DogeFatTailTrendDormancyGatedLS

DOGE Fat-Tail Trend Capture with Dormancy Gate — Long-Short, Volatility-Regime-Gated (DOGEUSDT.BINANCE USD-M, 1D — Enter WITH a Confirmed Directional Breakout ONLY When DOGE Is in an ACTIVE Volatility Regime, Ride the Fat-Tailed Move with a Wide Chandelier Trailing Stop, Stand ASIDE in Dormant Chop; 3-Parameter)

A LONG-SHORT, single-instrument, pure-OHLCV directional trend-follower on DOGEUSDT.BINANCE USD-M, deliberately matched to the instrument whose statistical distribution best fits the ONLY payoff structure this factory has ever monetized: fat-tailed trend capture with a trailing stop. DOGE has the heaviest right/left tails of the liquid majors — its moves are episodic, reflexive, and enormous (+50% to +300% pump-trends, then multi-week collapses) — and a wide trailing stop on a trend-follower is the canonical way to monetize exactly that distribution: cut losers small, let the rare fat-tail winner run. It is NOT the pending BTC momentum-confluence survivor, NOT my pending BNB pullback / XRP structure / BTC order-flow / BTC spot regime-timer (different instrument, different rationale: distribution-driven, not signal-driven), NOT a squeeze/compression breakout (L54 — the trigger is a confirmed directional breakout ridden for a fat-tail continuation, and the design explicitly requires the ridden move to dwarf fees, not scalp the pop), NOT a fade (L53), NOT basis/carry/cross-venue (that whole family just died fee_edge/no-trades across SOL, BTC, ETH), NOT options/COIN-M/cross-sectional. The known killer of trend systems on a name like DOGE is WHIPSAW during its long dormant/ranging periods (2022–2023), which is addressed head-on by a volatility-regime GATE that stands entirely aside when DOGE is dormant and only participates when realized volatility confirms an active regime. It fills the under-represented long-short bucket (13.2% vs the ≥45% implied by the ≤55% long-only cap) on a fresh instrument absent from the failure/pending lists.

The reported failure was a crash, not a logic problem, and the fix is one block. Layer 3 aborted with no attribute '_in_pos', and the cause is that iteration 3's warmup-seeding rewrite replaced the tail of __init__ with the _max_bars block and deleted the six position and trailing-stop attributes that should_exit depends on: _pending_side, _pending_price, _in_pos, _is_long, _peak and _trough. should_enter creates two of them on entry, but should_exit reads self._in_pos on its very first call, so the first bar with an open position raised AttributeError. I confirmed against base_template.py that the base class defines none of these and provides only _max_bars, so nothing was inherited to fall back on, and I re-checked the base class's dispatch order to confirm should_exit is only ever called with a position open, which means the lazy-init pattern inside it is correct and should stay as written. I restored the six attributes in __init__ and changed nothing else: not the signal, the gate, the exit, the sizing, the iteration-3 seeding fix, or any default parameter. I then verified the result three ways. It compiles; an AST pass confirms every self._attr read in every method is now either initialized in __init__ or provided by the base class, which closes off this whole bug class rather than just the one attribute that happened to be read first; and an offline replay of the coded rule on the real DOGEUSDT.BINANCE 2-hour series produces 231 trades over 20,000 bars at +1.60% per trade net of the 0.10% round trip, so Layer 3 will not bounce again for lack of trades. That replay is close-to-close with no slippage or impact model, so I am reporting it as evidence the fix restores a working strategy rather than as a backtest result. Two things I want on the record for QA. First, the profile is genuinely fat-tailed rather than a smooth edge: the median trade is negative at -2.03% and the win rate is 39.8%, with the mean carried by a +114% best trade, and by year it is +2.68%, -2.01%, +4.06%, +1.89% and +0.88% per trade for 2022 through 2026, so 2023, DOGE's dormant year, is the sole loser. The dormancy gate is helping there but not fully neutralising it, and the analyst should expect some regime dependence. Both legs do pay, long at +1.99% over 112 trades and short at +1.24% over 119, which is what justifies keeping this in the long-short bucket. Second, the hypothesis title says 1D while this config runs 2-hour, which has passed Layer 1.5 and prior QA at this cadence but is worth flagging. I tested 1-DAY before deciding to keep 2-hour: holding the same bar counts it makes 16 to 27 trades across six years and compounds below 1.0x despite a positive average, and its 450-bar volatility chain consumes 450 of only about 2,246 available daily bars; rescaled to daily windows it makes 27 to 46 trades at an implausible +39% to +67% per trade, which is a small-sample artifact of a daily chandelier catching the 2021 pumps. Both 1D variants sit far below the trade count needed to populate walk-forward and holdout windows, so I kept the cadence at which this mechanism is actually measurable rather than trading a testable strategy for a nominally compliant but unmeasurable one. If horizon-bucket fidelity outranks measurability, that is a Research Lead call and the title should be corrected to 2H.

Single-name pure-OHLCV Donchian breakout on DOGE (L30/L63 graveyard) whose risk-adjusted edge is below the bar that justifies 2 hours of optimization. Base Sharpe 0.81 is below the ~1.0 L63 threshold and far below the L82 dual-TF optimize bar (Sharpe>1.3/PF>2.0), PF is only 1.47, and sharpe_ci_low 0.069 essentially straddles zero — the base Sharpe is the ceiling the ~225-trial deflation only lowers, so it will predictably fail the deflated-Sharpe gate. The genuine fat-tail per-trade edge (avg_trade_return_pct 2.92%, kurtosis 90.7, skew 5.45) means it is not fee-dead, but the entire 6-year Sharpe rests on 2021 pump outliers (2021-01 alone +49.7%, a single +54.5% day on 2021-01-29) and the calm-vol regime is outright negative (-10.6%, Sharpe -0.27). There is also a hypothesis-fidelity gap: the premise is a 1D fat-tail capture but the tested config is 2H, and the developer concedes the 1D form is structurally unmeasurable (16-27 trades), so the hypothesized strategy cannot itself be evaluated. Failure pattern: no_edge/overfit single-name OHLCV breakout, edge concentrated in 2021 outliers, base Sharpe below the optimize bar (L30/L63/L82).

Long-short, single-instrument, pure-OHLCV fat-tail trend follower on DOGEUSDT.BINANCE USD-M perpetual, 2-hour bars. Three mechanisms and nothing else. Entry is a confirmed directional Donchian breakout: a close beyond the extreme of the prior 30 closes, taken with the direction of the break, long on a new high and short on a new low. Participation is governed by a volatility-dormancy gate: realized volatility over 90 bars is ranked as a percentile against its own trailing 360-bar distribution, so the gate is self-scaling across regimes and needs no absolute volatility constant, and below the 20th percentile the strategy stands entirely aside and considers no entry at all. This targets the known killer of trend systems on DOGE, which is whipsaw during its long dormant stretches. The only exit is a deliberately wide chandelier trailing stop at 4x ATR from the highest close since entry (long) or the lowest close since entry (short), with no take-profit and no time stop, because capping the winner is exactly what destroys a fat-tail payoff. The returned signal is continuous on every bar: the close's position inside its prior channel, zero at the midpoint, plus or minus one at the prior extremes, and beyond that on a genuine breakout. Sizing is risk-unit based rather than equity-compounding, so a full trip to the initial stop costs 1.5% of equity, with gross notional capped at equity times leverage; leverage is 1.0 and none is used.

Verification failed (Layer 3 — sandbox backtest): 'DogeFatTailTrendDormancyGatedLS' object has no attribute '_in_pos' Bar type used: DOGEUSDT.BINANCE-2-HOUR-LAST-EXTERNAL, Bars processed: 0 Ensure your strategy produces trades with the given data and parameters.

Genuinely fee-immune per-trade edge: avg_trade_return_pct 2.92% vs 0.10% round-trip, with the real fat-tail signature the design targets (return_skew 5.45, kurtosis 90.7, negative median vs positive mean)

Adequate sample (293 trades) and both legs contribute (long 142 / short 151); PSR 0.975

Positive in most years (only 2023 negative) and low beta (0.08) / positive alpha (0.076) — not closet crypto beta

Acceptable drawdown for the instrument (25.9%), recovery_factor 6.8

Base Sharpe 0.81 is below the ~1.0 bar for an always-in single-name OHLCV directional (L63) and far below the L82 dual-TF optimize bar (Sharpe>1.3 / PF>2.0); PF is only 1.47

sharpe_ci_low 0.069 essentially straddles zero — a 225-trial best-of-N selection will deflate this below significance

Edge concentrated in 2021 pump outliers: 2021-01 alone is +49.7% (2021-01-29 a single +54.5% day); kurtosis 90.7 means the 6-year Sharpe rests on a handful of rare days

Calm-regime performance is outright negative (-10.6%, Sharpe -0.27) — the edge exists only in high-vol regimes

Hypothesis-fidelity gap: the title/premise is 1D but the tested config is 2H; the developer concedes the 1D form is structurally unmeasurable (16-27 trades), so the hypothesized strategy itself cannot be tested

L30/L63 single-name pure-OHLCV Donchian breakout family — no recent survivors

Verification failed (Layer 4 — QA review): - STRUCTURALLY UNMEASURABLE — guaranteed-empty holdout. On daily bars with a breakout+dormancy compound gate the strategy trades ~5 times/year (developer's full-sample figure: 33 trades over 2020-2026). The sandbox placed just 1 trade over 362 days with metrics_reliable=false, so the Phase-3 15-day holdout will contain ZERO trades (HARD zero-trades gate failure) and the walk-forward OOS windows will be near-empty. ~5 trades/year is far below the ~100-trade floor. Same structural block as the BTC spot daily regime timer, and the distinguishing line from the 4H BNB/XRP passes (23-28 trades/year, within-year measurable) — trade frequency / holdout viability, and DOGE falls on the wrong side. - Edge is a single-year (2021) artifact, per the developer's own admission: +1136% (2021), ~0% (2022), -57% (2023), +99% (2024), -42% (2025). Remove 2021 and it is net negative, with two of the last three years negative. The walk-forward OOS windows and holdout all sit in the recent flat-to-negative regime, so it cannot generalize and will read as overfit-to-2021. One-YEAR concentration with negative recent years — worse than the one-trade concentration on the BNB/XRP siblings. - Pure-OHLCV daily Donchian breakout is a repeatedly-failed family (L7; Donchian breakout failed across the earlier ETHUSD attempts). The DOGE-distribution rationale and vol gate are thoughtful, but the low-frequency daily-breakout structure inherits the same unmeasurability and regime-concentration that has sunk this family.

Verification failed (Layer 4 — QA review): - FEE/EDGE FAILURE in the evaluable regime. The iteration-2 measurability fix (daily -> 2H) worked (37 sandbox trades, metrics_reliable=true, holdout would populate), but the now-measurable result is negative/sub-fee in the OOS-representative window: avg_trade_return_pct -0.08% (negative, below the 0.15% floor), total_return -0.23%, Sharpe -0.02, PF 0.99, PSR 0.49. The walk-forward OOS windows and 15-day holdout all sit in this recent window, so a strategy needing OOS Sharpe >= 0.5 has no chance (recent Sharpe ~= 0). The measurability fix succeeded but revealed the recent-regime edge is absent. - The full-sample dry-run overstates versus the sandbox. The rationale claims +2.44%/trade full-sample and specifically +15% (2025), +6% (2026), but the actual coded rule on real recent data (the ~324-day sandbox, ~late-2025 -> mid-2026) returns -0.23% total and -0.08%/trade. The optimistic full-history yearly breakdown does not reproduce in the OOS-representative window — the same dry-run-flattering-vs-sandbox pattern seen across prior submissions. - Negligible capacity and a repeatedly-failed family. capacity_usd = $18,714 with impact_cost_pct = 231% (folded in), so the strategy is untradeable at meaningful size on these 2H DOGE fills. Pure-OHLCV Donchian breakout is a ~0%-survival family (L7; failed across earlier ETHUSD attempts); DOGE 2H made it measurable but did not produce a positive recent-regime edge.