XRP Perp Failed-Breakout Reversal, Long-Short (Single-Instrument XRPUSDT.BINANCE USD-M — Fade LIQUIDITY-GRAB Breakouts That Fail: When Price Pierces an N-Bar High/Low Then Closes Back Inside the Range, Trade the Trapped-Trader Unwind, Defined-Risk Stop at the Failure Extreme, 3-Parameter)

A LONG-SHORT, SINGLE-INSTRUMENT, SINGLE-VENUE, PURE-OHLCV reversal strategy on the XRPUSDT.BINANCE USD-M perpetual built on the failed-breakout / liquidity-grab structure — a mechanism distinct from momentum, from generic RSI/Bollinger mean-reversion, from carry, and from anything requiring options/cross-venue/multi-leg machinery (the complexity that killed my last several ideas in verification Layers 2–4). The idea: breakouts of obvious N-bar highs/lows attract momentum entries and resting stop orders just beyond the level; when such a breakout FAILS — price pierces the level but closes back INSIDE the prior range within a few bars — those breakout buyers/sellers are trapped and their forced unwind fuels a sharp move back the OTHER way. The strategy SHORTS a failed upside breakout (trapped longs) and LONGS a failed downside breakout (trapped shorts). This is a genuine market-structure edge (stop-hunt/liquidity-grab), not a fitted price pattern, and it deliberately mirrors the ONLY archetype that has survived this factory: single-instrument, single-venue, pure-OHLCV, long-short, defined-risk, trailing-managed. It is LOW-FREQUENCY (4h bars, multi-day holds) so the ~0.10% USD-M round trip is negligible against multi-percent reversal targets (avoids the fee_edge graveyard), and it has a HARD, well-defined invalidation (the failure extreme) with a 2x leverage cap so it cannot produce the liquidation/-100% blow-up that abandoned the prior single-instrument MR strategy. XRP is chosen as a liquid, level-respecting, non-BTC/ETH/SOL/DOGE symbol (diversifying my pending set's instruments) whose breakouts frequently fail and reverse cleanly. 3 parameters only, to resist the overfit deaths that dominate the log.

Iteration 2 fixes the four QA findings in place; the signal logic that passed Layers 1-3 is byte-for-byte unchanged apart from caching bar.high/bar.low. (1) Sizing degeneration — the root cause QA identified was that should_enter imposed no minimum on |close - grab_extreme|, so a close resting on the failure extreme drove qty toward infinity until only the 2x leverage ceiling caught it. The stop LEVEL is now floored (min(grab_low, close - min_stop_atr*ATR) for longs, symmetric for shorts) so the level used by the exit and the distance used by the sizer are the same widened number, and position_size re-applies the same floor defensively. (2) Leverage amplifying the tail — QA showed the leverage cap was the binding sizing rule on precisely the trades that produced the -$32.8k loss, so leverage drops to 1.0 and the cap becomes max_notional_frac x equity; with a >=0.75-ATR stop (~1.3% on 4h XRP) and 1% risk, notional lands near 0.6-0.8x equity, making the risk budget — not a margin ceiling — the binding constraint. A 3x-stop gap now costs ~3% of equity instead of 33%. (3) Close-only stop — the stop is now tested against the bar's intrabar low/high so a bar that trades through it exits on that bar rather than 4 hours later; the target stays close-based so the asymmetry is conservative (stops early, targets late) and cannot flatter the backtest. (4) Inverted reward:risk — rather than hope the mechanism reverts, entries now carry a hard geometry gate: distance-to-midpoint >= min_rr x stop distance, which rejects the small-capped-win / fat-loss shape that produced profit_factor 0.50 and avg_loss 2x avg_win; and the breakeven trail arms at 1R of the trade's own risk instead of a fixed 1 ATR, so it can no longer cut a winner before it has earned its risk back. One honest caveat, echoing QA: these changes make the risk genuinely defined and the payoff geometry non-negative by construction, but they do not by themselves prove the failed-breakout reversal has positive expectancy on XRP — the min_rr gate is the parameter that will show it, and if the backtest is still a net loser with the tail contained, the mechanism itself should be abandoned rather than re-tuned. Parameter count grows from 4 to 7, but all three additions are risk knobs (min_stop_atr, min_rr, max_notional_frac); the 3 signal parameters the hypothesis specifies are untouched, so overfit surface on the edge itself is unchanged.

Decisive net loser with negative expectancy across the full sample. profit_factor 0.657 (<1.0), total_return -40.9%, avg_trade_return_pct -0.139% (negative), and expectancy -$134/trade over a decisive 303 trades, losing in every recent year (2023 -7.7%, 2024 -21.6%, 2025 -13.9%, 2026 -8.0%) with rolling Sharpe negative for most of the history. The iteration-2 redesign fixed the risk PLUMBING (no liquidation, largest loss cut from -$32.8k to -$7.7k, max_dd 55.8%->45.8%) but did NOT fix the EDGE: reward:risk is still inverted (avg_win $476 < avg_loss $844 at a 53.8% win rate), so profit_factor only moved 0.50->0.657 and stays below 1.0 — the capped midpoint target vs. running losers remains structurally negative, exactly the QA finding it was meant to cure. Fading failed breakouts on XRP simply has no positive expectancy net of costs; this is the single-name mean-reversion/fade family that has repeatedly died (the sibling MR strategy liquidated). There is no parameter region to optimize toward when PF is 0.66 and per-trade expectancy is negative across every regime — tuning would only overfit the two marginally-positive early years. Abandon per the PF < 1.0 / negative-return / negative-avg_trade_return_pct rules rather than spend 2 hours optimizing.

Verification failed (Layer 4 — QA review): - THE DEFINED-RISK STOP DOES NOT DELIVER DEFINED RISK, and the sandbox proves it: total_return -41.9%, max_drawdown 55.8% (CI to 85.8%), and a single largest_loss of -$32,788 (~33% of a ~$100k account, roughly 16x the intended 2% risk_per_trade budget). The hypothesis's central safety premise -- 'HARD, well-defined invalidation ... 2x leverage cap so it cannot produce the liquidation/-100% blow-up' -- is falsified. Root cause is a risk-sizing degeneration: position_size sets qty = equity*risk_per_trade / stop_distance where stop_distance = |close - grab_extreme|, but should_enter imposes NO minimum on that distance (a long only requires close > grab_low). When the close sits just inside the grab extreme, stop_distance -> ~0, qty explodes, and is capped only by max_qty = equity*leverage/price = 2x notional. Those tight-stop trades are therefore all sized at the 2x leverage ceiling (the developer's own dry-run: 22.4% of entries above 1x notional, 1.7% clamped at 2x). Compounding this, the 'stop' is not a stop-market order -- it is a CLOSE-based check in should_exit (close <= stop) evaluated once per 4-hour bar -- so on a violent bar price gaps far past the stop and the exit fills at the next close, many multiples of stop_distance away. A 2x-notional position exiting ~16% beyond a tiny stop = the observed 33%-of-equity single loss. The intended 2%-at-stop risk is thus illusory for the very trades that matter (the tail), which is exactly how a strategy that claims it 'cannot blow up' prints a 55.8% drawdown. - Inverted reward:risk of the breakout-fade / mean-reversion class. The target is the reference-range MIDPOINT (a bounded, small move) while the loss side is a stop that -- per the finding above -- can run far beyond its nominal distance. The sandbox shows the signature: avg_win $712 vs avg_loss $1436 (losers ~2x winners) at a 50% win rate => structurally negative expectancy (profit_factor 0.50, avg_trade_return_pct -0.079%), with return_skew -1.12 and kurtosis 12.5 (a few large losses dominate). This is the same fade geometry (small capped wins, fat-tailed losers) that has repeatedly been abandoned as a net loser in this factory. Even with the sizing defect fixed, the reviewer should verify the failed-breakout reversal actually reverts on XRP net of the trailing-to-breakeven behaviour before trusting it -- the mechanism being 'market-structure' does not by itself grant positive expectancy.

RISK FIXES VERIFIED, BUT THE SANDBOX SHOWS NEGATIVE EXPECTANCY — a likely abandon at backtest_review. The four risk-control defects I raised on iteration 1 are all genuinely fixed and the metrics prove it: max_drawdown fell 55.8% -> 11.7%, the largest single loss fell from -$32,788 (~33% of equity, 16x the risk budget) to -$3,593 (~3.6%, i.e. ~3x the 1% risk as designed), and avg_position_pct fell 91% -> 57% (no over-sizing). Defined risk is now real. HOWEVER, with the tail contained the strategy is a decisive net loser on the 362-day sandbox: total_return -7.1%, profit_factor 0.62, Sharpe -0.55 (CI [-2.06, 1.14]), and avg_trade_return_pct -0.359% (NEGATIVE — below any fee floor because the mechanism itself loses, not because fees eat a small move). The min_rr>=1.3 geometry gate contained the loss shape but did not produce positive expectancy: avg_win $611 is still below avg_loss $749 at a 43% hit rate, so the failed-breakout does not revert profitably on XRP in this sample. Per the developer's own stated criterion ('if the backtest is still a net loser with the tail contained, the mechanism itself should be abandoned rather than re-tuned'), this is a prime abandon candidate. It is not a QA code-correctness defect — the code is now correct and the edge call belongs to the analyst on the full multi-year backtest — but the analyst should weight the negative per-trade expectancy heavily and not re-tune the tail-contained loser.