses.
Full breakdown →
#19 Silver (XAG/USD) — Variant (LuxAlgo) — Strategy B Promising — thin sample
47 trades 12.77% win rate net avg +0.548R/trade net total +25.7R MC probability of profit 61.0% MC risk of ruin 0.0% median drawdown 21.2%
Why47 trades, net +0.548R/trade, high avg cost (0.148R) same as Strategy A on this market -- survives despite the cost drag, unlike silver's original engine.
WeaknessSample size is the core issue -- see the shared note below.
ImproveAccumulate more history (live-forward, since the backtest window is now the full available Dukascopy range) before sizing any of these up. Variant-engine samples grew with the 2022 extension (18-61 trades before, 34-99 now) but are still under the 200-trade Tier 1 bar and mostly still too thin for a reliable walk-forward check. Treat the headline numbers as "consistent with a real edge," not confirmed. This exact wall was hit again (2026-08-26) trying to roll the gold lookback_10/atrstop_15 stop mechanics out to the variant engine on every market: 20 of 24 walk-forward checks had zero complete folds, so no conclusion (positive or negative) is possible on the variant engine with the current data volume -- see All Analyses.
Full breakdown →

Tier 3 — Does not survive real costs

#20 S&P 500 (US 500) — Original — Strategy A Does not survive costs
1197 trades 31.83% win rate net avg -0.042R/trade net total -50.0R MC probability of profit 12.8% MC risk of ruin 68.5% median drawdown 68.0%
WhyCost-driven failure.
WeaknessGross was still positive (+0.045R/trade) but real Vantage index spread costs (0.086R/trade) flip it net negative (-50.0R). Same thin-edge-vs-cost mechanism as before the extension, now on more data.
ImproveA broker with a tighter S&P 500 index spread than Vantage's Raw ECN could plausibly flip this back positive -- worth shopping the actual spread before writing this market off. The gold mechanics rollout (2026-08-26) also tested atrstop_15 here (pre-registered for A): net avg R -0.042 -> -0.041, essentially no change, 1/3 folds. See All Analyses.
Full breakdown →
#21 Bitcoin (BTC/USD) — Original — Strategy B Does not survive costs
1129 trades 21.26% win rate net avg -0.049R/trade net total -55.1R MC probability of profit 13.3% MC risk of ruin 81.7% median drawdown 82.0%
WhyCost-driven failure.
WeaknessGross was solidly positive (+0.127R/trade) but Bybit's taker-fee-vs-tight-stop mechanics (0.176R avg cost, the highest of any original-engine combo) erase it (-55.1R net). Got worse with the extension, not better.
ImproveTrade BTC via the variant engine's wider-stop version instead (Tier 2, +0.512R and +0.928R/trade) -- same underlying market, structurally less fee-sensitive entry/stop logic. The gold mechanics rollout (2026-08-26) tested lookback_10 and no_event_bias here (both pre-registered for B): lookback_10 -0.049 -> -0.049 (no effect), 0/3 folds; no_event_bias -0.049 -> +0.033 (flips to surviving costs!) but only 2/3 folds -- the single closest near-miss in the whole rollout, still not validated. See All Analyses.
Full breakdown →
#22 Bitcoin (BTC/USD) — Original — Strategy A Does not survive costs
1693 trades 29.65% win rate net avg -0.070R/trade net total -118.7R MC probability of profit 4.3% MC risk of ruin 93.5% median drawdown 88.6%
WhyCost-driven failure.
WeaknessSame BTC fee-vs-tight-stop mechanism as Strategy B on this market (0.169R avg cost against a 0.099R gross edge).
ImproveSame fix as Strategy B -- the variant engine's version of this market is the better option, not a parameter tweak to this one. atrstop_15 (pre-registered for A) was also tested here (2026-08-26): -0.070 -> -0.059, 1/3 folds. See All Analyses.
Full breakdown →
#23 S&P 500 (US 500) — Original — Strategy B Does not survive costs
804 trades 25.25% win rate net avg -0.089R/trade net total -71.7R MC probability of profit 7.5% MC risk of ruin 76.3% median drawdown 70.9%
WhyStructural failure.
WeaknessGross was already essentially flat-to-negative (-0.005R/trade) before costs -- unlike Strategy A on the same market, this isn't primarily a cost problem.
ImproveNot cost-fixable -- would need a different entry/exit rule, not a cheaper broker. lookback_10 and no_event_bias (both pre-registered for B) were tested here (2026-08-26): lookback_10 -0.089 -> -0.104 (worse), 0/3 folds; no_event_bias -0.089 -> -0.221 (worse), 1/3 folds. See All Analyses.
Full breakdown →
#24 S&P 500 (US 500) — Variant (LuxAlgo) — Strategy B Does not survive costs
58 trades 8.62% win rate net avg -0.110R/trade net total -6.3R MC probability of profit 31.8% MC risk of ruin 0.0% median drawdown 26.2%
WhyFlipped by the data extension.
WeaknessThis combo was net POSITIVE (+31.1R) on the shorter Aug-2023-only sample and is now net NEGATIVE (-6.3R, -0.110R/trade) on the full 2022-2026 history -- a direct illustration of why the extension mattered: a result that looked like a Tier 2 survivor was actually period-dependent, same pattern as gold's filter but severe enough here to flip the sign entirely.
ImproveDo not trade this combo on the strength of the old number -- it no longer holds with more data.
Full breakdown →
#25 Euro / US Dollar (EUR/USD) — Original — Strategy A Does not survive costs
1263 trades 29.77% win rate net avg -0.136R/trade net total -171.5R MC probability of profit 0.0% MC risk of ruin 99.2% median drawdown 86.9%
WhyStructural failure.
WeaknessGross already negative (-0.071R/trade), consistent with this project's finding (now confirmed on more data) that EUR/USD shows no edge at all under the original engine.
ImproveNot fixable by cost optimization -- would need a genuinely different entry/exit rule to find an edge on this market. atrstop_15 (pre-registered for A) was tested here (2026-08-26): -0.136 -> -0.109, still deeply negative, 2/3 folds. As a secondary (non-pre-registered) check, no_event_bias on A hit 3/3 folds (-0.136 -> -0.107) -- one of 3 chance-level hits out of a 30-test sweep (~3.75 expected by chance alone), not treated as validated -- both are still deeply cost-negative regardless. See All Analyses.
Full breakdown →
#26 Euro / US Dollar (EUR/USD) — Original — Strategy B Does not survive costs
857 trades 23.92% win rate net avg -0.192R/trade net total -164.2R MC probability of profit 0.1% MC risk of ruin 98.7% median drawdown 86.1%
WhyStructural failure.
WeaknessGross already negative (-0.127R/trade) and gets worse net (-0.192R) -- worst Monte Carlo probability of profit on the page alongside silver's original engine.
ImproveNo cost fix helps here. lookback_10 and no_event_bias (both pre-registered for B) were tested (2026-08-26): lookback_10 -0.192 -> -0.188, 2/3 folds; no_event_bias -0.192 -> -0.184, 1/3 folds -- both still deeply cost-negative. See All Analyses.
Full breakdown →
#27 S&P 500 (US 500) — Variant (LuxAlgo) — Strategy A Does not survive costs
92 trades 16.3% win rate net avg -0.248R/trade net total -22.8R MC probability of profit 10.9% MC risk of ruin 0.7% median drawdown 30.3%
WhyStructural failure.
WeaknessGross already negative (-0.205R/trade), lowest win rate (16.3%) among original-pair variant-engine combos.
ImproveNot cost-fixable.
Full breakdown →
#28 Silver (XAG/USD) — Original — Strategy B Does not survive costs
812 trades 22.41% win rate net avg -0.289R/trade net total -234.6R MC probability of profit 0.0% MC risk of ruin 100.0% median drawdown 93.7%
WhyCost-driven failure, severe.
WeaknessGross was still positive (+0.015R/trade, thin but real) but net collapses to -234.6R (-0.289R/trade) -- silver's low unit price combined with tight relative stops makes even Vantage's small spread-only cost disproportionately punishing (0.304R avg cost, among the highest on the page).
ImproveSame fix idea as before the extension: wider stops relative to silver's typical spread, or larger position size per lot to amortize costs, would likely help more than changing the entry rule. lookback_10 and no_event_bias (both pre-registered for B) were tested (2026-08-26): lookback_10 -0.289 -> -0.288 (essentially no effect), 2/3 folds; no_event_bias -0.289 -> -0.216 (still deeply negative), 0/3 folds. See All Analyses.
Full breakdown →
#29 Silver (XAG/USD) — Original — Strategy A Does not survive costs
1177 trades 28.46% win rate net avg -0.339R/trade net total -398.5R MC probability of profit 0.0% MC risk of ruin 100.0% median drawdown 98.5%
WhyNow the single worst combo on the entire page.
WeaknessGross was mildly negative (-0.051R/trade) even before costs, and net collapses to -398.5R -- the largest total loss and, with 1,177 trades, not a small-sample artifact. Cost-in-R (0.287R) is nearly as high as Strategy B's on the same market.
ImproveThis is the strongest case on the page for simply not trading silver's original engine, Strategy A, at all rather than searching for a filter to rescue it -- the combination of a weak-to-negative gross edge and severe cost sensitivity leaves very little to work with. atrstop_15 (pre-registered for A) was tested here too (2026-08-26): -0.339 -> -0.424, WORSE not better, 0/3 folds -- this market's stop is already too loose relative to its near-zero edge, widening it further just bleeds more on losers. See All Analyses.
Full breakdown →