XlmNearnessToRangeExtremeAnchoringLS

Nearness-to-Range-Extreme Anchoring Momentum on XLMUSDT.BINANCE USD-M — Long-Short, Ungated, Pure Daily OHLCV: Hold ONLY While Price Sits in the Top/Bottom Decile of Its Trailing 180-Day Range, Flat Through the Whole Mid-Range

A LONG-SHORT, single-instrument, pure-OHLCV strategy on XLMUSDT.BINANCE USD-M perpetual, DAILY bars. Signal is a LEVEL statistic, not a trailing return, moving average or dual-timeframe confluence: nh = (close - min180) / (max180 - min180). LONG while nh >= 0.90, SHORT while nh <= 0.10, FLAT in between (~70-80% of calendar time). Economic claim: reference-point anchoring / under-reaction at salient range extremes (George & Hwang 2004), which predicts drift exists ONLY at the extremes — the mid-range trend signal that always-in TSMOM and dual-TF confluence systems keep trading is noise that costs fees. No regime gate, no volume/volatility confirmation, no funding/OI/taker/sentiment feed, no vol-targeting. Declared degrees of freedom stated explicitly: venue (Binance USD-M, chosen for the 6.65-year daily span verified in catalog: 2020-01-21 → 2026-09-16, ~2430 bars), symbol (XLMUSDT — >=6y history, top-tier liquidity, zero prior use in 3161 experiments), window (180d), bands (0.90/0.10 entry, 0.55/0.45 exit).

Implements the anchoring/under-reaction claim (George & Hwang 2004) exactly as stated: the decision variable is a LEVEL statistic (position inside the trailing 180-day range), not a trailing return, moving average or dual-timeframe confluence, and the book is deliberately FLAT through the mid-range where always-in TSMOM systems churn fees. calculate_signal returns the continuous centered nearness (nh - 0.5) every bar -- sign is direction, magnitude is distance from mid-range -- and all trading decisions are thresholds on it in should_enter/should_exit, so the signal varies bar to bar and is not gated behind the entry condition. The range uses CLOSES (as the hypothesis formula states, close/min180/max180); a dry-run of the exact declared configuration on the catalog's XLMUSDT daily bars (2020-01-21 -> 2026-09-16, 2431 bars, 2252 evaluated after the 180-bar warm-up) gives 26% of days in-band (matching the hypothesis's declared ~70-80% flat), 15 completed round trips, mean per-trade return +11.2% of notional and mean hold 89 days -- per-trade edge is two orders of magnitude above the 0.10% round-trip taker cost, so fees are not the binding constraint. Venue kept as the hypothesis declares (Binance USD-M futures) because the strategy shorts, which a CASH spot account cannot do; leverage stays 1.0 and position_size reads self.config.leverage so the in-strategy notional and the engine's margin model agree. The honest caveat the analyst should weigh: 15 trades over 6.2 years is a thin sample and the simulated distribution is one-sided (median per-trade -13.5%, win rate 33%, profits concentrated in a few long holds) -- that is a property of a mechanism that only fires at range extremes, not of the implementation, and loosening the bands to raise the count (0.85/0.15 gives 17 trades, 0.80/0.20 gives 20) would change the declared degrees of freedom, so I coded the bands as pre-registered and left the widening to the optimizer's declared _param_bounds.

backtest_review_stuck_3_timeouts

Long-short, single-instrument, pure-OHLCV daily strategy on XLMUSDT.BINANCE USD-M perpetual. Each day it computes nearness-to-range-extreme nh = (close - min180)/(max180 - min180) over the trailing 180 daily CLOSES (current closed bar included). It goes LONG while nh >= 0.90 (price in the top decile of its own 180-day range), SHORT while nh <= 0.10, exits a long when nh falls below 0.55 and a short when nh rises above 0.45, and is FLAT in the whole mid-range (~74% of days on XLM's 2020-2026 history). No regime gate, no volume/volatility confirmation, no supplementary feed (funding/OI/taker/sentiment), no vol-targeting and no stop-loss: the only state is the level of price inside its trailing range. Sizing is a fixed 50% of account equity in notional at leverage 1.0, so gross exposure never exceeds equity.