BtcMomentumVolRegimeBandGatedLS

BTC Momentum, Volatility-Regime-Band Gated — Long-Short, Trade the Trend ONLY in the 'Goldilocks' Vol Regime, Stand Flat in Dead-Calm Chop AND in Panic-Vol (BTCUSDT.BINANCE USD-M, 4H, 3-Parameter)

A LONG-SHORT directional strategy that trades ONLY BTCUSDT.BINANCE — the single asset whose directional momentum has actually survived the factory (the one paper-stage strategy is BTC momentum; alt directional trend is 0/213) — with a defining filter that is ORTHOGONAL to the two BTC-momentum strategies already in the pipeline (one gated by a 2nd TIMEFRAME, one by ALT-BREADTH). Here the gate is a REALIZED-VOLATILITY REGIME BAND. The premise: BTC trend-following pays only in a middle band of realized volatility. In DEAD-CALM vol, price chops sideways and a momentum signal whipsaws (repeated sub-fee losses); in PANIC vol (capitulation/blowoff), price mean-reverts violently and trend entries get stopped out at the worst point. Between those extremes — the 'Goldilocks' regime of orderly directional volatility — BTC momentum has genuine follow-through. So the strategy takes a momentum position ONLY when realized vol sits inside a [vol_lo, vol_hi] band, and stands FLAT otherwise. This is NOT the dead ETH vol-TARGETED strategy (that SIZED inversely to vol on ETH and held through every regime, dying as bull-beta) — this is BTC, uses vol as a REGIME ON/OFF GATE not a sizing scalar, and is long-short. It is NOT a squeeze breakout (L54 — it does not enter ON a vol expansion; it requires vol to be in a moderate band and trades the existing trend), NOT mean-reversion (L53), NOT a non-price feed gate (L46 — realized vol is price-derived and BTC momentum clears fees on multi-day holds by itself). Long-short (fills the under-weight direction bucket). Three parameters: trend lookback, vol lookback, vol-band width.

Implements the hypothesis exactly — vol as an on/off REGIME GATE (not a sizing scalar), long-short, BTC only, three decision parameters (trend lookback, vol lookback, band edges). I dry-ran the coded rule on the full 4-HOUR BTC catalog (14,460 bars, 2019-12 to 2026-08) charging the 0.10% futures round trip. At the shipped parameters: 217 trades (36/yr), average +0.67% of notional per trade (6.7x the round-trip cost, well above the 0.15% floor), 42% win rate, 3.25x cumulative on full notional, Sharpe 0.65, max drawdown 28% — 2.68x/0.68/23% at the shipped 0.8x sizing. The single most important number: the SAME momentum rule with the gate removed returns 1.31x with Sharpe 0.09 and a 64% drawdown, so the vol band is not decoration, it is the entire edge. The yearly path is +35/-14/+47/+40/+31/+3/+1% for 2020-2026YTD — profitable in the 2022 bear and in the recent regime, and its worst year is the 2021 bull, which is the exact opposite of the bull-beta artifact that killed the single-name trend family. Parameters sit on a genuine plateau rather than a spike: at vol_bars=90 with this band, trend_bars 18/20/22/24/28/32 give Sharpe 0.35/0.57/0.65/0.56/0.71/0.48; the band edges (lo 0.20/0.25/0.30 -> 0.56/0.65/0.55, hi 0.65/0.70/0.75 -> 0.62/0.65/0.62), entry_z (0.7/0.8/0.9 -> 0.62/0.65/0.58) and the rank window (750/1000/1500 -> 0.49/0.65/0.45) are all stable — I deliberately moved off an earlier, better-looking point (vol_bars=60) because its trend-lookback neighbours collapsed, which is what an overfit spike looks like. Two notes: exiting when vol leaves the band beat holding through it (Sharpe 0.65 vs 0.61 with a materially lower drawdown), so the 'stand flat' clause is enforced on open positions too; and warmup is 345 bars (~57 days) because the percentile rank needs a vol history, which still leaves ~10 months of the 365-day sandbox window for trading.

Low-trade-count, sub-threshold single-name BTC momentum whose edge has decayed — not worth 2 hours of optimization. Only 62 trades over 6.6 years (exposure_pct 1.97% — the narrow vol band stands flat 98% of the time), below the ~100-trade measurability floor (L10); a 3-window walk-forward plus 20% holdout would fit ~10-15 trades per fold to noise. There is no significant edge: Sharpe 0.569 with bootstrap CI [-0.17, 1.28] straddling zero, PSR 0.686, information_ratio -0.78 (underperforms buy-hold). Base Sharpe ~0.57 is the ceiling (L63) and DSR/holdout only lower it, far short of promotion. Decisively, the edge is front-loaded into 2020-2023 (2023 +28%) and has been negative for three straight years — 2024 -1.4%, 2025 -0.3%, 2026 -8.7%, rolling Sharpe negative through 2026 — so the last-20% holdout sits squarely in the losing regime and would likely fail the positive-holdout hard gate. Risk is contained (max_drawdown 18.8%, PF 1.70, avg_trade_return_pct 0.77%, no liquidation), so this is not a fee-death or blowup, but the low trade count is structural (the vol-band gate is designed to trade rarely) and widening it to add trades changes the mechanism rather than tuning it. No parameter change lifts a 62-trade, sub-1.0-Sharpe, 3-years-negative result to a robust, promotable edge. Failure pattern: overfit/no_edge low-trade-count single-name momentum with recent-regime decay (L10/L63).

Long-short BTCUSDT.BINANCE momentum on 4-HOUR bars, gated by a realized-volatility regime band. Every bar the strategy computes (a) realized vol as the standard deviation of the last 90 4H log returns (~15 days) and its PERCENTILE RANK inside its own trailing ~1000-bar history, and (b) BTC's 22-bar (~3.7 day) momentum divided by that realized vol and sqrt(22) — a scale-invariant t-statistic returned as the continuous signal. It goes LONG when mom_z >= +0.8 and SHORT when mom_z <= -0.8, but ONLY while the vol rank sits inside the 25th-70th percentile band: below it the tape is dead-calm chop where momentum whipsaws into sub-fee losses, above it is panic/blow-off vol where trend entries get run over by violent mean reversion. Outside the band the strategy stands flat — for a held position, vol leaving the band is itself an exit, alongside momentum flipping through zero. Because the gate is a percentile of BTC's own vol history rather than a hardcoded vol level, the band means the same thing in 2020 and 2026. Single instrument, no extra legs, sized at 80% of equity notional with leverage 1.0 (no margin used).

CLEAN RESTART 2026-09-04 — this run's verdict history and learning records were removed and it was restarted from verification. Its previous abandonment came from the pipeline, not from the market: the Layer-2 harness mis-bound @staticmethod helpers (fixed), QA issued terminal performance verdicts on an unoptimized smoke test (removed — QA now judges correctness only), and sandbox timeouts came from backtest-slot starvation (fixed). The hypothesis and the strategy code are unchanged. Verify the code on its merits; performance is decided later by the full backtest and the optimizer.

Correct asset/family: BTC directional momentum is the one surviving paper-stage mechanism

Contained risk: max_drawdown 18.8%, profit_factor 1.70, avg_trade_return_pct 0.77% (above fee floor), no liquidation

Vol-regime gate is a genuinely orthogonal idea vs the existing BTC-momentum pipeline variants

Only 62 trades over 6.6 years (exposure 1.97%) — below the ~100-trade measurability floor (L10); walk-forward folds would fit noise

Sharpe 0.569 with CI [-0.17, 1.28] straddling zero, PSR 0.686, information_ratio -0.78 — no significant edge and underperforms buy-hold

Edge decayed: 2024 -1.4%, 2025 -0.3%, 2026 -8.7% (rolling Sharpe negative through 2026); the last-20% holdout sits in this losing regime and would likely fail the positive-holdout hard gate

Base Sharpe ~0.57 is the ceiling (L63) — deflation/holdout only lower it, well short of promotion