SolSpotCapitulationBounceReversionLong

SOL Spot Capitulation-Bounce Reversion, Long-Only Intraday (Single-Instrument SOLUSDT.BINANCE_SPOT, 15m Bars — Fade Forced-Seller EXHAUSTION: Buy a Sharp Volume-Climax Flush Only in an Uptrend Regime, Exit on Reversion or Time-Stop, 3-Parameter)

A LONG-ONLY, single-instrument, INTRADAY mean-reversion strategy on SOLUSDT.BINANCE_SPOT that fades FORCED-SELLER EXHAUSTION rather than trading price momentum. The mechanism is microstructural, not a generic RSI/Bollinger dip-buy: it fires only on a genuine capitulation FLUSH — a sharp multi-bar drawdown accompanied by a volume climax (volume z-score spike) and, when the taker feed is available, extreme taker-SELL dominance that is beginning to normalize (aggressive sellers exhausting). Forced/panic sellers on a liquid high-volatility name overshoot intrabar, and liquidity providers who absorb that flow are paid a reversion premium as price snaps back. A long-term trend-regime gate (price above its slow moving average) restricts entries to dips WITHIN an uptrend, so the strategy buys absorbable flushes in bull phases and stands aside during structural downtrends (no falling-knife catching). This is NOT momentum (the exhausted BTC-perp OHLCV classes L9/L38/L41), NOT a pair/ratio trade (L40), NOT funding/options (my other pending hypotheses), and it is single-venue single-instrument so there is no cross-venue leg to swap (the failure mode that killed three prior ideas in QA). Fills three under-represented buckets at once: BINANCE_SPOT venue (12.0% vs ≥15% target), short-horizon (8.9% vs ≥10%, 15m bars), and a fresh MECHANISM (capitulation reversion) absent from my funding/options-heavy pending set. SOL is chosen because its capitulation flushes are large in ABSOLUTE terms (routinely >0.6-1.5% snap-backs), giving a real cushion over the 0.25% spot fee floor — this is an absolute single-name bounce, not a beta-cushioned ratio trade.

Implements the hypothesis mechanism directly: (1) FLUSH — a sharp multi-bar drawdown measured as a scale-free z-score of the flush_lookback-bar return, so the trigger adapts across volatility regimes instead of using a fixed absolute drop (per the lesson that absolute thresholds on any feed produce degenerate trade counts); (2) VOLUME CLIMAX — a volume z-score against the same trailing baseline, so only size-printing panic bars qualify, not quiet drifts; (3) TAKER-SELL EXHAUSTION — the recent PEAK taker-sell share z-scored against its baseline, measured just before entry so we are buying after aggression has peaked and begun to normalise. Crucially the three terms are SUMMED into ONE score thresholded once, rather than ANDed as three independent gates — that avoids the multiplicative selectivity that has repeatedly produced zero trades in the short Layer-3 window. The taker term is purely additive and degrades to 0.0 when the feed is absent (spot instruments may not receive the futures taker_ratio feed), so the strategy never silently switches edge mechanism and never goes trade-free on a missing feed. The trend-regime gate (close > 192-bar SMA) is the hypothesis's falling-knife guard and is roughly a 50% filter, not an 80%+ one. Venue is BINANCE_SPOT as specified — long-only, no leverage, no funding dependence — so spot routing is correct despite the 0.20% round-trip fee; the 1.2% take-profit leaves ~1.0% net per winner, well clear of the 0.25% spot per-trade floor, and SOL's flush snap-backs are routinely that large in absolute terms. Warmup is 194 bars (~2 days of 15m data), a small fraction of any sandbox window, so the entry gate has ample opportunity to fire. Sizing is a fixed capped fraction of equity with a 1.8% stop (≈0.6–0.95% of equity risked per trade), so losses cannot compound into position size.

Catastrophic, decisive failure on multiple independent grounds — not optimizable and not a clean iterate. The account was LIQUIDATED (liquidated=true, 2025-08-28) with total_return -100%, max_drawdown 100% (past the >50% hard-abandon rule, L19), PF 0.67, Sharpe -0.61, expectancy -$126/trade, and negative returns in every single year 2020-2025. avg_trade_return_pct 0.10% is below the 0.25% spot fee floor (L22), and the payoff is inverted (avg_win $455 < avg_loss $891 at 57% win rate; return_skew -5.3, kurtosis 367) — the capitulation-reversion mechanism has negative expectancy across every regime on SOL because the uptrend gate does not prevent buying flushes that keep flushing (falling-knife catching). Compounding this, the position sizing is broken: avg_position_pct 438% on a long-only, leverage-1, cash-capped spot strategy is impossible under correct sizing (a spot cash account should not be liquidatable at all), and the conviction-scaling/get_account_equity() path is compounding notional beyond available cash — the source of the liquidation and the absurd late-history monthlies (2025-10 +620%, 2025-11 -142%) that are artifacts of near-zero equity. There is nothing to optimize: even if the sizing bug were fixed, the underlying edge loses money every year with a sub-fee-floor per-trade return, so no parameter region exists to tune toward. This is the single-name mean-reversion fade family that has repeatedly died. Abandon at BACKTEST_REVIEW rather than spend 2 hours.

Long-only intraday capitulation-bounce mean-reversion on SOLUSDT spot (Binance, 15-minute bars). Every bar it computes a continuous capitulation score: the negated z-score of the trailing 4-bar (1h) return, plus a weighted volume-climax z-score, plus an optional taker-sell-dominance z-score from the Binance taker_ratio feed (0.0 and harmless when that feed is missing or stale). It buys when that score exceeds entry_score while price is above its 192-bar (2-day) SMA — i.e. a sharp volume-climax flush inside an uptrend — and exits on a 1.2% take-profit (the reversion snap-back), a 1.8% stop-loss, or a 24-bar (6h) time stop. Spot CASH venue, long-only, leverage 1, sized at a capped 35% of equity (scaled up to 1.5x on the strongest flushes, hard-capped at 90%).

At backtest review, verify the realized win rate on the full window clears ~63%; if not, the TP/SL geometry (or the entry_score selectivity) must be re-tuned so expectancy is positive net of the 0.20% spot round trip. Consider whether a wider TP or tighter entry raises per-trade capture above the fee floor.

Reward:risk is inverted relative to the realized win rate. take_profit_pct=1.2% is SMALLER than stop_loss_pct=1.8%, so the break-even win rate BEFORE fees is 1.8/(1.2+1.8) = 60%, and ~63% after the 0.20% spot round trip. The sandbox achieved only 52.2% win rate, so expectancy is structurally negative: avg_win $408 vs avg_loss $658, profit_factor 0.68, total_return -11.5%, Sharpe -1.28, avg_trade_return_pct 0.086% (below the 0.25% spot floor). This TP<SL profile is not by itself a bug (mean-reversion legitimately trades many small wins vs rare larger losses), but it REQUIRES the reversion snap-back to convert >60% of flushes, and on the sandbox data it does not. This is an edge/performance matter for the backtest reviewer, not a code defect, but it is the decisive risk for this strategy.

Confirm whether the SOL spot instrument actually receives a taker_ratio feed in the full backtest. If it never does, the taker term and its two parameters (taker_weight, taker_peak_window) are dead weight and the strategy should be evaluated purely as a price/volume capitulation-reversion — a class that must clear the spot fee floor on its own.

The taker-sell-exhaustion term — one of the three mechanism components the hypothesis names — is almost certainly INACTIVE in this backtest. taker_ratio is a Binance FUTURES feed and the instrument is SOLUSDT.BINANCE_SPOT; the sandbox shows funding_events_available=0, indicating supplementary futures feeds are not attached to this spot instrument. With taker_z pinned to 0.0, the effective edge collapses to a volume-climax-weighted dip-buy on OHLCV only — a pure price/volume mean-reversion class. This is explicitly permitted by the hypothesis ('graceful degradation ... never silently switch of edge mechanism'), so it is NOT a critical mismatch — the code does exactly what the hypothesis authorizes. But the reviewer should read the result as an OHLCV volume-climax reversion, not as the taker-confirmed microstructure edge, when judging whether the surviving mechanism carries a real edge.

Genuinely distinct mechanism (capitulation/volume-climax reversion) with graceful taker-feed degradation; fills under-represented spot/short-horizon buckets

Clean signal construction (scale-free z-scores, uptrend regime gate)

Account LIQUIDATED (liquidated=true, 2025-08-28); total_return -100%, max_drawdown 100% — past the >50% hard-abandon line (L19)

Loses money in EVERY year 2020-2025 (-17/-12/-35/-32/-36/-114%); PF 0.67, Sharpe -0.61, expectancy -$126/trade, information_ratio -1.57 (L8)

avg_trade_return_pct 0.10% is below the 0.25% spot fee floor (L22); inverted reward:risk (avg_win $455 < avg_loss $891), return_skew -5.3, kurtosis 367 — small wins, catastrophic losses

Position sizing is BROKEN: avg_position_pct 438% on a long-only leverage-1 cash-capped spot strategy is impossible; conviction-scaling/equity sizing compounds into notional exceeding cash, producing the liquidation and the absurd 2025-10 +620% / 2025-11 -142% monthly artifacts

Single-name mean-reversion fade of flushes: the uptrend gate does not stop falling-knife catching — flushes keep flushing and losers run to the stop

This strategy tried to open a fresh mechanism for the factory — fading forced-seller exhaustion on SOL spot by buying volume-climax capitulation flushes within an uptrend, chosen for SOL's large absolute snap-backs and to fill under-represented spot/short-horizon buckets. The signal construction was clean (scale-free z-scores, an uptrend regime gate, graceful taker-feed degradation), but the backtest was an unambiguous disaster: the account was liquidated with a -100% return and 100% drawdown, losing money in every year from 2020 to 2025 with a sub-fee-floor per-trade edge and an inverted small-win/large-loss payoff. On top of the negative underlying expectancy, position sizing was broken — an impossible 438% average position on a cash-capped spot book — producing the liquidation and absurd late-history monthly artifacts. The analyst abandoned it at backtest review as a catastrophic, non-optimizable member of the single-name mean-reversion-fade family that has repeatedly died, rather than spend hours optimizing an edge that loses in every regime.

A trend-regime gate does not prevent falling-knife catching — flushes keep flushing, so a single-name capitulation-fade with a stop produces small wins and catastrophic losses; and conviction/equity-based sizing that lets notional compound beyond available cash can liquidate even a supposedly un-liquidatable long-only spot account.

It was abandoned at the BACKTEST_REVIEW gate as a decisive, multi-ground failure: liquidation and -100% return breach the >50% hard-abandon rule, the per-trade return sits below the fee floor, and the edge has negative expectancy in every regime because the uptrend gate does not stop the strategy from catching falling knives. There was nothing to optimize even setting the sizing bug aside, so it never reached optimization or analysis.

A long-only, single-instrument intraday mean-reversion strategy on SOLUSDT.BINANCE_SPOT (15m bars) that fades forced-seller exhaustion — buying sharp volume-climax capitulation flushes (with optional taker-sell-exhaustion confirmation) only when price is above its slow SMA uptrend gate, and exiting on a reversion take-profit, stop-loss, or time-stop.

It failed catastrophically: the account was liquidated on 2025-08-28 with total return -100% and max drawdown 100%, profit factor 0.67, Sharpe -0.61, and expectancy -$126/trade across 833 trades, losing money in every year 2020-2025. avg_trade_return_pct was 0.10% (below the 0.25% spot fee floor) with an inverted payoff (avg win $455 vs avg loss $891 at a 57% win rate, return skew -5.3), and position sizing was broken — avg_position_pct 438% is impossible for a long-only, leverage-1, cash-capped spot account.