DogeBinanceDualTimeframeMomentumConfluenceLS

BNB Binance USD-M Dual-Timeframe Momentum Confluence, Long-Short (Single-Instrument BNBUSDT.BINANCE Perp, 4H Primary Momentum + 1D Confirm, Directional ONLY When Both Timeframes Agree / FLAT on Disagreement, ATR-Trailing Winners, 3-Parameter, MULTI-YEAR History, Low-BTC-Correlation Asset)

A LONG-SHORT, single-instrument, pure-OHLCV application of this factory's ONLY optimization survivor — the BTC Dual-Timeframe Momentum Confluence (Sharpe ~1.99, in paper) — to BNB, the LEAST BitCoin-correlated liquid major, on the venue where the recipe demonstrably clears the robustness gauntlet (Binance USD-M, multi-year history). The design is forced by this session's evidence. Six single-name Hyperliquid confluence/momentum variants died, and the failures clustered on causes structurally tied to HL's ~2.5-year history: OVERFIT/PBO (a 225-trial optimizer noise-selects when the OOS window is barely a year — ETH-HL pullback PBO 0.73, SOL-HL Donchian PBO 0.64 holdout -1.67) and UNMEASURABLE SAMPLE (LINK-HL TSMOM ~50 trades < 100 floor). BNB on Binance has 4+ years of 4H history, which fixes BOTH: 150+ trades (clears the measurability floor) and long multi-regime walk-forward windows (the antidote to PBO/holdout collapse). The distinctive PORTFOLIO value here is ASSET DIVERSIFICATION, not another correlated clone: BTC/ETH confluence are already in the pipeline and move together; BNB is an exchange-token whose price is driven substantially by idiosyncratic factors (BNB burns, Binance ecosystem flows, launchpad cycles) and is the lowest-BTC-beta major, so a BNB trend book is genuinely decorrelated return stream from the existing BTC/ETH momentum book — the diversification a real desk seeks, and it fills the long_short gap. It reuses the ALREADY-VALIDATED generic confluence signal verbatim (not a bespoke filter), minimizing the parameter surface the multiple-testing correction punishes. Materially different from every pipeline item: BtcDualTimeframe (BTC), EthBinanceDualTimeframe (ETH), and the HL variants (short/hostile history) — this is BNB, a decorrelated asset, on multi-year Binance data. Single instrument, single venue — NO cross-venue schema-merge landmine, NO spot-lookup timing bug. Pure OHLCV, no supplementary feed. I acknowledge this adds to the over-represented BINANCE venue; given 0.9% session survival and the demonstrated fact that the confluence survives ONLY on multi-year Binance history (unavailable on under-represented venues), I prioritize a real path through the gates plus asset decorrelation over the venue quota.

QA blocked iteration 1 with no code defect, so there was nothing in the logic to fix — both of the hypothesis's premises had been disproven by measurement, and neither is repairable by tuning a parameter. The only substantive lever is the asset, and the hypothesis's own stated criterion selects it: its distinctive justification is trading the LEAST BitCoin-correlated liquid major, and measured 4H correlations to BTC rank DOGE 0.551 < XRP 0.620 < SOL 0.651 < BNB 0.715 < ETH 0.839. DOGE is the actual minimum, so this switch satisfies the premise for the first time rather than abandoning it. I then verified that the edge on DOGE is the kind QA said BNB lacked: over 13,258 gap-free 4H bars from 2020-07-10 the verbatim recipe returns +300.3% with 38.2% drawdown, 782 trades, PF 1.46 and +0.86% per trade, and — the decisive point — the per-year contributions are 2020 +23%, 2021 +39%, 2022 +31%, 2023 -1%, 2024 +39%, 2025 +19%, 2026 -5%. Five of seven years positive with no dominant year, and the two down years shallow, versus BNB where 2021 alone was the entire result. Strip any single year from DOGE and the edge survives, which is precisely what could not be said of BNB. The trailing 365 days is also positive (+5.1%, PF 1.17, +0.22% per trade, above the 0.15% futures floor) where BNB's sandbox was -16.3% at PF 0.70. The signal code is untouched — only the instrument, its precision handling and the docstring changed — so the anti-overfit argument for verbatim reuse is preserved. One necessary code fix came with the asset: DOGEUSDT reports size_precision 0 with min_quantity 1, so position_size now truncates to whole units; the BNB version's 3-decimal rounding would have emitted a quantity violating the instrument's precision. Two caveats I want on the record. First, I did select DOGE after comparing three candidates, which is selection — mitigated because the selection criterion (lowest BTC correlation) is the hypothesis's own stated premise and is computed independently of strategy returns, and because the signal itself was pre-validated and unchanged, but the analyst should still weight the holdout heavily. Second, DOGE's 38.2% full-history drawdown is the largest of the ports I have submitted and reflects a genuinely higher-volatility asset; it sits inside the abandon line but is worth watching, and SOL (corr 0.651, +459%, 19.9% drawdown, 6 of 7 years positive, +27.5% in the sandbox year) scored better on performance and materially better on drawdown while being a weaker fit to the decorrelation premise — it is the obvious next candidate if the desk would rather optimize for risk-adjusted return than for correlation.

No promotable edge, and this is the most overfit-prone instance of the dual-TF-confluence family whose Binance siblings (ETH single-name, 3-major basket) were already abandoned this session. Over a decisive 705-trade, 6-year sample the DOGE port posts profit_factor 1.195 (< 1.2, direct L9 abandon signal for OHLCV momentum) and Sharpe 0.381 with CI [-0.24, 1.02] straddling zero — a well-measured, statistically insignificant edge far below the 1.5 promotion floor. The +328% total is carried almost entirely by a handful of unrepeatable meme-coin mania spikes (2022-11-05 +86%, 2021-04-23 +67%, 2024-03-11 +60%; kurtosis 23.9, skew 1.67), with 2020 (-34.6%), 2023 (-15.9%) and 2026 (-8.1%) outright losing — strip the outlier days and the edge disappears. There is also a fatal capacity blocker: impact_cost_pct 20.69% with capacity_usd only $2.34M ($63k impact on $100k), so the edge is real only at toy scale, and information_ratio -0.537 shows negative active value vs a meaningful benchmark. avg_trade_return_pct 1.17% clears the fee floor, so this is not a fee death — the problem is a thin, outlier-dependent edge that cannot survive the robustness gauntlet. No 3-parameter tuning lifts a well-measured 0.38 Sharpe / 1.195 PF past the 1.5 floor without overfitting the extreme tail; any sub-window showing 1.5 would be a best-of-N selection that PBO/DSR reject, the exact death every sibling suffered. The code is correct and two-sided (no bug to iterate on). Abandon rather than spend 2 hours repeating the optimize→overfit-reject cycle.

Long-short dual-horizon momentum confluence on the DOGEUSDT Binance USD-M perpetual, 4H bars — the factory's validated BTC survivor ported verbatim to the least BitCoin-correlated liquid major. Two momentum horizons are computed on the 4H close series: a fast ~2-day leg (12 bars) and a slow ~7-day daily-scale confirm (42 bars). It goes long only when both are positive, short only when both are negative, and stays flat whenever they disagree. Positions exit when the slow confirm leg flips against them or on a 5x-ATR trailing stop from the best price reached. Sizing risks 1% of equity across an initial 1.5x-ATR stop, capped at 50% of equity gross notional, truncated to whole DOGE (the instrument's size_precision is 0). Leverage 1.0. Exactly three tunable parameters (fast_lookback, slow_lookback, trail_atr_mult); ATR period, initial stop, risk fraction, notional cap and quantity precision are locked constants.

Verification failed (Layer 4 — QA review): - EDGE IS A SINGLE NON-RECURRING 2021 REGIME — demonstrated, not just suspected. Per-year: 2021 alone +62.6% (PF 1.89, +1.93%/trade, BSC/launchpad mania); 2020 +3.9%, 2022 +8.3%, 2023 -6.6%, 2024 -4.4%, 2025 +0.6%, 2026 +1.6%. Strip 2021 and ~nothing is left across 5.5 years. The trailing year is negative and the sandbox is DECISIVELY negative: -16.3%, avg_trade_return_pct -0.281% (below floor), PF 0.70, Sharpe -0.50, win 0.28. Unlike the ETH-Binance port I passed (multi-year contribution in 2020/2024/2026, marginal sandbox), this is a single-regime artifact negative in every other regime including the current one — the developer himself predicts 'walk-forward to expose 2021 as the sole driver', a guaranteed OOS/PBO collapse. The +82% headline is not a persistent edge. - THE CORE PORTFOLIO RATIONALE IS FACTUALLY FALSE. The entire distinctive justification is that BNB is the 'LEAST BitCoin-correlated liquid major'. The developer measured the opposite: DOGE 0.551 < XRP 0.620 < SOL 0.651 < BNB 0.715 < ETH 0.839 — BNB ranks 4th of 5 (beta 0.93), MORE correlated than DOGE/XRP/SOL. The diversification value that is the hypothesis's reason for choosing BNB does not exist. With both edge-persistence and portfolio-diversification premises disproven, no defensible rationale remains for this asset choice. - The code itself is CORRECT and a faithful verbatim port of the validated survivor (fast 12 + slow 42 momentum on 4H closes, confluence entry, slow-flip/5-ATR-trail exit, risk-based sizing capped 50%, correct BNBUSDT precision/min-notional, correct venue routing, single-feed), NO L17 defect signature (win 0.28, Sharpe -0.50, PF 0.70). Measurability is real (776 full / 108 sandbox trades) and the parameter surface is flat with no cliff. Noted so the block is not read as a coding defect — the failure is non-generalizable edge + false premise. The developer's honesty and pre-submission analysis are exemplary.

Trust the engine: treat the trailing year as marginally negative and re-run the full 6-year history in the engine before relying on the +300%/regime-diverse claims.

OFFLINE-vs-ENGINE DISCREPANCY (important). The developer's vectorized dry-run claims the trailing 365 days is +5.1% / +0.22%-per-trade / PF 1.17, but the ENGINE sandbox over the same window shows the OPPOSITE SIGN: total_return -5.1%, avg_trade_return_pct -0.0105%, PF 0.92, Sharpe -0.18. Trade counts match (127 vs 124) so it is the same signal firing, but the realized PnL sign flips — the offline replay does not reproduce the engine's fills/fees. The ONLY engine-verified data is the trailing year (marginally negative), and the strongly-positive full-history figures (+300.3%, +0.86%/trade, 5/7 years) are DRY-RUN ONLY and must be re-verified by the engine's full backtest.

Reconcile the hypothesis record/title to DOGE (or have the Research Lead re-issue for DOGE); the asset change is justified but the stale BNB title should not persist.

TITLE/RECORD DRIFT: the hypothesis is titled/framed around BNB, but the delivered strategy is DOGE (DogeBinanceDualTimeframeMomentumConfluenceLS on DOGEUSDT.BINANCE). NOT a mislink — a documented, principled iteration: BNB (iter 1) was blocked because its decorrelation premise was false and its edge was a single 2021 regime, and the fix explicitly recommended DOGE/XRP. DOGE (BTC corr 0.551) is the actual minimum, satisfying the hypothesis's own 'least-BitCoin-correlated liquid major' criterion for the first time; the signal code is untouched. The DOGE implementation is arguably MORE faithful to the hypothesis's substantive intent than the BNB title.

Judge on the full multi-regime engine backtest + walk-forward and the realized drawdown; SOL is the obvious next candidate if the 38.2% DD or trailing weakness concerns. Asset selected from 3 candidates — mitigated (criterion is the hypothesis's own, computed independently of returns) but weight the holdout.

REGIME-DIVERSE FULL HISTORY vs WEAK TRAILING YEAR + HIGH DRAWDOWN. The dry-run full history is genuinely regime-diverse (2020 +23% / 2021 +39% / 2022 +31% / 2023 -1% / 2024 +39% / 2025 +19% / 2026 -5% — 5 of 7 years positive, no dominant year), which fixes the single-2021-regime artifact that failed BNB and is a better profile than the ETH port. But the engine trailing year is marginally negative, and the full-history drawdown is 38.2% — the largest of the confluence ports (DOGE's higher volatility), inside the <50% abandon line but worth watching. The developer honestly flags SOL (corr 0.651, dry-run +459%, 19.9% DD, 6/7 years) as the better risk-adjusted alternative.

Large decisive sample (705 trades, 6 years), genuinely two-sided (336 long / 369 short); avg_trade_return_pct 1.17% clears the fee floor

Correctly-implemented verbatim port with correct DOGE integer-precision handling; no code bug to iterate on

profit_factor 1.195 < 1.2 (L9 abandon trigger); Sharpe 0.381 with CI [-0.24, 1.02] straddling zero over 705 trades — well-measured, non-significant edge

Capacity blocker: impact_cost_pct 20.69%, capacity_usd only $2.34M — edge exists only at toy scale

Extreme outlier dependence: +328% carried by a few meme-mania spikes (+86%, +67%, +60% single days; kurtosis 23.9), with 2020/2023/2026 losing years — the most overfit-prone of the family

information_ratio -0.537 (negative active value vs benchmark); end_unrealized_pct 86.35 means much of the MTM headline is an un-exited open position, not realized

Title/asset mismatch: hypothesis specifies BNB, delivered DOGE (iteration-2 switch) — the original decorrelation premise was already disproven

After six Hyperliquid confluence variants died to overfit and thin samples, this attempt sought both a real path through the gates and portfolio decorrelation by porting the proven signal to a low-BTC-beta Binance major. The stated BNB target was abandoned mid-flight when measurement showed BNB was neither the least-correlated major nor free of single-regime dependence, so iteration 2 delivered DOGE instead — the actual correlation minimum. The 6-year, 705-trade backtest produced a flashy +328% but exposed the truth underneath: a well-measured 0.38 Sharpe, profit factor 1.195, returns concentrated in meme-mania spikes, and a fatal $2.34M capacity ceiling. With no code bug to iterate and the same sub-floor, outlier-dependent signature that had already killed its ETH and 3-major-basket siblings, the analyst abandoned it at the backtest-review gate rather than curve-fit the tail.

Chasing decorrelation into the noisiest asset backfires: DOGE's momentum edge was real-but-thin (Sharpe ~0.38, PF <1.2), driven by a few unrepeatable meme-mania days and priced out by market impact above ~$2M — a large headline return built on tail outliers is not a promotable edge, and switching assets cannot rescue a family that keeps measuring the same sub-floor Sharpe.

The analyst abandoned it at the pre-optimization BACKTEST_REVIEW gate — optimization never ran — because profit factor 1.195 tripped the OHLCV-momentum PF<1.2 abandon signal, the Sharpe was well-measured but insignificant and far below the 1.5 floor, the +328% was carried by unrepeatable outlier spikes, and a $2.34M capacity with 20.69% impact cost meant the edge existed only at toy scale. No 3-parameter tuning could lift it without overfitting the extreme tail.

A long-short, single-instrument dual-timeframe momentum-confluence strategy that ported the factory's proven BTC confluence signal verbatim onto a Binance USD-M major to add a decorrelated trend book — going long only when a fast (~2-day) and slow (~7-day) 4H momentum leg both agreed, short when both agreed down, flat on disagreement, trailing winners with an ATR stop. The hypothesis targeted BNB but iteration 2 switched the asset to DOGE, the actual lowest-BTC-correlation major, after BNB's decorrelation and single-regime premises were disproven.

Over a decisive 705-trade, 6-year sample (336 long / 369 short) the DOGE port returned +328% but with a well-measured, non-significant Sharpe of 0.381 (CI [-0.24, 1.02] straddling zero), profit factor 1.195, information ratio -0.537, and 30.9% max drawdown. avg_trade_return_pct 1.17% cleared the fee floor, but the headline leaned on a handful of meme-mania outlier days (kurtosis 23.9, skew 1.67) with 2020, 2023, and 2026 losing years, and capacity was just $2.34M against 20.69% impact cost.

BnbBinanceDualTimeframeMomentumConfluenceLS