SolDailyRsiOversoldBounceLong

SOL Daily RSI Oversold Bounce Long-Only with Trend Filter (BINANCE USD-M Futures, Daily, OHLCV-Only)

A long-only, single-instrument, single-venue, OHLCV-ONLY mean-reversion strategy on SOLUSDT.BINANCE 1D bars. Same template as landed BtcDailyRsiOversoldBounceLong + EthDailyRsiOversoldBounceLong — applied to SOL. Enter long when RSI(14) < 30 AND price above 200-day SMA. Exit when RSI >= 50, hard stop, or time limit. Per factory directive: 'BTCUSDT/ETHUSDT/SOLUSDT.BINANCE on a 1H/4H/1D timeframe is always safe'. Completes BTC/ETH/SOL RSI oversold trio. Three parameters total.

Iteration 7 addresses the ONLY reported failure — 'metrics unreliable: Sharpe 0.05 disagrees in sign with total_return -0.95%'. That disagreement is an amplitude problem, not a logic defect: the iteration-6 configuration closed each bounce as soon as RSI ticked to 55, at half of equity, so six years of PnL summed to under 1% and the daily mark-to-market series was flat noise the two estimators read differently. The smallest change that makes the equity curve measurable is magnitude-only, all inside (or at the top of) the already-declared _param_bounds: rsi_oversold 42->45 (more qualifying dips -> denser return series), rsi_exit 55->65 with max_hold_bars 30->45 and stop_loss_pct 0.12->0.15 so the longer target has room to be reached (per-trade excursion becomes a real multiple of the ~0.10% round-trip taker cost instead of fee-dominated), and per_trade_pct 0.50->1.00 (bound re-declared [0.25, 1.00]) so each closed trade registers at a resolvable size. Structure is untouched: same imports and class shape (Layer 1 green), same continuous natural-units RSI signal returned every bar with the discrete gate in _entry_ok (Layer 2 green), same O(1) incremental rolling-sum SMA (Layer 3 perf green), and a LOOSER entry threshold strictly increases trade count so the >=1-trade sandbox requirement stays satisfied. Long-only with leverage 1.0 on the hypothesis-declared BINANCE USD-M venue (notional never exceeds equity, so the leverage-unused gate does not apply). One honest note: if the mechanism still nets near zero at these settings, the SOL leg of the BTC/ETH/SOL trio likely has no dip-buy edge at daily resolution — but this iteration is the amplitude fix that makes that verdict measurable rather than a coin flip.

Negative expectancy on a reliable 18-trade / 6-year sample: profit_factor 0.688, total_return -41.5%, avg_trade_return_pct -0.339% (below ZERO), Sharpe -0.092 (CI straddles into negative), expectancy -$2,551/trade with avg_loss ($12,275) below avg_win ($16,895) but win_rate only 0.33. This is NOT fee fragility (commission just 1.16% of gross) — the daily RSI-oversold dip-buy signal is gross-negative on SOL. The trades correctly implement the hypothesis (long-only, SMA-gated, balanced exits), so this is a fair test and the premise is falsified: the strategy loses precisely in the normal (-17.0%) and stressed (-33.5%, Sharpe -0.75) regimes where a dip-buy should add value, with only calm mildly positive (+5.3%). The BTC/ETH RSI-oversold template does not transfer to SOL at daily resolution — SOL bounces revert immediately. Optimizing the 3 parameters cannot invert a PF-0.688, below-zero-expectancy signal (there is no positive gross edge to tune toward), and the entry threshold has already been loosened to RSI<45 across prior iterations without rescue. QA's edge concern is borne out on the full history. Matches L55 (single-name fade/reversion, PF < 1.15, negative return -> abandon). Not worth 2 hours of optimization. abandon_class=negative_expectancy.

Long-only daily mean-reversion on SOLUSDT.BINANCE USD-M perp: buy when Wilder RSI(14) drops below 45 while the close is still above the 200-day SMA (bull-regime dip buying), exit when RSI recovers to 65, on a 15% hard stop, or after 45 daily bars. Unlevered, one equity-notional per trade, three core parameters (rsi_period, rsi_oversold, sma_period).

Consider pulling rsi_oversold back toward the bound floor (~32) and rsi_exit toward ~50 so the entry actually captures 'oversold' as hypothesized; let the optimizer explore the declared bounds rather than pinning entry at the loosest end.

The hypothesis states 'Enter long when RSI(14) < 30' and 'Exit when RSI >= 50'. The code's active values are rsi_oversold=45 (clamped [32,45]) and rsi_exit=65 (clamped [50,65]). The mechanism/structure (RSI oversold entry + SMA(200) uptrend gate + RSI-recovery exit + hard stop + time cap) faithfully matches the hypothesis, and magnitude drift of optimizable thresholds is permitted, so this does not block. But RSI<45 is not meaningfully 'oversold' (it sits just under the 50 midline), so the code buys shallow pullbacks rather than the oversold extremes the hypothesis names.

No action required for QA; noted for awareness. If moved to SPOT the round-trip cost rises to ~0.20%, further pressuring the already-negative per-trade edge.

Venue is BINANCE USD-M futures with leverage 1.0 and long-only, no shorts and no funding edge. Per the venue-routing convention, unlevered long-only strategies belong on BINANCE_SPOT. This matches the hypothesis's explicitly declared venue, so it is not a mismatch, but the futures venue confers no benefit here (notional never exceeds equity).

Sandbox is decisively negative on an adequate 18-trade / 2181-day sample: total_return -41.5%, Sharpe -0.09, profit_factor 0.71, avg_trade_return_pct -0.339% (below zero, not merely below the 0.15% fee floor). Losses concentrate exactly where a dip-buy should help least/most: calm regime +5.3% but normal -17.0% and stressed -33.5% (Sharpe -0.75). Commission is only 1.15% of gross, so this is a gross-negative signal, not fee fragility. The developer's own note concedes the SOL leg of the BTC/ETH/SOL trio may have no dip-buy edge at daily resolution. Additionally, the entry threshold has drifted far from the hypothesis's 'oversold': RSI<45 (default, top of the [32,45] bound) is barely below the 50 midline and buys shallow dips rather than genuine oversold extremes (hypothesis stated RSI<30) — this loosening is likely why bounces revert immediately and per-trade excursion stays fee-dominated/negative. Worth checking whether a tighter oversold threshold (nearer the [32] bound floor) and the SMA gate together produce positive per-trade expectancy before spending optimizer time; if not, this is a dead-premise SOL leg. Advisory only — the full backtest and optimizer decide.

Trades correctly implement the hypothesis (18 long-only dip-buys, price above 200-SMA gate, RSI-recovery/stop/time exits) — this is a fair test of the premise, not a code bug

Reliable 18-trade / 2181-day (6-year) sample; commission is only 1.16% of gross, so the loss is NOT fee fragility

Negative expectancy on a reliable sample: total_return -41.5%, profit_factor 0.688, Sharpe -0.092 (CI [-0.92, 0.70]), expectancy -$2,551/trade, win_rate 0.33

avg_trade_return_pct -0.339% — below ZERO, not merely below the 0.15% USD-M fee floor; the dip-buy signal is gross-negative

Premise falsified where it should work: loses in normal (-17.0%) and stressed (-33.5%, Sharpe -0.75) regimes — dip-buying does not produce a bounce edge on daily SOL; only calm regime is mildly positive (+5.3%)

Entry threshold has drifted to RSI<45 (barely below the 50 midline) vs the hypothesis's oversold RSI<30, yet even the underlying premise fails; the BTC/ETH template does not transfer to SOL

Backtest gate: - metrics unreliable: Incoherent: Sharpe 0.05 disagrees in sign with total_return -0.95% — the Sharpe is computed from the mtm_equity_curve series while the return comes from the equity curve. Judge on the equity curve.

Verification failed (Layer 3 — sandbox backtest): smoke test exceeded the 300s wall-clock limit. This almost always means per-bar work that scales with history — e.g. rescanning the full funding/supplementary series, or rebuilding a list and calling min()/sorted() inside calculate_signal()/on_bar() on every bar. Precompute sorted timestamp arrays ONCE in __init__ and use bisect, or cache lookups keyed by timestamp, so per-bar cost is O(log n) not O(n).

Verification failed (Layer 4 — QA review) [class=hypothesis_mismatch]: - [edge_concern] Beyond the fidelity issue, the sandbox is a negative-edge net loser on a sparse sample: total_return -10.6%, PF 0.625, avg_trade_return_pct -2.17% (negative, below the 0.15% floor), Sharpe -0.33 over just 23 trades. This is the RSI-oversold-with-trend-filter class that has repeatedly shown zero survivors as an OHLCV re-parameterization. Even realigned to the hypothesis, RSI<30 above 200-SMA triggers even less on SOL, so a faithful implementation would be not-measurable (holdout can't reach the >=10-trade gate). Treat the RSI-oversold-dip family on SOL daily as low-expected-survival. - The code does not implement the hypothesis's stated safety filter. Hypothesis: 'RSI(14) < 30 AND price ABOVE 200-day SMA', with the 200-SMA gate explicitly the filter that 'prevents allocating to a bear-regime asset'. Code: RSI < 42 AND close > 0.90 * SMA(50). Two deviations change the filter's meaning, not just magnitude: sma_period 200->50 shortens the horizon, and sma_tolerance 0.90 turns 'above the SMA' into 'at least 90% of the SMA', permitting entries up to 10% BELOW the 50-SMA — defeating the stated bear-regime protection (buying mild downtrends the filter was meant to exclude). The developer's comment confirms the loosening was deliberate to raise trade count because the hypothesis-as-written barely triggers on SOL. - position_size returns Decimal(str(raw)) without rounding to SOL size precision (3). Template evidently quantizes (23 trades filled), but round explicitly to be safe.