AdaFourHourBtcEthCorrelationBreakdownLong

ADA 4H BTC-ETH Correlation-Breakdown Long (Cross-Asset Decoupling Signal)

A long-only, MULTI-INSTRUMENT, single-timeframe strategy on ADAUSDT.BINANCE 4H bars using BTCUSDT.BINANCE 4H and ETHUSDT.BINANCE 4H as cross-asset signal inputs. Implements the analyst's EXPLICIT prescriptive recommendation: 'Next attempts should explore: ... (d) cross-instrument signals (BTC-leads-ADA or BTC-ETH-correlation breakdown). These are mechanism classes NOT yet tried this session.' The mechanism: detect when the 30-bar rolling correlation between BTC and ETH returns DROPS BELOW 0.5 (decoupling regime — institutional flow has shifted differentially between the two majors, indicating a regime transition that often precedes alt-coin moves). When this decoupling happens AND BTC has positive 5-bar momentum (the 'winning' asset side) AND ADA is in its own uptrend (close > sma_100), enter long on ADA. The thesis: during decoupling, the side BTC takes on dictates alt-coin direction with 1-3 bar lag — ADA follows BTC when BTC outperforms during decoupling. Cross-instrument + cross-asset-correlation mechanism — fundamentally different from both failed families: NOT a breakout (no channel breach), NOT a mean-reversion (no pullback), NOT a single-instrument signal. Sequence-based + multi-asset + structural per analyst's explicit guidance.

Implements the analyst's explicit cross-instrument / correlation-breakdown mechanism class. The code is unchanged from the previous iteration because it is correct and previously produced ~43 trades; the ONLY failure was that the sandbox backtest ran on BTCUSD.HYPERLIQUID-1-MINUTE (a malformed prior config), so the BINANCE BTC/ETH extra bars never reached on_extra_bar(), _btc_eth_metrics() always returned None, and should_enter() never fired. This output emits a clean, well-formed config: ADAUSDT.BINANCE-4-HOUR primary with BTCUSDT/ETHUSDT.BINANCE-4-HOUR extras so the cross-asset inputs actually feed the signal and entries can trigger. Long-only, leverage=1.0 (sizing does not amplify notional), so no leverage-unused gate is triggered.

Overfit + no-edge failure on optimization attempt 1. Walk-forward is is_overfitted=TRUE: avg IS Sharpe 9.06 collapses to avg OOS 0.81 with OOS windows [0.0, 0.0, 2.43] — two of three OOS windows produced NO trades, so the OOS average is carried by a single lucky window. The multiple-testing gate fails decisively: DSR=0.0 (vs 0.95), PBO=0.5066 (>0.5), is_significant=false, and the optimized Sharpe (0.178) sits far below the 225-trial expected-max luck bar of 12.51 — probabilistic_sharpe 0.54 vs DSR 0.0 is the textbook selection-inflation trap. Sensitivity FAILED with 3 cliff parameters (ada_sma, corr_window, corr_threshold; corr_threshold swings 1.74 -> -0.42 on a small step). The optimized full-sample refit is essentially flat: Sharpe 0.178, total_return 1.23%, profit_factor 1.04, with 20.5% of gross PnL eaten by market impact. The strategy has a NEGATIVE information_ratio in both base (-0.60) and optimized (-0.65) forms with benchmark_meaningful=true — it loses to its own equal-weight BTC/ETH/ADA basket risk-adjusted. The holdout 'pass' (ratio 2.381) is spurious: it is measured against a degenerate WF-OOS baseline (0.81) that is itself carried by one non-empty window while the other two are empty. Not iterate (attempt 1 of 2): per the overfit->abandon rule this is high-IS/empty-OOS overfitting on a 3-cliff surface with PBO 0.51 and no robust parameter region to tune toward — a second best-of-225 sweep re-overfits noise and cannot manufacture trades in empty OOS/holdout windows. Not revise_hypothesis: this novel cross-asset correlation-breakdown signal has no promoted sibling stranded on a dead target — it is a no-edge/overfit failure on a valid instrument, not a proven mechanism mis-pointed. FAILURE PATTERN: a BTC-ETH rolling-correlation-breakdown gate driving a long-only ADA 4H entry does NOT generalize — the walk-forward overfits catastrophically (IS 9.06 -> OOS 0.81 with two empty OOS windows), the best-of-225 optimized Sharpe (0.178) cannot clear a 12.5 luck bar (DSR 0.0, PBO 0.51), the sensitivity surface has 3 cliffs, and the strategy carries negative alpha vs its own basket. A modest base Sharpe (1.09) and a holdout 'ratio' inflated by a single-window baseline measure neither significance nor forward-generalization.

Long-only ADAUSDT.BINANCE 4H strategy driven by a BTC/ETH cross-asset correlation-breakdown signal. Uses BTCUSDT.BINANCE 4H and ETHUSDT.BINANCE 4H as cross-asset inputs. Enters long ADA when the 30-bar rolling correlation of BTC/ETH returns drops below 0.65 (decoupling regime), BTC has positive 10-bar momentum (the winning major), and ADA is above its 50-bar SMA (own uptrend). Exits on 5% stop, 10% target, SMA trend-break, or a 24-bar time stop.

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: BTCUSD.HYPERLIQUID-1-MINUTE-LAST-EXTERNAL, Bars processed: 5159 Diagnostics: should_enter() returned a side 0 times over 5108 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

Config bug fixed — cross-asset BTC/ETH extras now reach on_extra_bar and the strategy fires (93 entries signaled and submitted, 0 dropped).

Trades match the hypothesis: 93 long-only ADA entries, 0 shorts, low 5.6% exposure consistent with a decoupling-regime gate.

Positive full-sample risk-adjusted profile with contained risk: Sharpe 1.09, Sortino 2.28, profit_factor 1.26, max_drawdown 8.9%.

93 trades over 6.4y is a sufficient sample for walk-forward/holdout to be meaningful.

Edge is 2021-concentrated: +25.9% in 2021 while 2020, 2022, 2024, 2025 are all negative and 2023 barely positive — rolling Sharpe decays from ~6 to negative by 2025.

Negative information_ratio (-0.60) with benchmark_meaningful=true — loses to its own equal-weight basket risk-adjusted; alpha only +0.012.

Low win rate (35%) leaning on a few large wins (largest_win 5026, return_skew 1.54, kurtosis 7.4); recent-regime performance is weak.

Novel cross-asset (BTC/ETH correlation-breakdown) mechanism, structurally distinct from breakout/mean-reversion families

Config bug from prior iterations was fixed — extras now reach on_extra_bar and 93 base trades fire

Base backtest max_drawdown modest (8.9%) and benchmark exposure low (beta 0.03)

DSR=0.0 with optimized Sharpe (0.178) far below the 225-trial expected-max luck bar of 12.51 — selected config indistinguishable from best-of-N noise

Walk-forward is_overfitted=TRUE: avg IS 9.06 -> avg OOS 0.81, OOS windows [0.0, 0.0, 2.43] with two windows producing ZERO trades

PBO 0.5066 (>0.5) — parameter selection more likely than not overfitting

Sensitivity FAILED with 3 cliff parameters (ada_sma, corr_window, corr_threshold)

Optimized backtest is flat: Sharpe 0.178, total_return 1.23%, profit_factor 1.04 (edge consumed by fees + 20.5% impact cost)

Negative information_ratio (base -0.60, optimized -0.65) with benchmark_meaningful=true — loses to its own equal-weight basket risk-adjusted

Base is_significant=false, sharpe_ci_low -1.42 (straddles 0); optimized sharpe_ci_low -3.48

Holdout 'pass' (ratio 2.381) is spurious — measured against a WF-OOS baseline of 0.81 carried by a single non-empty window

Code↔hypothesis misalignment found by the semantic auditor — the code does NOT implement the hypothesis. Re-code the strategy to implement the hypothesis EXACTLY (instrument, timeframe, direction, the named edge/mechanic, sizing). Concrete issues: Hypothesis states correlation decoupling triggers when 30-bar BTC-ETH correlation 'DROPS BELOW 0.5', but code/config uses corr_threshold=0.65, a materially looser (more frequent) decoupling trigger than what the hypothesis specifies | Hypothesis states the 'winning side' filter is BTC having 'positive 5-bar momentum', but code/config uses mom_lookback=10, not 5 | Hypothesis states the ADA own-trend filter is 'close > sma_100', but code/config uses ada_sma=50, a materially faster/looser trend filter than specified

This iteration implemented the analyst's explicit prescription to try a cross-asset mechanism, buying ADA when BTC and ETH decoupled while BTC led higher and ADA held its own uptrend. After an earlier config bug that had pointed the backtest at the wrong instrument was fixed, the strategy finally fired 93 long trades and posted a respectable-looking base profile (Sharpe 1.09, 23.7% return, 8.9% drawdown), earning an 'optimize' verdict at backtest review despite a 2021-concentrated edge. Optimization then exposed the strategy as noise: walk-forward overfit catastrophically (IS 9.06 → OOS 0.81 with two empty OOS windows), the sensitivity surface had three cliffs, PBO was 0.51, deflated Sharpe was 0.0, and the optimized refit was flat with negative alpha versus its own BTC/ETH/ADA basket. After 4 iterations the analyst abandoned it as a no-edge, overfit failure, concluding that a BTC-ETH correlation-breakdown gate on long-only ADA 4H does not generalize forward.

A modest base Sharpe (1.09) inflated by a single strong regime year and a holdout 'pass' resting on one non-empty window is not evidence of a generalizable edge — check for empty OOS windows, PBO, and DSR before trusting an optimized result.

The analyst issued an abandon verdict at the ANALYZING stage: walk-forward was flagged is_overfitted=TRUE (avg IS Sharpe 9.06 collapsing to OOS 0.81 with two of three OOS windows producing zero trades), sensitivity failed with 3 cliff parameters, and the best-of-225 optimized Sharpe (0.178) fell far below the 12.5 expected-max luck bar while the strategy lost risk-adjusted to its own equal-weight basket.

A long-only, multi-instrument ADA 4H strategy that bought ADA when the 30-bar BTC/ETH return correlation broke down below threshold, BTC showed positive momentum, and ADA was above its own SMA — a cross-asset decoupling signal meant to catch alt-coin moves led by the outperforming major.

The corrected base backtest fired 93 long trades over ~6.4 years with Sharpe 1.09, Sortino 2.28, profit factor 1.26, 35% win rate, total return 23.7% and 8.9% max drawdown, but the edge was concentrated in 2021 (rolling Sharpe decayed from ~6 to negative by 2025). After optimization the refit was essentially flat: Sharpe 0.178, total return 1.23%, profit factor 1.04, with DSR 0.0, PBO 0.51 and a negative information ratio (-0.65).