ETH 15-Minute Large-Impulse-Bar Continuation (Single-Instrument USD-M, Long-Short, Fast Intraday, Ride Directional Impulse Bars, Tight Initial Stop, Trailing-Stop Winners, Discrete Capital-Capped, Pure OHLCV)

A LONG-SHORT, SINGLE-INSTRUMENT, pure-OHLCV FAST-INTRADAY MOMENTUM-CONTINUATION strategy on ETHUSDT.BINANCE USD-M perpetual futures (15-MINUTE bars) -- the intersection of the two strongest validated facts in the factory's impulse-continuation family. (1) ASSET: ETH is the PROVEN impulse asset -- EthImpulseBarContinuationTrendLS1H PASSED optimization at Sharpe 1.40 and is PROMOTING (the only family member to actually pass so far). (2) TIMEFRAME: 15m is the now-validated sub-1H grain -- BtcImpulseBarContinuationTrendLS15m shipped as the first sub-1H instance. Combining proven-asset x proven-grain maximizes transfer probability while filling the single most under-represented horizon bucket (short_1m_15m, 2.6% vs >=10% target). The analyst's standing guidance after the funding-carry and basket-refinement failures is 'route ALL novelty to the validated intraday continuation family'; this is that family at its highest-confidence asset and the under-served fast grain. MECHANISM (mirrors the proven ETH 1H template, faster grain): an IMPULSE bar is a 15m bar whose directional body (close - open) exceeds impulse_mult x ATR(atr_period) AND closes in the strong third of its high-low range -- a high-conviction directional thrust. The strategy enters in the impulse direction to capture short-horizon CONTINUATION, with a TIGHT initial stop and a TRAILING stop that lets winners run. WHY IMPULSE-CONTINUATION IS THE RIGHT VEHICLE FOR SUB-1H (not the banned scalping anti-pattern): it is EVENT-DRIVEN and SELECTIVE -- a high impulse_mult means only RARE, LARGE 15m thrusts qualify, whose continuation move is large enough to clear fees; it does NOT trade every bar or chase sub-0.15% ticks. WHY ETH at 15m: ETH is the proven asset AND deeply liquid -> the cleanest 15m impulse signal and continuation moves large enough to clear fees at the fast grain (the one real risk). WHY IT GENERALIZES WHERE OTHER SINGLE-ASSET INTRADAY DECAYED: impulse-continuation is event-driven/selective, the higher-quality signal that passed the ETH 1H holdout, unlike VWAP-continuation (rode noisy excursions, decayed) and single-asset TSM/continuous-trend (whipsawed -- a DIFFERENT mechanism). WHAT IT AVOIDS (every closed path): NOT funding/carry/cross-venue (comprehensively closed -- uncreditable funding + HL-data wall); NOT a basket refinement (prohibited axis); NOT single-asset TSM/continuous-trend (dead); NOT reversion (dead); NOT continuous-rebalance/netting-flip (the 316%/385% sizing-artifact cause -- this uses DISCRETE enter-then-exit-to-FLAT, like the cleanly-sized ETH 1H sibling). EVERY design choice respects the lessons: (1) DISCRETE entries, flat between signals, sized against STABLE realized equity (excl. unrealized PnL), hard-capped <=100% (leverage 1x). (2) FEE-AWARE BY CONSTRUCTION: a high impulse_mult restricts entries to large 15m thrusts whose continuation clears the ~0.10% RT floor -- NOT scalping; the backtest must confirm avg net trade > 0.15%. (3) DENSE: 15m gives ample impulse events -> robust 3-window walk-forward population. (4) PURE OHLCV. Per the 'favor simpler' mandate, 3 tunables (impulse_mult, trail_mult, init_stop_mult) mirroring the proven ETH config.

Implements the validated intraday impulse-continuation family at its highest-confidence asset (ETH, the only family member to pass optimization, Sharpe 1.40, promoting) and the validated sub-1H grain (15m, shipped via the BTC 15m sibling), maximizing transfer probability and filling the under-represented short_1m_15m bucket. The impulse definition mirrors the proven template: a directional body exceeding impulse_mult*ATR(14) with the close in the strong third of the range (close_pos >= 0.667 for bullish, <= 0.333 for bearish), so only rare, large 15m thrusts qualify -- event-driven and selective, NOT the banned every-bar scalping anti-pattern, and a high impulse_mult keeps continuation moves large enough to clear the ~0.10% RT fee floor. calculate_signal returns the continuous signed body/ATR so Layer-2 frozen-signal detection passes, while the discrete impulse decision lives in should_enter. Exits use one rising-floor stop combining a tight init_stop_mult*ATR initial stop (cuts failed impulses fast) with a trail_mult*ATR trailing stop from the running extreme (lets winners run), plus a 96-bar (~24h) safety cap; entry/extreme context is captured on entry and reconstructed from avg_px_open if lost. Critically it uses DISCRETE enter-then-exit-to-FLAT (one position, no continuous rebalance / netting flip), the cleanly-sized pattern of the ETH 1H sibling that avoids the 316%/385% sizing artifact; sizing is risk-based against stable realized equity (get_account_equity reads the futures balance, excluding unrealized PnL) and hard-capped at max_notional_frac (50%) of equity at leverage 1.0, so exposure stays <=100% and there is no oversizing. Single-instrument pure-OHLCV on a deeply-liquid major avoids the funding/carry/cross-venue, basket-refinement, and tick/options walls; 15m gives ample impulse events to populate a 3-window walk-forward. Three tunables mirror the proven ETH config. leverage stays 1.0, referenced only as a non-amplifying sizing multiplier, so the unused-leverage gate does not apply. The backtest should confirm avg net trade return > 0.15% (the one real risk at this grain); if continuation does not clear fees on 15m ETH, that is the signal to abandon, but the proven ETH-1H edge plus validated 15m grain make this the highest-probability fast-grain instance.

The impulse-continuation family (validated at ETH 1H, promoted at Sharpe 1.40) does NOT transfer to the 15m grain: it LIQUIDATED on 2022-08-04 (total_return -100%, max_drawdown 100%, dead 2023-2026) with a losing signal (profit_factor 0.757 < 1.0, Sharpe -0.30, win_rate 27%, losing every year). The assassin is FEES: at 15m the strategy fired 4244 trades at 229x turnover for a commission_pct_of_gross of 20.35% ($100k commission), and avg_trade_return is -$37 -- the average NET trade is negative, directly failing the hypothesis's own stated fee-viability gate ('the backtest must confirm avg net trade > 0.15%'). The 'fee-aware by construction' claim (a high impulse_mult restricts entries to large thrusts whose continuation clears fees) is empirically falsified: 15m impulse continuation moves do not clear the ~0.10% round-trip at this firing rate. Compounding it, the discrete-entry sizing blew past its own cap (avg_position_pct 103% vs the intended max_notional_frac 0.50, ~2x) and produced physically-impossible death-spiral prints (-651.9% on 2022-07-29, +324.4% on 2022-07-18, return_kurtosis 755) into the liquidation. Not optimize: a series that liquidates with impossible bars and PF 0.757 cannot be optimized, and the 3 tunables cannot lift a sub-fee edge above 20%-of-gross costs. Not iterate: fixing the sizing blowup would only surface a money-loser whose gross edge is below fees at 15m, and the ETH 1H instance already captures the working version. FAILURE PATTERN: the impulse-bar-continuation mechanism is grain-specific (ETH at 1H) -- pushing it to a faster 15m grain inverts the fee arithmetic (commission 20% of gross at 229x turnover, avg net trade NEGATIVE), turning PF>1 into PF<1, and the discrete-entry over-exposure/stacking bug (avg_position 103% vs a 50% cap, impossible -650% prints) amplifies the bleed to a -100% liquidation -- the same liquidation signature as the XRP 1H impulse sibling. 'Proven asset x proven grain' is falsified: 15m is not a proven grain for ETH impulse; it is fee-death territory, and a high impulse_mult does not make a sub-1H impulse strategy fee-viable.