BtcVolatilitySqueezeExpansionBreakoutLS

Volatility-Compression Expansion Breakout, Single-Instrument Long-Short (BTCUSDT.BINANCE USD-M, 1H Bars — Trade the Directional Expansion ONLY After a Genuine Multi-Week Volatility SQUEEZE (Coiled Range at a Realized-Vol Extreme Low); Enter on the First Decisive Range Break, Cut False Breaks Fast, Ride the Expansion with an ATR Trailing Stop, 3-Parameter)

A LONG-SHORT, SINGLE-INSTRUMENT, SINGLE-VENUE volatility-regime strategy on BTCUSDT.BINANCE USD-M perpetual (~0.10% RT taker), on 1H bars. It is a DIFFERENT family from the factory's trend-confluence survivor (which does not transfer off BTC — the SOL clone overfit) and from every falsified positioning/flow and reversal family: the signal is the VOLATILITY REGIME, not price direction per se and not any leverage/flow feed. The mechanism is the well-documented volatility cycle: realized volatility mean-reverts, so an unusually TIGHT, coiled range (realized vol at a multi-week extreme low = a 'squeeze') is mechanically followed by an expansion, and the expansion's initial break tends to run because it releases pent-up order flow (stops and breakout entries stacked at the range edges). The strategy does nothing during normal volatility and acts ONLY when a genuine multi-week compression is present, then trades the FIRST decisive break of the compression range in whichever direction it breaks (long or short), cutting false breaks fast and riding true expansions with an ATR trailing stop. BTC is chosen because it is the highest-survival instrument in this factory (2% vs ~1% elsewhere), has the cleanest, most liquid squeezes, and produces expansion moves of 1-3% that clear the fee floor with margin. Firing only on rare true squeezes keeps turnover low despite the shorter timeframe. Deliberately 3 parameters to resist the overfitting that is the factory's #1 killer. It also targets the under-represented short horizon bucket (9.2% vs ≥10%).

IMPLEMENTATION. The hypothesis's three ingredients map onto one continuous decision variable plus one regime gate: signal is price's position inside the compression range in half-range units (varies every bar, squeeze or not, so nothing freezes), and the squeeze test lives in should_enter. Both the squeeze test and the break test are scale-invariant by construction — the vol test is a PERCENTILE RANK within a 30-day window rather than any absolute vol number (L5), and 'decisive' is defined as a fraction of the RANGE WIDTH rather than ticks or ATR, so a tighter coil requires proportionally less follow-through. Freezing the broken edge and width at entry matters: the rolling range keeps advancing with the expansion, so a live range would move the false-break level under the trade. VALIDATION BEFORE CHOOSING DEFAULTS (measured on the real catalog: BTC 1H, 57,720 bars, 2019-12 to 2026-07, 0.05% per side). Frequency first, because over-selective gates are this factory's most expensive failure: the squeeze fires on 12-17% of bars, and decisive breaks inside one give ~14 trades/year — the 365-day Layer-3 sandbox will see roughly that many, far above the ≥1 requirement, and warmup (30-day rank window + 72-bar range) consumes ~33 days of it. Edge second: at the shipped defaults the strategy nets +0.421% average per trade over 92 trades (win rate 0.36, median hold 26h, 5 of 7 years positive) — about 4x the ~0.10% round trip and well clear of the analyst's 0.15% floor. Stability third, which is what I actually selected on: the FULL 27-cell structural family (squeeze_lookback 48/72/96 x squeeze_pct 0.10/0.15/0.25 x atr_trail_mult 2/3/4) is positive, +0.12% to +1.31% per trade, so the optimizer will be searching a broad plateau rather than a spike. TWO CHOICES I MADE FROM EVIDENCE, NOT TASTE. (1) The decisive-break buffer is essential, not cosmetic: entering on ANY close beyond the edge gives +0.03% per trade with a 4h median hold — the book just pays fees on failed breaks — while requiring 10% of the range width lifts it to +0.42% with a 26h hold. (2) The false-break cut is what the hypothesis calls for and it earns its place: removing it drops the same configuration to +0.08% per trade and 2/7 positive years, because failed breaks are then held all the way to the ATR stop. I did NOT ship the best cell. squeeze_lookback=96 / squeeze_pct=0.15 / atr_trail_mult=4 measures +1.31%, roughly 3x the defaults; picking it would be best-of-N selection on 59 trades. The shipped 72 / 0.15 / 3.0 is a mid-family choice defensible ex ante (a 3-day coil, a bottom-15% vol reading, a conventional 3-ATR trail) and leaves the optimizer honest room. HONEST CAVEATS. The per-year record is uneven — 2020 +2.08% carries a disproportionate share and 2022 (-0.85%) and 2026 (-0.45%) are negative — so a walk-forward will likely show a lower OOS Sharpe than the pooled number suggests, and the win rate is 0.36 (this is a fat-right-tail book: most trades are small false-break cuts, a few expansions pay for them). Trade count is ~14/year, so the holdout window will be thin. Venue is BINANCE USD-M because the strategy shorts; leverage is left at 1.0 and sizing never references it, so there is no unused-leverage flag.

The edge is essentially absent, so it will not clear the post-optimization robustness gates. Sharpe is 0.224 with a 95% CI [-0.42, 0.78] straddling zero (not distinguishable from noise), total_return is only 7.9% over ~6.5 years (~1.1%/yr — flat-to-noise), and information_ratio is -0.70. The very low 3.7% max drawdown is a function of tiny risk deployment, not a strong risk-adjusted edge — the book barely does anything, and annual returns are ±1-3% noise that is negative in 2022, 2024 and 2026. This is the OHLCV breakout/volatility-expansion class (L9/L41, near-zero survival); return_kurtosis 58.6 shows the little PnL there is comes from a handful of outlier trades. profit_factor 1.28 and avg_trade_return_pct 0.388% clear their bars, so this is not fee-fragility, but a Sharpe of 0.22 with a CI including zero and negligible cumulative return cannot reach the 0.5 OOS minimum, and best-of-225 selection would inflate this noise and near-certainly fail deflated-Sharpe/holdout in ANALYZING. There is no strong parameter region to tune toward — the mechanism simply produces almost no return. Abandon at BACKTEST_REVIEW rather than spend 2 hours optimizing.

Long/short BTCUSDT USD-M perp on 1H bars. Realized volatility over the last 72 bars is percentile-ranked against its own trailing 30-day distribution; a squeeze is active when that rank is in the bottom 15% (BTC is at a multi-week vol low). Signal = (close - range_mid) / (0.5 * range_width) of the prior 72-bar compression range — 0 at the coil's middle, ±1 at its edges, beyond ±1 once broken. Inside a squeeze, a close 10% of the range width beyond an edge (signal ≥ ±1.2) opens the trade in the break's direction. Exits: the false-break cut (price closes back inside past the frozen edge by 10% of the frozen width) and a 3×ATR(14) trailing stop from the best close reached. Risk-first sizing (1% of equity risked at the trailing-stop distance, capped at 30% of equity notional), leverage 1x. Pure OHLCV — no supplementary feed. Three tunable parameters: squeeze_lookback_bars, squeeze_pct, atr_trail_mult.

Not code-blocking — the strategy is measurable over the full catalog. The analyst should weight the pooled +0.42%/trade against the developer's own honest caveats at BACKTEST_REVIEW: uneven per-year contribution (2020 +2.08% carries it; 2022 -0.85% and 2026 -0.45% negative), likely lower OOS than pooled Sharpe, and a thin holdout. Consider whether the holdout gate can be meaningfully populated before spending the full 3-phase optimization.

Low trade frequency makes the downstream OOS/holdout gates thin. At ~14-16 trades/year (sandbox: 16 over 363 days; developer's full-catalog validation: 92 over ~7 years), the 15-day holdout window will contain <1 expected trade and each walk-forward window only ~14-28. Over the FULL multi-year backtest the count reaches ~92-100 — at the lower edge of statistical measurability but genuinely measurable (unlike the data-limited liquidation family), and comfortably above the analyst's >=5-trade floor. The sandbox Sharpe is 0.065 with CI [-1.57, 1.77] straddling zero, and win_rate 0.375 with a fat-right-tail payoff (a few expansions pay for many small false-break cuts), so the edge is real but fragile.

Informational for the analyst. The squeeze gate (percentile-rank, scale-invariant) and range-width-relative break test are correctly scale-free, which is the right defense against the L5 absolute-threshold failure; the concern is generalization, which the walk-forward and PBO gates will test.

Pure-OHLCV mechanism class has a poor historical survival rate in this factory, and the edge shows signs of recent decay (2026 negative in the developer's per-year record; sandbox — the most recent 363 days — nets only 0.209%/trade vs the 0.421% full-catalog pooled figure). Volatility-squeeze-expansion is a distinct family from the abandoned Donchian/50-day-high breakout and VWAP-deviation classes, so it is not a re-parameterization of a dead mechanism, but it remains OHLCV-only with no orthogonal confirmation.

Not backtest-blocking. Risk Officer / Portfolio Manager should note the low capacity if this reaches promotion; it does not change the code or the sandbox metrics.

Sandbox reports impact_cost_pct 26.93% and capacity_usd only ~$1.38M against a small per-trade edge (total_impact_usd $140 vs $380 gross PnL). At the $100k backtest scale the fills are trivially liquid on BTC 1H so this does not affect backtest validity, but it flags a real-money capacity ceiling for any future scaling.

Disciplined construction: rare squeeze gate, fast false-break cut, ATR trail; very low max_drawdown (3.7%); pure OHLCV, verification-safe

profit_factor 1.28 (>1.2) and avg_trade_return_pct 0.388% (clears fee floor); adequate trade count (103)

Sharpe 0.224 with 95% CI [-0.42, 0.78] straddling zero — edge not distinguishable from noise

total_return only 7.9% over ~6.5 years (~1.1%/yr) — the strategy is essentially flat; information_ratio -0.70

Low drawdown is a function of tiny risk deployment, not a strong edge — the book barely does anything

Annual returns are ±1-3% noise, negative in 2022/2024/2026 (recent years negative)

return_kurtosis 58.6 — the little PnL comes from a few outlier trades; OHLCV breakout/expansion class (L9/L41, near-zero survival)

This strategy tried to exploit the volatility mean-reversion cycle, sitting idle until BTC coiled into a genuine multi-week squeeze and then trading the first decisive range break in either direction with disciplined false-break cuts and an ATR trailing stop. The 6.5-year backtest was clean and verification-safe with a very low 3.7% drawdown and per-trade returns above the fee floor, but the edge was essentially absent: total return of only ~7.9% (~1.1%/yr), Sharpe 0.224 with a confidence interval straddling zero, and annual returns that were ±1-3% noise, negative in several recent years. The analyst judged that the low drawdown reflected minimal risk deployment rather than a real edge and that best-of-225 optimization would only inflate noise and fail later robustness gates, so it was abandoned at backtest review on its first iteration without proceeding to optimization.

Passing the fee-viability bars (profit factor >1.2, per-trade return above the floor) is not enough — a strategy also needs a Sharpe whose confidence interval clears zero and a meaningful cumulative return, otherwise a low drawdown just signals the book is barely deploying risk and optimization would only overfit noise.

It was abandoned at the pre-optimization backtest review with an 'abandon' verdict: the edge was statistically indistinguishable from noise (Sharpe CI includes zero, ~1.1%/yr return, negative in 2022/2024/2026), the low drawdown reflected tiny risk deployment rather than a real edge, and there was no strong parameter region to optimize toward — so it never reached the 3-phase optimization, analyst, or risk stages.

A long/short single-instrument BTCUSDT USD-M volatility-regime strategy on 1H bars that stayed flat during normal conditions and traded only the first decisive break of a genuine multi-week volatility squeeze (realized vol at a 30-day percentile extreme low), cutting false breaks fast and riding true expansions with an ATR trailing stop, using 3 parameters.

Over a ~6.5-year backtest (2019-2026, 103 trades) it produced a total return of only 7.9% (~1.1%/yr) with Sharpe 0.224 (95% CI [-0.42, 0.78] straddling zero), profit factor 1.28, avg_trade_return_pct 0.388% (above the fee floor), max drawdown 3.7%, and information ratio -0.70, with return kurtosis of 58.6.