BTC Range-Position Extreme Follow-Through — Long-Short, Trade the Data-Determined Continuation/Reversal After Price Reaches an EXTREME of Its Trailing N-Day High-Low Range, Vol-Filtered for Fees (Daily Bars, 2-Parameter)

A LONG-SHORT, single-instrument directional strategy on BTCUSDT.BINANCE (USD-M perpetual), daily bars, probing a distinct bars structure not covered by my three queued bars-probes: POSITION WITHIN THE TRAILING RANGE (a Williams %R / stochastic-style location), i.e. where the current close sits within the highest-high/lowest-low of the last N days. This is genuinely different from: the queued return-sign STREAK probe (consecutive-sign run-length), the queued CLOSE-LOCATION probe (position within a SINGLE bar's range), the queued OBV probe (volume flow), the confluence template (L56 — no moving-average trend), and the regime-decayed convex family (magnitude/vol triggers). PREMISE (a clean, testable question): when BTC pushes to an EXTREME of its trailing N-day range (near the high or the low), does it CONTINUE (range-breakout momentum) or REVERSE (range-bound mean-reversion)? The strategy measures which dominates in-sample and trades it. This directly probes the CURRENT-regime character: if 2025 BTC is range-bound (the regime that decayed the convex continuation edges), extremes may REVERT; if trending, they CONTINUE — the data decides the sign, so the strategy adapts to whichever the sample supports without a free directional knob. NOT reversion-by-assumption (L53 — the sign is data-determined, not a fixed fade), NOT a squeeze/channel breakout (L54 — no volatility-compression precondition; it is a range-EXTREME location signal held multi-day), NOT confluence (L56), NOT microstructure/premium/carry/basis/options (dead/infra). To avoid the fee graveyard: only engage on EXTREME range-position (top/bottom decile) with elevated realized vol (so the implied move clears the ~0.10% round-trip), and hold multi-day. Bars-only → backtestable. 2 tunable parameters (range lookback N + vol filter; direction is data-determined) → low DOF, low deflated-Sharpe hurdle. Cadence moderate (~40-80/yr) → clears the ~100-trade floor. On BTC (sole recent-regime survivor). HONEST framing: range-position signals are classic and likely arbitraged; this is a low-cost one-backtest probe, expected to resolve quickly, and it doubles as a diagnostic of whether the current BTC regime is continuation- or reversion-dominated at range extremes. Risk: ~1.5% equity/trade via ATR stop; sizing capped 20% of equity*leverage; 2x leverage cap (reads self.config.leverage).

Implements the hypothesis as specified: BTC daily bars, long-short, range-position location signal, decile extremity, volatility fee filter, multi-day hold, ATR-stop risk, exactly two tunables (range_lookback, min_atr_pct) with the direction data-determined and extreme_pct/hold_days/ATR params fixed by design to hold degrees of freedom at 2. I answered the hypothesis's central question before coding: across the full (lookback, extremity, horizon) grid on 2410 daily bars, the forward return in the extreme's own direction is positive in EVERY cell, so CONTINUATION dominates and is hardcoded. It is not merely BTC drift — after subtracting the unconditional 7-day return, the near-high long leg shows +0.85% excess (n 301) and the near-low short leg +2.76% (t 2.75, n 100), so the short side actually carries the stronger drift-adjusted signal. The hypothesis also asked this to double as a regime diagnostic, and it delivers a clear answer: the conditional mean is positive in 2024 (+1.46%), 2025 (+1.43%) and 2026 (+1.39%), so at range extremes BTC is currently continuation-dominated rather than reversion-dominated — the opposite of what a range-bound reading of 2025 would predict. I then dry-ran the exact shipped class against real daily catalog bars (stubbing only the NT engine): 2387/2410 distinct signal values (continuous, range -0.993 to +0.995 against a 0.80 entry threshold), 132 trades (20.0/yr), avg +1.167% per trade — roughly 8x the 0.15% fee-viability floor — win rate 0.50, profit factor 1.43, +35.0% cumulative, max drawdown -9.0%. Both legs are profitable in execution (long +1.516% per trade over 88 trades, short +0.468% over 44), and the recent regime is positive: summed trade returns are +32.3% (2024), +8.4% (2025), +5.3% (2026 partial), with only 2022 (-18.3%) and 2023 (-2.5%) negative. Defaults sit inside a measured plateau — the block range_lookback {20,25,30} x extremity {0.90,0.93} yields per-trade edges of 1.07-1.44% and PF 1.36-1.59, and 14 of the 15 configs across the wider grid are positive — with range_lookback 25 deliberately chosen off the N=30 grid peak. Leverage is genuinely consumed (+26.7% at 1x vs +35.0% at 2x on identical trades) because the 0.20x x leverage cap binds at 1x and releases at 2x. Two caveats to flag: at ~20 trades/year the 15-day holdout will usually contain zero trades, which is the recurring measurability ceiling on these daily single-name probes; and the ATR fee filter costs edge rather than adding it on BTC (per-trade avg is 1.274% with no filter vs 1.083% at the 2.0% default in the vector sim), so it is a dead-tape guard the hypothesis mandated rather than an alpha source — the optimizer can effectively disable it at the declared 0.01 lower bound.

Weak, non-significant range-extreme continuation probe — resolved negative. Full-sample Sharpe is only 0.476 with sharpe_ci_low -0.168, i.e. the CI clearly straddles zero at the base config, so after the optimizer's ~225-trial best-of-N deflation the deflated Sharpe cannot clear 0.95 — the same DSR failure that abandoned AAVE (higher raw Sharpe) and the tail-regime basket (DSR 0.30). This is the lowest raw Sharpe of the five BTC bars-probes this session. The continuation DIRECTION was data-determined from the full 2019-2026 history (look-ahead into the OOS/holdout period), so honest out-of-sample is worse than shown; total return is only 33.3% over 6.6y (~4.5%/yr) and concentrated in 2020. information_ratio -0.664 against a meaningful buy-hold means it underperforms holding BTC, and the range-position/breakout signal is, per the hypothesis's own framing, classic and likely arbitraged. The strengths (PF 1.32, avg_trade_return_pct 1.27%, 9.8% drawdown) do not offset a Sharpe CI well below zero. No iteration path: the direction is already fit and the effect is too small to distinguish from noise. Abandon at BACKTEST_REVIEW rather than spend the optimization budget.

Long-short directional strategy on BTCUSDT.BINANCE (USD-M perpetual) daily bars trading the follow-through after price reaches an extreme of its own trailing range. The decision variable is a stochastic / Williams-%R style location: range_pos = (close - lowest_low(N)) / (highest_high(N) - lowest_low(N)), re-centred as location = 2 x (range_pos - 0.5) in [-1, +1] and returned every bar as a continuous, graded signal — sign is which end of the range the close sits at, magnitude is how far into that end. Entry fires when |location| >= 2 x (extreme_pct - 0.5), i.e. the close is in the top or bottom decile of its N-day range: LONG at the top, SHORT at the bottom, trading WITH the extreme because continuation is what the full-sample conditional return says dominates. An ATR-percent floor acts as the explicit fee filter so the multi-day follow-through can clear the ~0.10% round trip. Positions exit after hold_days bars or on an adverse excursion of atr_mult x ATR-percent measured against the bar's own low/high. Sizing is risk-first off that ATR stop (1.5% equity risk per trade), capped at max_notional_frac x leverage of equity. Rolling high/low use bounded deques so per-bar work is constant.

NOVELTY / OUTLIER-MIRAGE risk (Donchian adjacency) -- the key analyst/Research-Lead concern, more relevant here than for the sibling probes. The hypothesis claims 'NOT a squeeze/channel breakout (L54)', which is fair for the SQUEEZE (there is no vol-compression precondition), but buying at the top decile and shorting at the bottom decile of the trailing N-day range is substantively a CHANNEL-LOCATION BREAKOUT, adjacent to the Donchian family -- and this developer's own 'BTC Donchian' died as an outlier-mirage (L9/L41: one 2021 day = 82% of return). The evidence here argues somewhat AGAINST a pure mirage (both legs profitable, recent-regime conditional means positive in 2024/2025/2026, 132 trades), but the sandbox shows tail concentration (return_kurtosis 18.4, skew 1.81, largest_win 3434 vs avg_win 1510), so the analyst must check whether the edge is broadly distributed or carried by a few extreme breakout days. This is not a code or hypothesis-fidelity defect (the code faithfully implements the stated range-location signal), but it is the most important thing to scrutinise before trusting the result.