Volatility-Compression (Squeeze) Breakout, Long-Short Futures (Single-Instrument ETHUSDT.BINANCE USD-M)

Implements the hypothesis directly: BB-inside-KC compression detection, no directional guess during the squeeze, entry only on release confirmed by the momentum-histogram sign, chandelier/ATR-trailing exit, never averages down, long+short on BINANCE USD-M (correct venue — it shorts and uses leverage), and exactly three searched parameters (squeeze_len, mom_thresh, chandelier_atr_mult). Rather than pick defaults by feel I replayed the exact indicator stack and gate on the real ETHUSDT 4-HOUR catalog bars. Over the last 365 days (2,191 bars) the squeeze is on 24% of the time and the release gate fires 196/371/590 times at release_window 3/6/10, so the entry condition is comfortably non-degenerate for Layer 3 (L21/L25: no warmup-vs-window or over-selective-AND problem — warmup is 45 bars against a 2,190-bar window). I then simulated the complete strategy including the chandelier exit and a 0.10% round-trip fee across a 3x3 grid of mom_thresh x chandelier_atr_mult: every cell produced 33-69 trades/year with a POSITIVE average net per-trade return of +0.61% to +2.29%, i.e. 4-15x the 0.15% futures fee floor, and win rates near 40% with positive expectancy (the classic breakout profile — the compression filter really does suppress turnover as the hypothesis claims). On the full 6.5-year history the same config gives 351 trades at +0.215% net average, still above the floor. I deliberately did NOT tune to the best cell (chandelier 4.0 scored highest); the shipped defaults are the canonical Chandelier Exit multiple of 3.0 and the standard LazyBear length of 20, leaving the genuine search to the optimizer rather than handing it a pre-overfit point. Being pure OHLCV, the strategy reads no supplementary feed at all, so the whole L35 class of missing-supp-data verification loops cannot apply and there is no data-availability or staleness path to degrade. Per L15 sizing is double-capped — 2% of equity risked at the ATR chandelier AND an independent gross-notional ceiling — so drawdown cannot compound through position size, and the single-position, no-add design means exposure can never grow after entry. Leverage 2.0 is actively consumed in position_size via self.config.leverage, not inert. The chandelier is tested intrabar and before ratcheting on the same bar's extreme, so a violent bar exits on that bar rather than a full 4h later, and the trailing distance never loosens.

Fails the fee-viability gate at BACKTEST_REVIEW: avg_trade_return_pct is 0.093%, below the 0.15% futures fee floor (L22), which is disqualifying regardless of the positive headline. The hypothesis's core claim — that the volatility-compression filter suppresses turnover enough to make the breakout fee-survivable — is directly falsified: the per-trade edge sits below the round-trip cost, impact modeling consumes 39.3% of gross PnL, and capacity_usd is only $647k (the edge exists only at toy scale). profit_factor 1.02, Sharpe 0.100 with a 95% CI [-0.544, 0.726] straddling zero, and information_ratio -0.74 confirm there is essentially no risk-adjusted edge; total_return is just 11.2% over ~6.5 years with lumpy, often-negative years (2021 -13.9%, 2022 -5.8%, 2026 -11.1%). This is the OHLCV breakout/momentum-continuation class (L9/L41, near-zero survival). A sub-fee-floor per-trade edge cannot be fixed by parameter tuning — there is no parameter region where a 0.09% edge clears a ~0.10% cost plus 39% impact drag, and best-of-225 selection on a Sharpe-0.10 signal would tune to noise and fail deflated-Sharpe/holdout. Abandon rather than spend 2 hours optimizing.

Long/short volatility-compression breakout on ETHUSDT.BINANCE USD-M perpetual (4-HOUR bars, 2x leverage), implementing the LazyBear Squeeze Momentum archetype. Every bar it computes Bollinger Bands (sma +- 2.0 x stdev) and Keltner Channels (sma +- 1.5 x sma(true_range)) over `squeeze_len` bars: the squeeze is ON when the bands sit entirely inside the channels (compression) and OFF when they sit entirely outside (expansion). It also computes the LazyBear linear-regression momentum histogram — the endpoint of a least-squares line fitted to close minus the average of the Donchian midpoint and the SMA — and returns it divided by its own rolling standard deviation as the continuous signal, so the value varies every bar and its sign is the breakout direction. Direction is never guessed during the squeeze: an entry requires a genuine RELEASE (squeeze OFF on this bar AND ON at some point in the previous `release_window` bars) plus |mom_z| >= `mom_thresh`, then goes long on a positive histogram and short on a negative one. The only exit is an ATR chandelier at `chandelier_atr_mult` x ATR(14) behind the running favourable extreme since entry, ratcheting so it never loosens and tested intrabar against the bar's low/high. Sizing risks 2% of equity at the initial chandelier distance, hard-capped at 25% x leverage (= 50%) of equity notional, with no add or average-down path.