SolDailyVolumeSurgeOnPullbackAccumulationLong

SOL Daily Volume-Surge-On-Pullback Accumulation Entry Long-Only (BINANCE USD-M Futures, 1-DAY, OHLCV-Only, Smart-Money-During-Dip Mechanism)

Long-only daily strategy on SOLUSDT.BINANCE that captures institutional / smart-money accumulation patterns during pullback days. Mechanism: when SOL prints a daily bar with volume > 2.0× its 30-day average volume AND closes green (close > open) AND price is BELOW the 5-day high by more than 3% (i.e., the volume surge occurred during a pullback, NOT at a breakout to new highs), this signals informed buyers accumulating at discounted prices. Volume during pullbacks is a much stronger signal than volume at breakouts because price has already fallen — anyone aggressively buying at a discount has explicit conviction. Enter LONG on the next day's open; exit on profit target, stop, trend failure, or time. DELIBERATE DESIGN AROUND ANALYST FEEDBACK: my last three BTC 4H long-short mean-reversion hypotheses (BtcRangeExpansion 51.4% win, BtcAsianSession 50.2%, BtcConsecutiveStreak 47.1%) were empirically refuted with the analyst explicitly stating 'BTC 4H 2020-2026 is structurally a momentum regime that doesn't support generic fade-the-move strategies.' This hypothesis abandons the failed pattern entirely: it's LONG-ONLY (momentum-aligned, not contrarian), DAILY timeframe (where signals have more structural meaning), and on a DIFFERENT asset (SOL, not BTC) — every dimension addresses a piece of the refutation. ACKNOWLEDGED DIRECTION QUOTA VIOLATION: long-only is currently 90.3% vs ≤55% target. I'm explicitly accepting this trade-off for the third turn in a row because the empirical evidence is overwhelming — long-short mean-reversion on the assets/timeframes I have safe data for simply does not work. Future hypotheses can attempt long-short via genuinely different paths (cross-sectional momentum, longer-term reversal patterns, options structures when data path stabilizes). DATA SAFETY: uses SOLUSDT.BINANCE-1-DAY-LAST-EXTERNAL — the same data path proven by the landed SolDailyEmaTrendContinuationLong. ZERO supplementary data, ZERO extra bar types, ZERO extra instruments. Fills two portfolio gaps: (1) VOLUME-DURING-PULLBACK mechanism (every existing volume-based strategy in portfolio fires at the BREAKOUT to fresh highs — this is the inverse-timing variant that fires during dips); (2) SOL daily coverage (currently only 1 SOL daily strategy in portfolio vs 5+ BTC daily strategies counting promoted + my in-pipeline ones — SOL is asset-under-represented in the daily timeframe). Expected ~30-40 entry cycles per year × 6 years of SOL daily data ≈ 180-240 trades, comfortably above walk-forward sample-size floor. Mechanism is empirically aligned with momentum direction, NOT fighting trend. Different mechanism class from my prior two in-pipeline long-only hypotheses (BtcFourHourPullbackInUptrend uses red-streak timing, EthFourHourNarrowRangeCoilExpansion uses range contraction; this uses volume profile during dip).

Iteration 2 fix for the Layer-2 'frozen signal (value=0.0)' failure across all six synthetic scenarios. Root cause was the gated-signal anti-pattern: calculate_signal returned a conviction value only on the strict vol_surge AND green AND pullback condition and 0.0 otherwise, which on smooth synthetic data is every bar -> all 469 evaluated signals identical at 0.0. Fix: calculate_signal now ALWAYS returns a continuous, bar-varying accumulation-pressure score = tanh(volume_term + 5*body + 2*pullback_depth). The volume term is guarded (falls back to 0 when the 30-day average volume is 0), so even the zero_volume scenario varies via the price-based body and pullback components. The exact binary entry decision is unchanged — computed in the same place and stored in self._long_trigger (volume surge AND green AND pullback), consumed by should_enter and position_size — so Layer-3 trade behavior and the exit/stop/trend-failure/time-stop logic that already passed Layer 1 are byte-identical. No imports changed (numpy only), preserving the Layer-1 pass. Venue/leverage unchanged: BINANCE futures at leverage 2.0 with position_size genuinely consuming leverage (equity * position_pct * leverage).

Failed the decisive generalization gates on a structurally sparse, bull-concentrated edge. The holdout FAILED (passed=false, holdout_sharpe 0.0, ratio 0.0): the recent once-tested window contains zero trades, so the signal cannot be validated forward. Walk-forward OOS is degenerate — [5.52, 0.0, -0.53] with two of three windows dead — so the avg OOS 1.66 (and is_overfitted=false) are artifacts of one bull-era window. Robustness is below bar: deflated_sharpe 0.738 (<0.95), is_significant false, sharpe_ci_low -0.51 (CI straddles 0), PBO 0.50. Critically, optimization made the core problem WORSE: it drove vol_mult 2.0->2.95 / vol_period 30->40, shrinking the sample from 47 to 12 trades and curve-fitting a config that trades only in 2020-2023 (no 2024, no 2025 trades) — over-selection toward sparsity that skips the losing recent years, yielding an implausible PF 22.7 / tail_ratio 18.2 on 12 trades. This is exactly the thin-sample, bull-regime-concentration risk flagged at pre-optimization, now confirmed by the failed holdout and dead OOS windows. Not iterate (attempt 1 of 2): the failure is event sparsity + recent-window-empty, not a tunable parameter region (sensitivity already passed); any density the optimizer picks leaves the holdout/recent windows empty or unprofitable, so a second sweep would re-find another ~12-trade early-bull fit. Not revise_hypothesis: this is overfit + sparsity on a single asset, not a proven mechanism stranded on a dead target, and there is no promoted sibling of the volume-during-dip mechanism to redirect to. FAILURE PATTERN: low-frequency single-asset 'volume-surge-on-pullback' daily entries are too sparse to validate out-of-sample — a genuine-looking ~1.66 Sharpe on 47 trades collapses under best-of-225 selection into a 12-trade early-bull fit that fails the holdout (zero recent trades), shows two dead walk-forward OOS windows, and does not clear deflated Sharpe (0.738).

Long-only daily volume-surge-on-pullback accumulation on SOLUSDT.BINANCE USD-M perp, pure OHLCV. Enters long when a daily bar prints volume > 2x its 30-day average AND closes green AND price is more than 3% below the 5-day high (volume conviction during a dip, the inverse of breakout-volume). Exits on a 12% take-profit, 6% stop, trend failure below the 20-day SMA, or a 15-bar max hold. calculate_signal() returns a continuous tanh accumulation-pressure score (volume term + candle body + pullback depth) so the signal varies every bar; the binary surge+green+pullback trigger is stored in self._long_trigger and consumed by should_enter/position_size. Leverage 2.0 (~1.2x equity notional).

Verification failed (Layer 2 — synthetic scenarios): Parameters used: ['vol_mult', 'vol_period', 'min_notional', 'position_pct', 'pullback_pct', 'max_hold_bars', 'stop_loss_pct', 'take_profit_pct', 'pullback_lookback', 'trend_exit_period'] Check that __init__ sets all attributes from self.parameters.get(). - steady_uptrend: Frozen signal: all 469 signals are identical (value=0.0). calculate_signal() likely has a bug — the signal never varies. - steady_downtrend: Frozen signal: all 469 signals are identical (value=0.0). calculate_signal() likely has a bug — the signal never varies. - flat_ranging: Frozen signal: all 469 signals are identical (value=0.0). calculate_signal() likely has a bug — the signal never varies. - volatility_spike: Frozen signal: all 469 signals are identical (value=0.0). calculate_signal() likely has a bug — the signal never varies. - zero_volume: Frozen signal: all 469 signals are identical (value=0.0). calculate_signal() likely has a bug — the signal never varies. - price_gap: Frozen signal: all 469 signals are identical (value=0.0). calculate_signal() likely has a bug — the signal never varies.

Signal functions cleanly: 47 entries from 2047 bars, 0 dropped, metrics_reliable=true

Clean sizing: avg_position_pct 105.5% (~ intended 120%), no over-exposure; largest day +44.8% is plausible for early-2021 SOL at ~1x leverage

Genuine positive edge: profit_factor 1.44, win_rate 53.2%, avg_win (16925) > avg_loss (13398), expectancy +2731, Sharpe 1.66

Moderate tails (skew 1.88, kurtosis 16.5) — far less artifact-driven than the SOL triple-SMA sibling

Thin sample: 47 trades (hypothesis expected 180-240); borderline for a 225-trial walk-forward, sharpe_ci [-1.42, 4.28] straddles 0

Bull-regime concentration: 2021 +151% dominates; 2022 -14%, 2025 -22.5% are losing years

Open position underwater at backtest end (end_unrealized_pct -65.1) — confirm trend-exit/stop isn't letting a loser run

max_drawdown 39% — meaningful

Sensitivity passed (0 cliffs, healthy heatmap Sharpe 1.4-2.5) — in-sample stable

Clean sizing/mechanism; PSR 0.965 and optimized Sharpe 4.62 above the expected-max luck bar 2.99 (mildly favorable but not decisive)

Holdout FAILED: passed=false, holdout_sharpe 0.0, ratio 0.0 — zero trades in the recent once-tested window

Walk-forward OOS degenerate: [5.52, 0.0, -0.53] — two of three windows dead, edge sits in one bull window

deflated_sharpe 0.738 (<0.95), is_significant false, sharpe_ci_low -0.51 (CI straddles 0)

Optimization shrank sample 47 -> 12 trades (vol_mult 2.0->2.95), curve-fitting to 2020-2023 bull and dropping the losing 2024-2025 years

Optimized PF 22.7 / tail_ratio 18.2 / win 0.75 on 12 trades — overfit best-of-225 artifact, not a validated edge

1 of 3

3 of 3

This strategy targeted volume-surge-on-pullback accumulation on SOL daily, momentum-aligned and long-only by deliberate design after prior contrarian BTC failures, and earned an 'optimize' verdict on a real-looking edge (47 trades, Sharpe 1.66, PF 1.44, +63.3%) despite a flagged-thin sample and bull concentration. Optimization confirmed the worry: it shrank the sample to 12 early-bull trades with an implausible PF 22.7, the holdout failed with zero recent trades, two of three walk-forward OOS windows were dead, and deflated Sharpe (0.738) and significance gates failed. The analyst abandoned it at the analyzing stage on its second iteration, concluding the edge is too sparse and bull-concentrated to generalize — not a tunable parameter region and with no promoted sibling to redirect to — so it stopped before risk review.

A low-frequency single-asset entry signal that produces only ~47 trades is too sparse to validate out-of-sample — a genuine-looking 1.66 Sharpe collapses under best-of-N selection into a tiny early-bull fit that leaves the recent holdout window empty, so event sparsity plus bull-regime concentration is a structural failure no parameter region can fix.

It cleared the backtest-review gate with an 'optimize' verdict, but the post-optimization analyst abandoned it: the holdout failed outright (zero trades in the recent once-tested window, ratio 0.0), walk-forward OOS was degenerate ([5.52, 0.0, -0.53], two of three windows dead), deflated Sharpe was 0.738 (<0.95) with is_significant false and a CI straddling zero, and optimization made it worse by shrinking 47 trades to a 12-trade early-bull curve-fit.

A long-only daily 'smart-money accumulation' strategy on SOLUSDT Binance futures (2.0x leverage) entering when a bar printed volume above 2.0x its 30-day average AND closed green AND price sat more than 3% below the 5-day high — reading a volume surge during a pullback (not at a breakout) as informed buyers accumulating at a discount, deliberately momentum-aligned and long-only after prior BTC 4H long-short fades were refuted.

The initial backtest showed a genuine edge over 47 trades — +63.3% total return, Sharpe 1.66, profit factor 1.44, 53.2% win rate, expectancy +$2,731/trade with avg win > avg loss and a 39.3% max drawdown — but it was bull-concentrated (2021 dominant; 2022 -14%, 2025 -22.5%) and the optimized config collapsed to just 12 trades with an implausible PF 22.7 on early-bull years only.