EthSpotRangeGridVolatilityHarvestLong1H

ETH Spot Range-Grid Volatility Harvest, Long-Only (Single-Instrument ETHUSDT.BINANCE_SPOT, 1h Bars, Buy Fixed Steps Down / Sell Fixed Steps Up Around a Slow Reference, Regime-Gated OFF in Trends, Fee-Aware >0.5% Grid Step, Pure OHLCV, 3-Parameter)

A LONG-ONLY, single-instrument RANGE-GRID strategy on ETHUSDT.BINANCE_SPOT that harvests the volatility/rebalancing premium — profiting from oscillation around a slow reference WITHOUT predicting direction — rather than betting on a trend or fading a single extreme. It is mechanistically distinct from every recent failure: not carry (the pending set is carry-heavy), not a directional momentum/ignition bet (which just liquidated an account at -100%), not a single-extreme fade with inverted reward/risk (LINK/ADA/SOL deaths), not a cointegration pairs bet (SOL-AVAX -82%), not buy-and-hold accumulation (LinkSpot drawdown-accum). It fills the under-represented BINANCE_SPOT venue (12.1% vs >=15%, and binance_spot has the 2nd-best tag survival at 2%) with pure OHLCV. The two design choices that keep it out of the graveyards: (1) a fee-aware grid step > 0.5% so every completed buy-low/sell-high cycle nets above the ~0.25% spot round-trip floor — no sub-fee scalping; (2) a hard REGIME GATE that stops buying in confirmed downtrends and a range-break stop that liquidates inventory if the range is decisively broken — directly addressing the one way grids blow up (bag-holding a trending decline). Only 3 parameters. It generates a large, measurable trade sample (hundreds of completed cycles over multi-year history) in exactly the ranging regimes where directional strategies stall — a genuine diversifier.

Implements the hypothesis exactly: a fee-aware, regime-gated, range-break-stopped long-only spot grid that profits from oscillation without a directional view, filling the under-represented BINANCE_SPOT venue with pure OHLCV. Design choices map to the brief: (1) grid_step_pct default 0.8% keeps every completed cycle above the ~0.20% spot taker round-trip and the 0.25% spot viability floor — no sub-fee scalping; (2) the regime gate (buys off when price is below a falling slow SMA) plus the range-break liquidation (sell all if price breaks 12% below the slow reference) address the sole way grids blow up, bag-holding a trending decline. It is built on the factory's proven multi-order driver pattern (all order management inside calculate_signal, standard hooks neutralised) because a grid scales one netted spot position in and out rather than opening a single position via should_enter; the ladder anchors to the last fill price (geometric step walk) so buys deepen on dips and sells unwind on rallies. Spot CASH constraints are respected — only BUYs open inventory and SELLs are capped to the net long (the base's spot-short guard is never tripped). Sizing deploys deploy_frac (85%) of equity spread across max_units tranches so total exposure is bounded and independent of the tranche count, avoiding the compounding-sizing blowups that killed leveraged strategies (leverage stays 1.0, spot). The slow SMA warmup (~122 bars ≈ 5 days on 1h) is a small fraction of the multi-year history and of any Layer-3 sandbox window, and ETH's hourly range crosses 0.8% steps many times per day, so the grid fires frequently and returns to flat regularly — generating a large, measurable sample of completed cycles (avoiding the low-trade-count measurability failure) precisely in the ranging regimes where directional strategies stall. Only three core parameters keep it clear of the overfit graveyard; the SMA/slope/deploy settings are fixed structural conventions. calculate_signal returns a continuous oscillator (percent deviation from the slow SMA) that varies every bar, so the signal is never a flat constant.

Structural grid failure plus a capacity wall — not optimizable. Over 8 years it is a net loser (total_return -2.45%, profit_factor 0.997, expectancy -$3.79/trade) with a 61.9% max drawdown (past the 50% hard-abandon line). Despite a 92% win rate, avg_loss ($16,166) is ~12x avg_win ($1,322) with a -$43k largest loss: the long-only grid banks tiny steps and periodically liquidates the whole accumulated bag into trends — the exact inverted reward/risk that kills grids, and the regime gate/break stop did not prevent it (the deep drawdowns ARE those liquidations). On top of that, modeled impact_cost_pct is 112.6% (impact exceeds gross PnL) with capacity_usd only $78,870, so any deployable size erases the edge. It is also not the advertised market-neutral harvest — benchmark_correlation 0.84 / beta 0.40 make it whipsawed long-ETH beta. Optimizing grid_step_pct/break_exit_pct/max_units cannot invert the small-win/large-loss payoff geometry inherent to the mechanism, nor fix the impact/capacity ceiling. Abandon rather than spend 2 hours.

Long-only single-instrument range-grid on ETHUSDT.BINANCE_SPOT (1h bars, pure OHLCV) that harvests the volatility/rebalancing premium. A laddered grid buys one equity-fraction tranche each time price falls a full grid step (grid_step_pct, default 0.8% > the fee floor) and sells one tranche each time price rises a full step above the ladder, capturing the step on every buy-low/sell-high cycle. All inventory is a single netted spot LONG scaled in/out via the driver (market orders); it never sells more than it holds (long-only, no shorts). A regime gate suppresses new buys in a confirmed downtrend (price below a slow 96-bar SMA whose slope is falling), and a range-break stop liquidates the entire inventory if price breaks break_exit_pct below the slow reference — directly guarding the grid's only blow-up mode (bag-holding a trend). All execution runs in calculate_signal; the standard entry/exit/size hooks are neutralised. Three core parameters: grid_step_pct, break_exit_pct, max_units.

Tighten break_exit_pct or scale it to ladder depth / add a per-tranche stop so a trending breakdown is cut before the ladder is deep underwater.

The range-break stop (break_exit_pct=0.12) is too wide to contain the blow-up mode it claims to guard. At 12% below the slow SMA the full 6-tranche ladder is deeply underwater, so the one-shot liquidation crystallizes a large loss: avg_loss 13,033 = 16.4× avg_win 795, largest_loss -14,339, PF 0.41. The mitigation is coded but fires too late to keep the tail near the size of the banked steps.

Analyst should evaluate on the full multi-year history including true ranging regimes; if correlation to buy-hold stays ~0.9 it is not the neutral diversifier claimed.

Framed as direction-neutral, but the sandbox shows a long-biased ETH accumulator: benchmark_correlation 0.89, beta 0.45, ~9d20h holds. The -30.7% largely reflects ETH's decline, not oscillation capture; the 'diversifier' claim isn't realized.

Confirm the full multi-year backtest produces a materially larger cycle count; if it stays in the low tens the result can't distinguish edge from a few unlucky trend breaks.

Thin completed-cycle sample: total_trades 31 over 344 days (404 fills net to ~31 round trips) versus the 'hundreds of cycles' claim; the loss is dominated by a few range-break liquidations, so the edge rests on few independent samples.

Large, well-sampled trade population (646 net trades / 6,571 fills over 8 years) and pure OHLCV — the mechanism runs as designed

Fee-aware 0.8% step did keep individual completed cycles above the spot round-trip cost (avg_trade_return_pct 0.97%)

Net loser over 8 years: total_return -2.45%, profit_factor 0.997 (<1), expectancy -$3.79/trade, Sharpe 0.088 (CI straddles zero)

max_drawdown 61.9% (CI to 96.6%) — past the 50% hard-abandon line

Inverted reward/risk: win_rate 92% but avg_loss $16,166 is ~12x avg_win $1,322 (largest_loss -$43k) — banks small, loses huge on range-break liquidations; the regime gate/break stop did not fix bag-holding trends

Capacity wall: impact_cost_pct 112.6% (impact exceeds gross PnL) and capacity_usd only $78,870 — edge only at toy scale

Not market-neutral: benchmark_correlation 0.84 / beta 0.40 — whipsawed long-ETH beta, not a direction-agnostic vol premium

EthSpotRangeGridVolatilityHarvestLong1H proposed harvesting ETH spot oscillation by laddering buys down and sells up around a slow reference, protected by a fee-aware 0.8% step, a downtrend regime gate, and a range-break liquidation. The 8-year backtest confirmed the mechanism ran as designed — 646 net trades, a 92% win rate, and per-cycle returns above the fee floor — but it was a net loser (total_return -2.45%, profit_factor 0.997) with a 61.9% max drawdown, because rare range-break liquidations produced losses ~12x the size of its typical wins. Compounding the failure, modeled market impact (112.6%) exceeded gross PnL at a capacity of just $78,870, and a 0.84 benchmark correlation showed it was really whipsawed long-ETH beta rather than a direction-agnostic vol premium. The analyst abandoned it at the backtest-review gate before optimization, judging the inverted payoff geometry and capacity ceiling structural and not fixable by tuning its three parameters.

A long-only accumulation grid inherently trades many small wins for rare catastrophic losses and carries long-beta rather than being market-neutral, so a regime gate and range-break stop alone cannot make it viable — nor can they overcome a capacity/impact wall where trading costs exceed gross PnL.

The analyst issued an 'abandon' verdict at the pre-optimization backtest-review gate: the win/loss geometry was fatally inverted (avg_loss $16,166 ≈ 12x avg_win $1,322, largest loss -$43k) because the grid banks tiny steps then liquidates its accumulated bag into trends, and the max drawdown of 61.9% blew past the 50% hard-abandon line — problems the regime gate and break stop failed to fix and that parameter optimization cannot invert.

A long-only, single-instrument spot range-grid on ETHUSDT.BINANCE_SPOT (1h) that aimed to harvest the volatility/rebalancing premium by buying fixed steps down and selling fixed steps up around a slow SMA reference, with a fee-aware 0.8% grid step, a downtrend regime gate, and a range-break liquidation stop.

Over ~8 years (2018-07 to 2026-07, 70,196 bars) it generated a large sample of 646 net trades / 6,571 fills with a 92.4% win rate and avg_trade_return_pct of 0.97%, but was a net loser overall: total_return -2.45%, profit_factor 0.997, expectancy -$3.79/trade, Sharpe 0.088 (CI straddling zero), and a 61.9% max drawdown (CI up to 96.6%). Modeled impact_cost_pct was 112.6% with capacity_usd of only $78,870, and benchmark_correlation was 0.84 (beta 0.40).