DogePriorDayRangeFailedBreakoutTrapReversalLS

Prior-Day-Range Failed-Breakout Trap Reversal, Long-Short (Single Instrument DOGEUSDT.BINANCE Perp, 1h Bars, Fade FALSE Breakouts of the Prior-Day High/Low That Close Back Inside the Range — Reversion to Range Midpoint, Hard Stop Beyond the Poke, 3-Parameter)

A LONG-SHORT, single-instrument intraday REVERSAL strategy on DOGEUSDT.BINANCE (USD-M perp, 1h bars) that fades FAILED breakouts of the prior-day range — the classic 'bull trap / bear trap'. The prior-day high (PDH) and prior-day low (PDL) are obvious levels where breakout traders and resting stops cluster; when price pokes beyond them but fails to hold and closes back INSIDE the range, the trapped breakout crowd is forced to unwind, driving a reversion back toward the range midpoint. This is the deliberate OPPOSITE polarity of the breakout-CONTINUATION strategies that repeatedly blew up (SOL vol-expansion breakout -100% liquidation, SOL Donchian trend-ride) — those bet WITH the breakout and lost to false signals; this monetises exactly those false signals. It is also distinct from the deployed VWAP-band reversion (that fades statistical band deviations; this fades a discrete level-based trap event). DOGE is chosen for its very high liquidity and heavy retail participation, which makes PDH/PDL false breakouts frequent and reversion-prone. Kept to 3 parameters to resist overfitting (the #1 killer, 239x), single-instrument to avoid the basket equity-accounting bugs that killed the ETH/BTC and breadth baskets, and hard-stopped to prevent the leverage blow-ups that liquidated prior long-short perp strategies.

Implements the bull-trap/bear-trap edge: PDH/PDL are the levels where breakout stops cluster, so a poke-and-fail (close back inside the range) traps the breakout crowd and forces an unwind back toward the range midpoint. This is the deliberate OPPOSITE polarity of the breakout-continuation systems that repeatedly failed to false signals — here the false signal IS the entry trigger — and it is level-event-driven rather than a statistical-band fade, keeping it distinct from the deployed VWAP-band reversion. DOGE is chosen for very high liquidity and heavy retail participation, which make PDH/PDL false breakouts frequent (good measurability on 1h bars) and reversion-prone. Single-instrument by design to avoid the multi-leg equity-accounting bugs that killed the basket strategies, and every trade is HARD-STOPPED just beyond the poke with conservative 0.5x-gross, leverage-1.0 sizing so a stop-out costs a small fraction of equity and cannot cause the leverage liquidations that wiped prior long-short perp attempts (avg loss ~ risk_per_trade x equity). Kept to exactly 3 tuned parameters (poke_pct defines a genuine poke, stop_pct sets the hard stop, target_frac sets the reversion exit) to resist overfitting; the time cap, risk fraction, and gross cap are fixed structural choices. Uses the base single-instrument on_bar/dispatch unchanged: calculate_signal derives PDH/PDL by scanning the buffered bars for the last completed UTC day (robust to warmup, no incremental-state drift), caches the trap booleans + bracket levels, and returns a continuous location-in-range score that varies every bar; should_enter (called only while flat, so no pyramiding) fires the fade, and should_exit runs the entry-anchored target/stop/time bracket seeded via the _was_in_pos pattern. Fee headroom is ample: reverting a poke at PDH to the midpoint is half the prior-day range (typically 1.5-2.5% on DOGE), an order of magnitude above the ~0.10% USD-M round-trip, against a ~0.5-0.8% hard-stop risk (favorable reward/risk).

Account LIQUIDATED on 2025-05-24 (total_return -100%, max_drawdown 100%), fatal for a hypothesis whose whole premise was being hard-stopped against blow-ups. The edge is decisively negative and regime-universal: on 1,573 well-sampled trades avg_trade_return_pct is -0.063%, profit_factor 0.826, expectancy -$67/trade, and EVERY year is red (2020 -11%, 2021 -47%, 2022 -5%, 2023 -7%, 2024 -47%, 2025 -123%). Position sizing is also broken — avg_position_pct 263% despite the claimed 0.5x notional cap, with 2025 daily returns of -360%/-8033%/+443% as equity went negative and sizing off collapsing equity exploded (return_kurtosis 677) — which is what drove the liquidation. Fixing the leverage defect cannot create edge from a mechanism that loses in all six years. This is the intraday reversion-fade class (fading failed prior-day-range breakouts) that has died repeatedly this session; optimizing poke_pct/stop_pct/target_frac cannot flip negative expectancy, and there is no salvageable reframe of DOGE failed-breakout fading. Abandon rather than iterate or revise.

Long-short intraday reversal on DOGEUSDT.BINANCE (USD-M perp), 1h bars, that fades failed breakouts of the prior-day range. Computes the prior completed UTC day's high (PDH) and low (PDL). Upside trap -> SHORT: the bar's high pokes >= poke_pct (0.15%) above PDH but the bar closes back below PDH (and above the range midpoint); downside trap -> LONG: the low pokes >= poke_pct below PDL but closes back above PDL (and below midpoint). Each trade is bracketed and entry-anchored: reversion target = target_frac of the way from entry to the range midpoint, a HARD STOP stop_pct beyond the poke extreme, and a 24-bar time cap. Sizing risks 1% of equity over the hard-stop distance, capped at 0.5x equity gross (leverage 1.0, real margin buffer so a stop-out cannot liquidate). 3 tuned params (poke_pct, stop_pct, target_frac).

Seed the bracket from the trap that caused the entry: capture stop/target at should_enter time (or stash keyed to the pending entry) and attach on fill, instead of reading mutable self._last_trap in _track_position.

Intermittent bracket-seeding bug. The entry-anchored stop/target are seeded in _track_position from self._last_trap on the FIRST bar the position is visible — the bar AFTER the trap/entry bar (on the trap bar, _track_position runs during calculate_signal BEFORE should_enter submits, so the position isn't yet open and _was_in_pos stays False). calculate_signal recomputes traps each bar and overwrites _last_trap when a NEW trap forms. So if a fresh trap forms on the bar right after entry, _last_trap is overwritten before _track_position seeds the open position, and that trade gets stop/target from the wrong (later) trap. Intermittent (only on consecutive trap bars), cannot blow up (risk bounded by the 0.5x cap), not the driver of the loss — but a genuine correctness defect.

Analyst: 0.062%<0.15% floor and PF 0.90 argue for abandonment. If optimized, require a stricter trap gate (larger poke_pct, volume confirmation, wider target) to lift per-trade return clearly above 0.15% OOS; fix the seeding bug first.

Marginal/net-negative edge — analyst should apply the futures floor and most likely ABANDON. On 253 well-sampled trades (metrics_reliable=TRUE): total_return -12.8%, PF 0.90, win_rate 0.486 (near coin-flip), avg_win 945 ≈ avg_loss 993, avg_trade_return_pct +0.062% (net-positive equal-weighted but dollar expectancy -$50.83, below the 0.15% floor), Sharpe -0.30 (CI [-1.93,+1.25] straddles zero). The failed-breakout fade is essentially a coin flip with slightly unfavourable payoff; the tiny gross edge (~0.16%/trade) clears the ~0.10% round-trip so it's NOT a gross-can't-clear-fees death, but it doesn't net positive. High turnover (159) makes it fee/impact-sensitive (total_impact 8,176).

Well-sampled (1,573 trades, balanced 805 long / 768 short) so the negative verdict is statistically decisive.

Account LIQUIDATED on 2025-05-24 (total_return -100%, max_drawdown 100%, CI low 88%) — directly contradicts the hypothesis's 'hard-stopped, no blow-ups' claim.

Negative per-trade expectancy on 1,573 trades: avg_trade_return_pct -0.063%, profit_factor 0.826, expectancy -$67/trade.

EVERY year negative: 2020 -11%, 2021 -47%, 2022 -5%, 2023 -7%, 2024 -47%, 2025 -123% — no profitable regime exists.

Sizing broken: avg_position_pct 263% despite the claimed 0.5x cap; 2025 daily returns of -360%/-8033%/+443% show equity went negative and sizing exploded (kurtosis 677).

Intraday reversion-fade class that has died repeatedly; Sharpe 0.52 is a pre-blowup artifact with CI [-0.23, 0.92] straddling zero.

The strategy set out to monetize false prior-day-range breakouts on DOGE — deliberately taking the opposite side of the breakout-continuation strategies that had repeatedly blown up — using a single instrument, 3 parameters, and hard stops to avoid the leverage failures of prior attempts. Backtested over 1,884 days it generated 1,573 balanced trades but lost money every single year, with negative per-trade expectancy (-0.063%, profit_factor 0.826) and, fatally, liquidated the account to -100% on 2025-05-24. The liquidation traced to broken sizing — average position notional hit 263% of equity despite a claimed 0.5x cap — the very blow-up the hypothesis promised to prevent. The analyst abandoned it at the backtest-review gate before any optimization, judging the negative edge regime-universal and unfixable, so later pipeline stages were never reached.

A hard-stop premise is meaningless if position sizing is defective (avg_position_pct 263% vs a 0.5x cap drove the blow-up), and fixing leverage cannot manufacture edge from an intraday reversion-fade mechanism that loses across every regime.

The analyst issued an 'abandon' verdict at the pre-optimization BACKTEST_REVIEW gate: the account was liquidated (max_drawdown 100%) despite the strategy's whole premise of being hard-stopped, expectancy was decisively negative across all six years, and position sizing was broken (avg_position_pct 263% vs the claimed 0.5x cap).

A long-short single-instrument (DOGEUSDT.BINANCE perp, 1h) reversal strategy that faded FAILED breakouts of the prior-day high/low — buying/shorting the trapped breakout crowd as price closed back inside the range and reverted toward the midpoint, with a hard stop beyond the poke and only 3 tuned parameters.

Over 1,884 data days it took 1,573 well-sampled trades (805 long / 768 short) with a 47.9% win rate, but delivered a total_return of -100% (account liquidated 2025-05-24), profit_factor 0.826, and avg_trade_return_pct of -0.063% — negative in every year (2020 -11%, 2021 -47%, 2022 -5%, 2023 -7%, 2024 -47%, 2025 -123%).