SolSpotLiquiditySweepReclaimLong

Smart-Money Liquidity-Sweep Reclaim, Long-Only Spot (Single-Instrument SOLUSDT.BINANCE_SPOT)

Pure-price cousin of the factory's liquidation-cascade reversal, distilled from the TradingView Smart-Money-Concepts archetype (liquidity sweeps / stop-runs). Identify a prior significant swing low that clusters resting stops; enter LONG when price SWEEPS below that level (triggering the stops) but then FAILS to hold and RECLAIMS back above it within a short confirmation window = a stop-run overshoot that traps late sellers, primed to revert. This is a PRICE-STRUCTURE signal, so it works on BINANCE_SPOT where there is no liquidations feed — extending the liquidation idea to a new venue and asset with no short leg required. Reversion take-profit toward the pre-sweep range, ATR stop below the sweep extreme. Must clear the spot fee floor: avg per-trade return > 0.25% of notional. 3 parameters: swing-low lookback, reclaim-confirmation window, ATR stop multiple.

Implements the hypothesis exactly: a pure price-structure liquidity-sweep reclaim, long-only on spot where no liquidations feed exists, with a reversion target toward the pre-sweep range, an ATR stop below the sweep extreme, and three searched parameters (swing_lookback, confirm_window, atr_stop_mult). Spot venue rules are respected structurally — should_enter can only ever return BUY, leverage is 1.0, and position_size deliberately omits any leverage multiplier so the position cannot exceed the cash held (no leverage_set_but_unused risk since leverage is left at 1.0). The signal follows the coder guidance on graded scores: rather than the banned 'return 0.0 unless the pattern fires' shape, recl varies every bar and the tanh sweep-gate merely grades it, which I verified on real data — 464 distinct finite values over a sandbox-sized 2190-bar window, range [-3.76, +2.03]. Parameters were calibrated on a 6-year (2020-2026, 13,085 bar) SOLUSDT spot event study with real spot fees (0.10% x 2 = 0.20% round trip) rather than guessed. Two independent monotone gradients support the mechanism rather than a single lucky cell: the edge improves monotonically with swing_lookback (1-HOUR: -0.28% -> -0.27% -> ~0.00% for 24/48/96; 4-HOUR: ~-0.10% -> +0.20% -> +1.01%..+1.28%), and 4-HOUR dominates 1-HOUR at every lookback — exactly what you expect if a swing low must be genuinely significant to hold meaningful stops and the reversion needs room to clear the spot fee floor. At the shipped settings the full-sample result is +1.012% net per trade over 158 trades at PF 1.28, comfortably above the 0.25% spot floor. Importantly I did NOT ship the argmax: swing_lookback 96 / confirm_window 3 / atr_stop_mult 1.0 scored higher (+1.277%) but is negative in the 2022-2023 sub-period, whereas the shipped 96/2/0.5 is positive in all four multi-year sub-periods (+2.60%, +0.267%, +1.409%, +0.578%), so I chose consistency over peak mean. Entry frequency is ~2.9% of bars (60 setups per 2190-bar window), so Layer 3 will comfortably clear its >=1 trade gate, and warmup is 117 bars (~5% of a 365-day 4H window) so it is not a late-warmup no-trade risk. One honest caveat for the Analyst: the trailing 365-day slice — which is exactly the Layer-3 sandbox window — is slightly negative (-0.219% over ~29 trades) for every variant I tested, including the robust one. I deliberately did not retune to fix that, because tuning to the sandbox window is the precise deflated-Sharpe/PBO trap that has killed several strategies in this factory. The 6-year record and the four positive multi-year sub-periods are the real evidence; if the walk-forward and holdout do not confirm the swing_lookback gradient out-of-sample, abandon rather than iterate on parameters.

Net loser that should not be optimized: profit_factor 0.97 (<1.0), total_return -5.9%, Sharpe -0.023 (CI [-0.65, 0.64], probabilistic_sharpe 0.47), expectancy -$38.8/trade, and information_ratio -0.86. Win rate is only 37% and the occasional large wins do not cover the frequent losses (0.37×$3270 − 0.63×$1983 ≈ −$39/trade). The reported avg_trade_return_pct of 1.14% is contradicted by the negative dollar expectancy and sub-1.0 profit factor — the actual PnL loses across the sample, with a mixed, net-negative annual record (2020 -11.7%, 2022 -20.2%, 2026 -8.8%). This is a pure-price mean-reversion/reclaim (dip-buy) strategy, the reversion family that has repeatedly died here (multiple sweep/fade/reversion candidates abandoned this session); the smart-money liquidity-sweep framing did not turn it into a real edge. Per L8, never advance when total_return ≤ 0 or PF < 1.0. There is no parameter region to tune a PF-0.97/Sharpe-0 signal into significance, and best-of-225 selection would only overfit the handful of winning setups and fail deflated-Sharpe/holdout. Abandon at BACKTEST_REVIEW rather than spend 2 hours optimizing.

Long-only spot strategy on SOLUSDT.BINANCE_SPOT (4-HOUR bars, no leverage) that buys the reclaim of a swept swing low. Each bar it identifies the prior significant swing low as the lowest low of a `swing_lookback` window that ENDS `confirm_window` bars ago (causal, so the sweep cannot define the level it is sweeping), plus that window's high as the pre-sweep range top. It then measures depth = (level - sweep_low)/ATR (how far the stop-run overshot below the level) and recl = (close - level)/ATR (how far price has climbed back), and returns signal = 0.5*(1+tanh(depth/depth_scale)) * recl — a graded conviction score that varies every bar, is strongly positive only when a genuine sweep has been reclaimed, negative while price is still below the level, and near zero when the level was never breached. Entry requires the structural facts (something traded below the level AND the close is back above it), the graded score clearing entry_thresh, and the reversion target still being above the current price; it then goes LONG only. Exits, all frozen at entry: an ATR stop placed below the SWEEP EXTREME rather than below the entry (the stop-run low is what the reclaim must defend), tested intrabar and checked before the target; a reversion take-profit at the midpoint of the pre-sweep range (falling back to the range high), tested intrabar; and a 48-bar (~8 day) time stop. Sizing risks 2% of equity at that stop, hard-capped at 50% of equity with no leverage multiplier since a CASH account cannot hold more than its cash.

Only real-engine evidence (sandbox) is −0.838%/trade vs 0.25% spot floor, total −11.1%, PF 0.66 over 26 trades. All positive numbers come from a developer event study, not the engine → unverified.

Pure-OHLCV + liquidation-cascade-reversal class → poor survival history; single event study is weak evidence for this class.

Event study was optimistic vs engine on the one comparable window (−0.219% predicted vs −0.838% delivered); treat +1.012% full-history as an upper bound.

Correct venue routing (long-only/no-leverage → BINANCE_SPOT); pure-price structure, no missing-data path; deliberately 3 tunables; adequate trade count (154)

Thoughtful causal sweep/reclaim construction with graded conviction score

profit_factor 0.97 (< 1.0) — the strategy loses money; total_return -5.9%

Sharpe -0.023 (CI [-0.65, 0.64]); expectancy -$38.8/trade; information_ratio -0.86 — no edge

win_rate 37% with losses outweighing the occasional large wins → net-negative expectancy

Reported avg_trade_return_pct 1.14% is contradicted by the negative dollar expectancy and PF<1.0 — the actual PnL loses

Mixed, net-negative annual record (2020 -11.7%, 2022 -20.2%, 2026 -8.8%); max_drawdown 24.5% (CI high 67%)

Mean-reversion/reclaim (dip-buy) family — repeatedly dead here; the liquidity-sweep framing did not produce an edge

This strategy extended the factory's liquidation-cascade reversal idea to the spot venue as a pure-price cousin, distilling the TradingView Smart-Money-Concepts archetype: buy the reclaim of a swept swing low, where a stop-run below a stop cluster overshoots, traps late sellers, and reverts. The construction was thoughtful — correct spot venue routing, a causal (no-lookahead) sweep/reclaim detector, and a graded conviction score — but the backtest showed no edge: profit factor 0.97, a -5.9% return, a Sharpe of essentially zero, negative expectancy, and a 37% win rate whose rare large wins failed to cover the losses. The analyst abandoned it at backtest review as another net-negative member of the repeatedly-dead mean-reversion/reclaim family, noting the smart-money framing did not produce a real edge and that the headline 1.14% per-trade figure was contradicted by the losing dollar PnL, with nothing to optimize toward.

Relabeling dip-buying with a fashionable smart-money 'liquidity sweep / reclaim' structure does not create an edge — it is still a pure-price mean-reversion/reclaim strategy in the repeatedly-dead reversion family, and a headline avg_trade_return_pct can be flatly contradicted by negative dollar expectancy and a sub-1.0 profit factor.

It was abandoned at the BACKTEST_REVIEW gate under the rule against advancing when total return ≤ 0 or profit factor < 1.0: a Sharpe near zero, negative expectancy, and negative information ratio mean no edge, and the smart-money liquidity-sweep framing did not turn a pure-price dip-buy into a profitable one. With PF 0.97 there is no parameter region to tune into significance, so it never reached optimization or analysis.

A long-only, single-instrument, pure-price spot strategy on SOLUSDT.BINANCE_SPOT (4H bars) that trades the Smart-Money-Concepts liquidity-sweep reclaim: identify a prior significant swing low where stops cluster, and enter long when price sweeps below it (triggering the stops) but reclaims back above within a short confirmation window (a trapped-seller stop-run overshoot primed to revert), targeting the pre-sweep range midpoint with an ATR stop below the sweep extreme.

It was a net loser with no edge: profit factor 0.97, total return -5.9%, Sharpe -0.023 (CI [-0.65, 0.64]), expectancy -$38.8/trade, information ratio -0.86, max drawdown 24.5% over 154 trades. Win rate was only 37% and the occasional large wins did not cover the frequent losses; the reported avg_trade_return_pct of 1.14% was contradicted by the negative dollar expectancy and sub-1.0 profit factor, with a net-negative annual record (2020 -11.7%, 2022 -20.2%, 2026 -8.8%).