DogeWeeklyGatedDonchianBreakoutLS

DOGE 4H Breakout Gated by the Weekly Trend — Long-Short, Single-Instrument Directional Continuation: Take a 4H Range Breakout ONLY in the Direction of the Prevailing WEEKLY Trend, Hard ATR Stop + Drawdown-Capped (BINANCE DOGEUSDT USD-M, 4H Primary + 1W Confirm, Pure OHLCV, 3-Parameter)

A LONG-SHORT, single-instrument directional-continuation strategy on DOGEUSDT.BINANCE USD-M perp. It keeps the ONE edge class that empirically survives in this factory — direction-aligned momentum on a liquid perp (the ADA/BTC 4H+1D confluence successes) — but is deliberately reconstructed to sit far from every abandoned variant in mechanism, confirm-timeframe, instrument, venue and direction, and to fix the specific death causes. WHAT IS DIFFERENT FROM THE DEAD/COLLIDED IDEAS: (1) SIGNAL TYPE — the entry is a 4H RANGE BREAKOUT (Donchian channel break), not the EMA-cross 'momentum confluence' that the abandoned XRP-CM / SOL-HL clones used; (2) CONFIRM TIMEFRAME — the regime gate is the WEEKLY trend (a stronger, rarer, higher-timeframe filter), not the daily confirm used by every prior confluence variant, which both reduces whipsaw and only takes breakouts aligned with the dominant multi-week regime; (3) INSTRUMENT/VENUE/DIRECTION — a single high-beta USD-M perp (DOGE), long-SHORT with leverage, not the long-flat ETHBTC spot ratio that just collided/abandoned, not a two-leg pair, not a rotation basket (0/127), not a ratio pair. It uses ONLY OHLCV (multi-year Binance history — no data-availability or short-history death, no supplementary feed). It structurally answers the recurring killers: L19 DRAWDOWN BREACH — a HARD ATR stop on every trade plus a rolling-equity kill-switch (flatten and pause new entries when rolling drawdown exceeds a cap) bounds drawdown, directly fixing the 56-76% breaches that killed the un-stopped trend clones; FEE GRAVEYARD — 4H breakouts on high-beta DOGE that are weekly-trend-aligned run several percent, so per-trade capture is many times the 0.10% round trip, and the weekly gate keeps cadence moderate. 3 tunable parameters to resist the overfit graveyard.

Implements the hypothesis as specified and then reports what the data says about it, because that second part is worth more than the code here. (1) Build: the signal type is a Donchian range break rather than an EMA cross, the confirm timeframe is a genuine 1-WEEK feed (extra_instruments repeats DOGEUSDT.BINANCE so the resolver's positional pairing actually attaches it; 1-WEEK is slower than the 4H primary so the base's cross-leg alignment barrier never stalls the primary), the instrument/venue/direction are a single high-beta USD-M perp traded long-short, and there is no supplementary feed, no second leg and no basket. calculate_signal returns the continuous Donchian channel position on every bar; the weekly gate and the break test live in should_enter. Windows are rolled after the read, so no lookahead. (2) The two named killers are addressed structurally: every trade carries a hard ATR stop that ratchets and is tested intrabar, sizing is solved from that stop distance so a loss is bounded by ~0.8% of equity rather than by the equity path, and a peak-to-trough kill-switch flattens and pauses entries at a 20% drawdown. (3) THE HONEST MEASUREMENT, which the analyst should weigh before authorising an optimization run. I simulated this exact rule on the real DOGE 4H + 1W history with 0.10% round-trip fees. At the submitted defaults it produces 132 trades at avg +1.42%/trade with PF 1.38 — but the median trade is -2.6%, removing only the three largest winners flips the total from +187% to -91%, and the yearly split is 2021 +235%, 2022 -37%, 2023 -20%, 2024 +14%, 2025 +29%, 2026 -25%, with the trailing 365 days slightly negative. In other words the edge is concentrated in the 2021 mania and is thin-to-negative outside it: this is the OHLCV-trend-clone profile the hypothesis hoped to escape, and the sandbox window will most likely come back weak. (4) One finding worth flagging as a methodology point: the same rule WITHOUT the ratcheting trail (a pure fixed stop plus weekly-flip exit, which is the literal reading of 'hard ATR stop') backtests at avg +24%/trade with PF 5+ — entirely because a single 2020 entry rode DOGE for +1658% unmanaged. I added the chandelier ratchet specifically so this submission cannot report that artefact as an edge; it is the same class of spike-driven, un-exited-position headline that has been abandoned at backtest review before. (5) Defaults are taken from a well-populated, robust cell (donchian_n 30, weekly_lookback 8, stop_atr_mult 4.0) rather than the sweep argmax, which sat at the edge of the stop-width grid; the monotone preference for wider stops is a real trend-following property but I did not want to ship a boundary value.

ohlcv_trend_clone_tail_regime_concentration: single-instrument pure-OHLCV Donchian-breakout / weekly-trend directional continuation on DOGE — the OHLCV trend/breakout family with 0/213 survival (L9/L41/L63). The headline (Sharpe 1.069, PF 2.42, +69%) is a thin-right-tail artifact: skew 10.9, kurtosis 251.9, largest_win 17563 vs avg_win 1523, and the developer's own pre-submission measurement shows removing the 3 largest winners flips +187% into -91% with the edge concentrated in the 2021 mania (2022 -37%, 2023 -20%, 2026 -25%, trailing 365 days negative; rolling Sharpe decayed ~8 to ~1.7). information_ratio is -0.61 (negative active edge) and base Sharpe 1.069 with CI low 0.567 is marginal and only degrades under deflation. Not a fee death (avg_trade_return_pct 5.58%), but a directional OHLCV book's base Sharpe is its ceiling (L63) and a ~225-trial optimization on 155 trades whose PnL lives in a few 2021 outliers would curve-fit the tail and die at DSR/holdout. Not worth the 2-hour optimization run.

Long-short single-instrument directional continuation on DOGEUSDT.BINANCE USD-M. The regime gate is the WEEKLY trend, read from extra 1-WEEK bars of the same contract in on_extra_bar: weekly_dir is the sign of the return over the last weekly_lookback weekly closes. The entry is a 4H DONCHIAN RANGE BREAK (close beyond the highest high / lowest low of the prior donchian_n bars, extremes excluding the current bar), taken ONLY when the break points the same way as the weekly regime. Exits, in priority order: a hard ATR stop that ratchets chandelier-style from the best price reached and is tested intrabar against the bar's low/high; a weekly regime flip against the position; a time stop at max_hold_bars; and a portfolio drawdown kill-switch that flattens and blocks new entries when equity falls max_dd_pct from its peak, resuming only once the drawdown recovers to resume_dd_pct. Position size is solved from the initial stop distance so each trade risks risk_pct of equity, with gross notional capped. Pure OHLCV, one instrument, one venue, three tunable parameters: donchian_n, weekly_lookback, stop_atr_mult.

Set entry/stop state only after a successful order submission, or reset it when position_size() returns 0; increment _entries only on a non-zero size.

should_enter() mutates entry state (self._side, self._stop_px, self._extreme, self._entry_stop_dist, self._bars_held) and increments self._entries BEFORE the order is confirmed. If position_size() then returns Decimal('0') (notional below min_notional), no position opens but self._side is left non-zero and _entries is over-counted. The base template gates should_exit() on an actually-open position, so stale _side is overwritten at the next real breakout and does not corrupt trading — but the on_stop entries summary can overstate real fills.

No code change. Analyst should weigh regime-concentration and trade-count before authorizing optimization.

Measurability: sandbox produced 21 trades over 362 days (~21/yr), ~132 over the 6-year history. Low sample for distinguishing edge from noise, and the developer honestly discloses the edge is concentrated in 2021 (removing 3 winners flips +187% to -91%). Pure-OHLCV directional-continuation profile. Not a code defect — flagged for the analyst's backtest_review edge decision.

Clean, well-instrumented implementation with hard ATR chandelier stop and drawdown kill-switch; max_drawdown only 6.0%

avg_trade_return_pct 5.58% clears the futures fee floor — this is not a fee-viability death

155 trades over 6 years is an adequate sample; no data/short-history problem (pure OHLCV, multi-year Binance history)

OHLCV single-instrument trend/breakout clone — the 0/213-survival family (L9/L41/L63)

Returns are tail-dependent: skew 10.9, kurtosis 251.9, largest_win 17563 vs avg_win 1523; developer confirms removing 3 winners turns +187% into -91%

Edge concentrated in the 2021 mania — 2022 -37%, 2023 -20%, 2026 -25%, trailing 365 days negative per the developer's own measurement; rolling Sharpe decayed from ~8 to ~1.7

information_ratio -0.61 (negative active edge); base Sharpe 1.069 (CI low 0.567) is the ceiling and only degrades under deflation/holdout

DogeWeeklyGatedDonchianBreakoutLS was a carefully engineered long-short DOGE perp strategy that gated 4H Donchian breakouts by the weekly trend and added a ratcheting ATR stop plus a drawdown kill-switch to fix the drawdown breaches that killed earlier trend clones. On six years of data it posted a respectable-looking Sharpe 1.069, PF 2.42, +69.3% return and only 6% max drawdown across 155 trades, clearing the fee floor with an avg_trade_return_pct of 5.58%. But the profits lived almost entirely in a few 2021 outliers — skew 10.9, kurtosis 251.9, a negative information ratio, and negative recent years — which the developer's own pre-submission analysis confirmed. The analyst abandoned it at the backtest-review gate as another tail-dependent, regime-concentrated OHLCV trend clone whose marginal Sharpe would only degrade under optimization and holdout, judging it not worth the optimization run.

A high headline Sharpe/PF built on a handful of outlier winners in one regime is a mirage; directional pure-OHLCV breakout strategies should be screened for tail-concentration and out-of-regime decay before spending an optimization run.

It was abandoned at the pre-optimization BACKTEST_REVIEW gate: the analyst flagged it as a single-instrument OHLCV trend/breakout clone (a 0/213-survival family) whose edge is a thin right tail concentrated in the 2021 mania — removing the 3 largest winners flips +187% into -91%, and 2022/2023/2026 plus the trailing 365 days were negative — with Sharpe below the 1.5 threshold, so it was not deemed worth the ~2-hour optimization run.

A long-short DOGEUSDT USD-M perp strategy taking 4H Donchian range breakouts only in the direction of the prevailing weekly trend, protected by a ratcheting ATR chandelier stop and a drawdown kill-switch.

Over 2,244 days of pure OHLCV history it made 155 trades with Sharpe 1.069 (CI low 0.567), profit factor 2.42, total return +69.3%, and avg_trade_return_pct 5.58% at only 6.0% max drawdown. However returns were extremely tail-dependent (skew 10.9, kurtosis 251.9, largest win 17,563 vs avg win 1,523) with a negative information ratio of -0.61.