SolEfficiencyRatioGatedTrendBreakout

SOL Trend Capture Gated by a Trend-EFFICIENCY Regime Filter — Long-Short: Take a Directional Breakout ONLY When the Kaufman Efficiency Ratio Confirms the Market Is in an EFFICIENT (Trending) Regime, Stand Flat in Choppy/Inefficient Regimes, ATR-Trailing Exit (SOLUSDT.BINANCE USD-M, 4H, 3-Parameter)

A LONG-SHORT, single-instrument directional strategy whose LOAD-BEARING signal is a REGIME classifier — the Kaufman Efficiency Ratio (ER = |net price change over M bars| / sum of absolute bar-to-bar moves over the same M bars) — not a moving-average crossover. The premise directly attacks the #1 reason directional strategies die in this factory: pure momentum/breakout bleeds because it fires in CHOPPY, mean-reverting regimes where whipsaw + fees dominate. ER measures exactly that — it is ~1.0 when price travels in a straight (efficient, trending) line and ~0 when it oscillates (inefficient, choppy). We take a directional position in the direction of a channel breakout ONLY when ER ≥ threshold (the market is demonstrably trending), and we STAND FLAT whenever ER is low — precisely the periods where the factory's dead momentum/breakout strategies lost money. This is deliberately DISTINCT from: the promoted BTC dual-timeframe MA momentum-confluence (different confirmation mechanism — a regime scalar, not a second timeframe of the SAME momentum), the dead cross-sectional/relative-strength MOMENTUM baskets (single-instrument directional, not a beta-neutral rank), the dead OI-trend (no supplementary data), and the dead pure Donchian breakout (which had NO regime gate and fired in chop). SOL is chosen as a liquid, high-trend-persistence major with multi-year 4H history (a NEW mechanism on a fresh symbol, not a symbol-swap of a dead mechanism per L62). 3 parameters: ER lookback M, ER threshold, ATR-trailing multiple.

NOTE ON THE ITERATION: the `previous_code` supplied in this brief was BtcEthBetaHedgedSpreadReversionPair (an ETH/BTC 1H beta-hedged stat-arb) -- a pipeline mislinkage, exactly what QA flagged as a whole-strategy swap. There was no prior code for THIS hypothesis to edit, so this submission implements the hypothesis as specified from the correct starting point: SOLUSDT.BINANCE single instrument, 4-HOUR bars, long-short, Kaufman Efficiency Ratio regime gate on a channel breakout, ATR-trailing exit, exactly 3 parameters (ER lookback M, ER threshold, ATR multiple). Every mismatch QA listed (instrument set, timeframe, core edge, parameter set) is resolved. IMPLEMENTATION CHOICES: (1) the ER lookback doubles as the Donchian channel length, so the breakout and the regime measurement observe the same horizon and the parameter count stays at 3 as the hypothesis requires; (2) the returned signal is the continuous normalised channel position (varies every bar, crosses +-1 exactly at a breakout), with ER applied as a separate gate in should_enter rather than multiplied into the signal -- a strong regime must not be able to substitute for an absent breakout; (3) the channel and ATR are read from the PRIOR bars before this bar is pushed, so the breakout test can never compare the close to a channel that already contains it (no look-ahead). VALIDATION on the real 12,917-bar SOLUSDT 4H catalog series (2020-09..2026-08), engine-free, charging the full 0.10% Binance USD-M taker round trip and filling at the next bar's open: the shipped config (er_period 30, er_threshold 0.20, atr_mult 4.0) gives 173 trades, +4.04% net per trade (vs the 0.15% futures floor), t = 2.39, win rate 46.8%, ~29 trades/year, average hold ~44 bars (~7 days), and is POSITIVE IN ALL 7 CALENDAR YEARS including the trailing 12 months (32 trades, +0.44%/trade) -- so the Layer-3 sandbox window should show trades and a positive per-trade figure, not the negative one that killed the previous submissions. THE ER GATE IS LOAD-BEARING BUT MODEST, and I would rather state that than oversell it: the same breakout with the gate removed (er_threshold = 0) gives 213 trades at +3.45%/trade, t = 2.24, 6/7 years positive. The gate therefore removes ~19% of trades, lifts per-trade net by ~17%, lifts the weak second half from +1.45% to +2.18%/trade, and turns the one negative year positive -- real, directionally consistent with the hypothesis, but the breakout itself carries most of the edge. ROBUSTNESS: a 75-cell scan of the declared bounds box (er_period 20-45 x er_threshold 0.15-0.40 x atr_mult 3-5) is positive in 74 of 75 cells (worst -0.12%/trade), so _param_bounds sits on a broad plateau rather than a spike; the chosen cell is the highest-t-stat cell of the full sample, not the best trailing-year cell. CAVEATS for the analyst: (a) this is trend-following, so win rate is <50% and the equity path is right-tailed -- my crude per-trade path shows ~38% max drawdown at half-notional sizing, which is why sizing is ATR-risk based (3% of equity at the initial trail distance) and gross is capped at 0.5x equity with leverage 1.0; (b) the second half of the sample (+2.18%/trade) is materially weaker than the first (+5.92%), the usual trend-following decay, though it stays clearly above costs; (c) SOL 4H ATR is large, so a 4x ATR trail is a ~10% initial stop -- deliberate, since a tighter trail (atr_mult 3) is the weakest corner of the bounds box.

Single-name OHLCV trend/breakout with a regime-confluence gate (ER) — the trend-confluence-clone class with 0/213 OHLCV and 0/97 Hyperliquid survival (L41/L9/L62). Textbook mid-trade-count overfit signature (172 trades, single symbol, breakout+confirmation), and L41 shows this construction dies in the 3-phase run even at higher Sharpe than this 0.95. The 394% headline is concentrated in the 2021 (+63%) and 2023 (+98%) explosive SOL trends and a handful of outlier days (kurtosis 9.0), with other years modest and information_ratio -0.64 (worse than holding SOL). The moderate base Sharpe is the ceiling; ~225-trial deflation plus a 15-day holdout will lower it below the OOS floor. The attractive metrics (Sharpe CI clears zero, PSR 0.998, DD 20.8%) are exactly the flattering-first-backtest trap L41 warns to override with the class prior rather than spend 2 hours optimizing. Failure pattern: single_name_trend_confluence_clone_bull_concentration.

Single-instrument, long-short directional trend strategy on SOLUSDT.BINANCE USD-M perps, 4-HOUR bars. Every bar it computes Kaufman's Efficiency Ratio over er_period bars (ER = |net close change| / sum of |bar-to-bar close changes|, ~1 in a straight trend, ~0 in chop) and the close's position inside the prior er_period-bar Donchian channel normalised by the channel half-width (chpos; |chpos| >= 1 means a new M-bar extreme, i.e. a breakout). calculate_signal returns chpos as a continuous decision variable. Entry is taken ONLY when the regime is efficient: flat + ER >= er_threshold + chpos >= +1 goes LONG, chpos <= -1 goes SHORT; in inefficient/choppy regimes (ER below threshold) it stands flat and never enters. The only exit is an ATR-trailing stop at atr_mult x ATR(14) from the highest high (long) / lowest low (short) since entry, so winners run while the trend stays efficient. Sizing is ATR-risk based and capital-relative: qty = equity * risk_frac / (atr_mult * ATR), capped at 50% of equity of notional. Three tunables: er_period (M), er_threshold, atr_mult.

Verification failed (Layer 4 — QA review): - WHOLE-STRATEGY SWAP / MISLINKAGE. The hypothesis specifies a single-instrument, long-short SOLUSDT.BINANCE 4-HOUR directional breakout gated by the Kaufman Efficiency Ratio (trade a channel breakout only when ER >= threshold, stand flat in chop, ATR-trailing exit; 3 params = ER lookback M, ER threshold, ATR mult). The submitted code is BtcEthBetaHedgedSpreadReversionPair — a two-instrument, market-neutral BTC/ETH beta-hedged spread MEAN-REVERSION stat-arb on ETHUSDT+BTCUSDT 1-HOUR bars (3 params = window, entry_z, exit_z). Instrument set, timeframe, and edge mechanism all differ. Critical fail. - INSTRUMENT MISMATCH (hypothesis SOLUSDT single-instrument; config ETHUSDT primary + BTCUSDT extra), TIMEFRAME MISMATCH (hypothesis 4-HOUR; config 1-HOUR), and CORE EDGE ABSENT (no Efficiency Ratio, no breakout channel, no ATR trail anywhere — the code does rolling-regression beta + residual z-score mean-reversion). - Moot given the swap, but the swapped-in pair is a self-declared failure: its own rationale recommends ABANDON (2/72 configs net-positive OOS, zero clear the 0.15% floor at t>2), and its sandbox is negative (total_return -2.80%, Sharpe -0.64, PF 0.92, avg_trade_return_pct +0.091% < 0.15% floor).

Analyst should confirm the ER gate's contribution survives walk-forward OOS; if it adds nothing OOS, this reduces to a plain Donchian breakout (a repeatedly-abandoned OHLCV class).

The hypothesis frames the Kaufman Efficiency Ratio gate as the LOAD-BEARING signal, but the developer's own validation shows it is real-but-modest: gateless (er_threshold=0) gives 213 trades @ +3.45%/trade (t=2.24) vs 173 @ +4.04% (t=2.39) with the gate. The breakout carries most of the edge; the ER gate removes ~19% of trades and lifts per-trade net ~17%. The gate IS implemented and beneficial (not a core-edge-absent critical), but its contribution is smaller than the narrative implies.

Analyst/Risk should weight recent-period OOS and holdout heavily and confirm the ATR-risk sizing keeps drawdown within limits.

Trend-following decay disclosed by developer: second-half +2.18%/trade vs first-half +5.92%; sandbox Sharpe only 0.32 with 16.9% max DD (CI to 32.8%). Right-tailed sub-50% win rate is expected for the family, but the decaying edge and deep drawdown are the binding risks.

Healthiest metrics of the class: Sharpe 0.95 with sharpe_ci_low +0.227 (CI clears zero), PF 1.65, PSR 0.998, max DD 20.8%

Positive trend-following payoff shape: 46.5% win rate but avg_win ~1.9x avg_loss, positive skew 0.81, tail_ratio 1.18

Positive every full year including the 2022 bear (+0.7%) — regime gate did avoid catastrophic trend-fighting; avg_trade 4.14% clears fees easily

Single-name OHLCV trend/breakout + regime-confluence gate — class with 0/213 OHLCV and 0/97 Hyperliquid survival (L41/L9/L62)

Mid-trade-count (172) overfit signature (L41): flattering first backtests on this exact construction die in the 3-phase run every time, even at higher Sharpe than this

Return concentrated in 2021 (+63%) and 2023 (+98%) explosive trends and a handful of outlier days (kurtosis 9.0); other years modest — walk-forward OOS windows will be marginal

information_ratio -0.64: worse risk-adjusted than simply holding SOL

0.95 base Sharpe will be lowered by ~225-trial deflation + 15-day holdout, likely below the 0.5 OOS floor

SolEfficiencyRatioGatedTrendBreakout tried to fix the #1 reason directional strategies die in the factory — firing in choppy regimes — by gating a SOL 4H Donchian breakout on Kaufman's Efficiency Ratio so it only trades when the market is demonstrably trending, holding the trend with an ATR trail. It produced the healthiest backtest of its class: a 394% headline return, Sharpe 0.95 with a CI clearing zero, PF 1.65, PSR 0.998, and a 20.8% max drawdown, positive in every full year. But the analyst abandoned it on the second iteration at the backtest_review gate, reading the metrics as the flattering-first-backtest trap: the return was concentrated in the 2021 and 2023 explosive trends with a negative information ratio (-0.64, worse than holding SOL), and the single-name trend-confluence-clone class has 0/213 OHLCV survival and reliably dies under 3-phase optimization and holdout deflation. It never advanced to optimization, analysis, or risk review.

Flattering first-backtest metrics (Sharpe CI clearing zero, PSR 0.998, low drawdown) are not enough to override a strong class prior — a single-name trend/breakout-plus-confirmation with a bull-concentrated return and negative information ratio matches a 0/300+ survival pattern that consistently dies at holdout, so the class prior should override spending hours optimizing it.

It was abandoned at the pre-optimization backtest_review gate on iteration 2: despite passing the Sharpe, drawdown, and profit-factor thresholds, the analyst applied the class prior — single-name OHLCV trend/breakout with a regime-confluence gate has 0/213 OHLCV and 0/97 Hyperliquid survival, a mid-trade-count overfit signature that L41 shows dies in the 3-phase run even at higher Sharpe, so ~225-trial deflation plus the 15-day holdout would likely push the 0.95 base Sharpe below the 0.5 OOS floor.

A long-short, single-instrument directional strategy on SOLUSDT.BINANCE USD-M 4H bars that takes a Donchian channel breakout only when Kaufman's Efficiency Ratio confirms an efficient (trending) regime, stands flat in choppy/inefficient regimes, and exits via an ATR-trailing stop, with three parameters (ER lookback, ER threshold, ATR-trailing multiple).

This was the healthiest backtest of its class: total_return 393.6%, Sharpe 0.95 (CI low +0.227, clears zero), profit_factor 1.65, PSR 0.998, max_drawdown 20.8%, win_rate 0.47 with avg_win ~1.9x avg_loss and positive skew 0.81 over 172 trades, and it was positive every full year including the 2022 bear (+0.7%). But the return was concentrated in the 2021 (+63%) and 2023 (+98%) explosive SOL trends and a handful of outlier days (kurtosis 9.0), and information_ratio was -0.64 — worse risk-adjusted than simply holding SOL.

BtcEthBetaHedgedSpreadReversionPair