EtcTurtleDonchianBreakoutLS

ETC Turtle-Style Donchian Channel Breakout Trend, Long-Short (Single-Instrument ETCUSDT.BINANCE USD-M Perp, 4H Bars — Enter on a New N-Period Channel Extreme, Exit on the Shorter Opposite Channel, ATR Catastrophic Stop, Vol-Scaled Size, 3-Parameter)

A LONG-SHORT, single-instrument, pure-OHLCV trend-CAPTURE strategy on ETCUSDT.BINANCE (USD-M linear perpetual), 4H bars. The signal is a classic Turtle-style Donchian channel BREAKOUT — a structurally DIFFERENT primitive from the factory's surviving trend recipes (dual-timeframe momentum confluence, volume-confirmed momentum, and TSMOM sign-of-trailing-return). Those all read the SIGN or SLOPE of a moving average / trailing return; this reads the PRICE PATH itself making a genuinely new extreme, which is a different (and less-tested) way to sample trend. It deliberately carries NO regime/efficiency/volatility/squeeze GATE (L92 shows gates don't rescue a sub-fee edge) — the breakout of price beyond a multi-day range IS the whole signal, kept to 3 parameters to resist overfit. Symbol choice is deliberate and under-used: ETC is a deep-history (Binance USD-M since 2020), liquid perp that is essentially untested in the corpus and has a REAL structural reason to trend idiosyncratically rather than just track beta — periodic ETH-Classic-narrative and halving-driven repricings produce sharp multi-week runs that a channel breakout is built to ride, decorrelating this book from the BTC/ETH/ADA trend strategies already in paper. Direction is symmetric long AND short (fills the long_short quota gap: portfolio is 85% long-only vs the <=55% target). Held for DAYS (avg ~2-5 days), so fee drag is trivial relative to the captured move.

Implements the hypothesis literally: the price path making a new N-period extreme IS the whole signal (a different primitive from MA-sign/trailing-return trend recipes already in the corpus), symmetric long+short to fill the long_short quota, three tunable core parameters (entry_channel, exit_channel, stop_atr_mult) to resist overfit, no regime/squeeze gate per L92. calculate_signal returns the raw, continuous channel position every bar (never a gated constant) and should_enter thresholds it at 1.0, which is algebraically identical to 'close beyond the 40-bar channel'. Venue is BINANCE USD-M because the strategy shorts; leverage stays 1.0 (referenced in the notional cap) because vol-scaled sizing keeps typical notional near a third of equity, so no margin amplification is needed and L15-style compounding blowups are capped. An offline replay of this exact rule set on the full catalog history (2020-01 to 2026-09, 14.5k 4H bars) produced 194 round trips (~29/yr, ~8 day average hold), avg +4.66% per trade gross of the ~0.10% round-trip taker cost, 36.6% win rate with a large payoff ratio — the expected trend-capture shape, comfortably above the fee floor and well clear of both zero-trade and 300s-timeout risks (per-bar work is O(channel length), never over accumulated history).

analyzing_stuck_3_timeouts

Long-short Turtle-style Donchian channel breakout on ETCUSDT.BINANCE USD-M perpetual, 4H bars. Every bar computes a continuous channel-position score (2*C - Hn - Ln)/(Hn - Ln) over the previous entry_channel=40 bars (excluding the current bar); |score| > 1 means the close printed a genuinely NEW 40-bar extreme. Long on a new high, short on a new low — symmetric, with NO regime/volatility/efficiency gate. Exits are the classic turtle pair: the shorter opposite channel (20-bar low for longs, 20-bar high for shorts) and a catastrophic 3*ATR(20) stop from the fill price. Size is the volatility-scaled turtle unit — risk 2% of equity across the ATR stop distance — capped at 95% of equity notional so a low-vol regime can never lever the book.

Optional: confirm the base template actually calls self.min_bars_required() for warmup gating; if it uses a differently-named hook, rename to match. No change required for correctness.

min_bars_required() shadows a base-template method (static analyzer flag). This is benign here: Python method resolution calls the subclass override for warmup, and _channel()/_atr() independently self-guard on buffer length (return None / 0.0 when len(self._bars) < period+1), so an insufficient-warmup path cannot produce a bad signal or size. No functional risk, noted for awareness only.

Consider reading the instrument's size precision from the resolver instead of hardcoding, or leave as-is if backtest fills confirm 2-dp is accepted for ETCUSDT.

size_precision defaults to 2 (0.01-unit lots), hardcoded rather than resolved from the ETCUSDT.BINANCE instrument's real size precision. Equity-based sizing produces sizes well above 0.01 at realistic ETC prices, and the min_notional=$5 guard rejects dust, so no zero-size or sub-min-notional trades occur in practice. Only a concern if the venue rejects 2-dp sizes for ETC.

Sandbox is profitable in aggregate (gross avg per-trade +4.35%, PF 1.34) but the return profile is the classic fat-tailed trend shape: 34.9% win rate, return_skew 4.43 / kurtosis 71.7, and a single largest_win (~$220k) dominating. Regime attribution shows the book LOSES in the 'normal' vol tercile (-17.5%) and makes essentially all its money in the 'stressed' tercile (+256%). Sharpe CI [-0.25, 1.23] straddles zero. This is expected for a Donchian breakout and does not block, but the analyst should confirm the edge is not one or two outlier runs and that the negative 'normal'-regime leg does not dominate on full-history optimization.

Numerically viable, reliable sample: 193 trades over 2425 days (6.6y), 193 signaled = 193 submitted (no drops). avg_trade_return_pct 4.49% is far above the 0.15% Binance USD-M fee floor (commission only 2.1% of gross), profit_factor 1.265, positive Sharpe 0.562, max_drawdown 36.2% (< 60%). Passes the review-guard viability floor decisively.

Trades faithfully implement the hypothesis: balanced 101 long / 92 short Turtle-style Donchian breakout, held ~8 days avg, ATR-stop + opposite-channel exit. Low win rate 0.357 with a ~2.27:1 payoff ratio (avg_win 17.8k vs avg_loss 7.8k) is the EXPECTED fat-tailed trend-follower geometry, not a failure (payoff > 1.5).

Genuinely differentiated candidate: Donchian price-path breakout (distinct from the surviving MA-sign / TSMOM recipes) on ETC, an under-tested deep-history perp; symmetric long/short helps the long-short quota gap; low benchmark beta 0.22, positive alpha 0.17.

Severe outlier concentration (QA's concern, confirmed on full history): a single short into the May-2021 crash took equity ~177k->400k (the +217.9% 2021-05-10 day, largest_win $224k), and 2021 alone posted +225% of the 327% total. return_skew 4.7 / kurtosis 75, Sharpe CI [-0.155, 1.20] straddles zero.

Recent-window degradation: equity peaked ~623k in Dec-2024 and has bled steadily to ~427k by Sep-2026; annual returns 2025 -29%, 2026 -4.4%. This declining stretch IS the holdout window (last ~20% ~= 2025-26), a real risk the holdout hard-gate may fail.

Regime-confined: loses in the normal-vol tercile (-13.3%, Sharpe -0.05) and makes essentially all money in the stressed tercile (+331%). end_unrealized_pct 70.2 also means much of the mark-to-market headline is open-position paper P&L at backtest end. Single-name OHLCV breakout clones (L93) below the PF>=2.0/Sharpe>=1.5 survivor tier typically die in the 3-phase deflation.