AvaxHyperliquidEfficiencyRatioGatedMomentumLS

AVAX Hyperliquid Trend-Quality-Gated Momentum, Long-Short (Single-Instrument AVAXUSD.HYPERLIQUID Perp, 4H Bars, Kaufman Efficiency-Ratio Gate — Take Directional Momentum ONLY When the Trend Is 'Clean'/Efficient, Sit Flat in Choppy/Noisy Regimes, ATR-Trailing, 3-Parameter)

A LONG-SHORT, single-instrument, pure-OHLCV trend-following strategy on AVAXUSD.HYPERLIQUID perpetual futures. It stays in the ONLY family that has survived optimization here (directional momentum/trend, long-short, pure OHLCV — the BTC dual-timeframe confluence at Sharpe ~1.99), but its distinguishing mechanism is a NOISE FILTER, not a new entry primitive: it computes the Kaufman Efficiency Ratio (ER = |net move over N bars| / sum of absolute per-bar moves over N bars) and only takes a directional position when ER is HIGH — i.e. price is moving in a clean, persistent, low-noise line — and sits FLAT whenever ER is low (choppy/whipsaw regime). This directly attacks the #1 and #2 killers in this factory (overfit 249x, fee_edge 193x): most momentum/breakout strategies die because they keep firing in choppy regimes where whipsaw losses and fee drag overwhelm the occasional clean trend; the ER gate structurally suppresses trading in exactly those regimes, so the strategy only pays fees when a real directional move is underway. It is materially different from the three strategies already in this factory's pipeline: it is NOT a channel breakout (SOL HL Donchian), NOT an RSI pullback entry (ETH HL continuation), and NOT a two-signal momentum confluence (BTC dual-TF) — the trade filter is a trend-cleanliness metric, and it uses a fresh asset (AVAX) on the most under-represented data-safe venue (HYPERLIQUID at 5.9% vs 20% quota). Fills the long_short direction gap (13.7% vs target). Uses 4H bars on Hyperliquid (~2.3 years usable candle history — above the HL sub-hourly wall), pure OHLCV, no supplementary feed (best-surviving tag family at 4%). Structurally avoids every recent death mode: no options-expiry cadence, no cross-venue schema-merge/spot-lookup bug, no funding fee-floor, no 90-day liquidation-feed wall, no COIN-M booking defect.

Implements the hypothesis directly: Kaufman Efficiency Ratio as a trend-cleanliness gate on 4H AVAXUSD.HYPERLIQUID, long-short, pure OHLCV, ATR trailing, three tunable parameters. Four notes. (1) The mechanism collapses into ONE continuous decision variable rather than a conjunction of filters: keeping the sign on the efficiency ratio makes magnitude=conviction and sign=direction, so the entry test is a single threshold. That matters because stacked compound gates are the top cause of zero-trade verification failures in this factory — here there is literally nothing to stack, and the signal is returned raw every bar in its natural [-1,1] units with no clipping or gated default. (2) I calibrated the threshold on real data before shipping instead of picking a round number: AVAXUSD.HYPERLIQUID has 5,124 4H bars back to 2024-02-07, and with er_window=20 the |signal| distribution runs median 0.21 / 75th pct 0.34, so the 0.35 default admits roughly the cleanest quartile of regimes while still leaving 236 long and 383 short eligible bars inside the 365-day Layer-3 window. Selective enough to be a real filter, nowhere near a zero-trade gate. (3) This is a single-feed strategy — 4H bars only, no extra instruments, no second timeframe, no supplementary data — which removes the whole class of failures where an extras leg is silently dropped or misaligned, and keeps per-bar cost O(1) (running sums for both the efficiency denominator and the ATR, with periodic fsum re-seeding against float drift). (4) Risk controls follow the compounding-blowup lesson: 1% of equity risked per trade across the initial 3-ATR stop, gross notional hard-capped at 1x equity, no leverage set or referenced, and the re-entry cooldown is anchored to bar timestamps rather than a bar counter so it is restart-safe and identical in backtest and paper. One honest caveat for the analyst: an efficiency gate is a lagging regime read — it confirms a clean trend only after N bars of it — so the entry is later than a breakout's by construction; the bet is that the improved signal-to-noise more than pays for the worse fill, and er_window is the first parameter I would scrutinise if it does not.

No edge to optimize. Across a decisive 146-trade sample the ER-gated momentum strategy is a net loser: profit_factor 0.81, avg_trade_return_pct -0.282% (negative, below the fee floor), expectancy -$75/trade, Sharpe -0.21 (CI [-1.17, 0.89] straddling zero), total_return -10.3%. It deteriorates monotonically — 2024 +6.2%, 2025 -4.4%, 2026 -12.4%, rolling Sharpe ending near -8 — so this is a persistent absence of edge, not a single bad regime. The Kaufman Efficiency-Ratio gate was the whole thesis (only trade 'clean' trends to dodge chop/fees); despite favorable trade geometry (avg_win $992 > avg_loss $582) the 32% win rate shows the gate does not select regimes where AVAX's direction call is correct. This is the pure-OHLCV momentum/trend class (zero survivors) with a per-trade edge that is negative net of costs. Tuning er_window/er_threshold/atr_trail_mult cannot flip a negative expectancy that holds across 146 trades and every recent year; it would only overfit the lone positive 2024 window and fail walk-forward OOS/holdout. Abandon rather than spend 2 hours optimizing.

Long-short 4H trend-following on the AVAXUSD Hyperliquid perpetual, gated by trend quality rather than by a new entry trigger. Each bar it computes the signed Kaufman Efficiency Ratio — net price displacement over the last 20 bars divided by the total distance price actually travelled — so magnitude in [0,1] measures how clean/straight the move is and the sign gives direction. It goes long when that value is at or above +0.35, short at or below -0.35, and stays flat in between (choppy, low-efficiency regimes), so it only pays fees while a genuine directional move is underway. Exits use a 3x-ATR chandelier trail from the extreme reached since entry, or an efficiency sign flip against the position, with a 12-hour calendar-anchored cooldown after each exit. Sizing risks 1% of equity across the initial stop with gross notional capped at 1x equity; leverage 1.0. Exactly three tunable parameters (er_window, er_threshold, atr_trail_mult); ATR window, risk fraction, notional cap and cooldown are locked constants.

Judge on the FULL ~2.3yr HL history; given avg_trade_return_pct is deeply below the floor, be prepared to abandon at backtest_review unless the full window + er_window tuning lift the win rate above break-even.

CLEARLY NEGATIVE SANDBOX — the weakest of the HL batch. total_return -11.7%, profit_factor 0.54, win_rate 24.6%, avg_trade_return_pct -0.88% (well below the 0.15% floor) over 57 trades. Materially worse than the SOL Donchian sibling (-0.276%, PF 0.86). NOT statistically decisive though: per-trade SE ~0.66% gives t ~-1.3, Sharpe CI [-2.34, +1.10] straddles zero — one bad 320-day window within noise. Reward:risk geometry is FAVORABLE (avg_win $987 vs avg_loss $595 = 1.66x), so the exit is sound; the entire deficit is win rate (24.6% vs ~37.6% break-even).

Test whether a shorter er_window (earlier confirmation) or ER-as-filter-on-a-separate-trigger raises the win rate. If the low win rate persists across the full history, the gate confirms trends too late to trade — abandon, don't curve-fit.

PLAUSIBLE STRUCTURAL LATE-ENTRY FLAW. A high |ER| (>=0.35) means the trend has ALREADY traveled far and cleanly over the prior 20 bars, so entering with-trend then is a late entry near exhaustion — consistent with the 24.6% win rate. Not a code bug: the developer honestly flagged it and named er_window as the first param to scrutinize. Kaufman's own use of ER is as a position-sizer/AMA smoother, not a standalone entry trigger, which corroborates.

Require robust walk-forward with OOS Sharpe clearly above zero before promoting; do not read the dry-run's 236/383 eligible BARS as trade count — the engine produced 57 trades.

Pure-OHLCV momentum/trend class prior (L7) is unfavorable (zero OHLCV-only survivors). The ER trend-quality gate is a genuinely fresh filter within the surviving momentum family and the code is clean, but the class prior plus the deeply negative sandbox warrant a high OOS bar.

Clean, decisive sample (146 trades, 872 days) — the negative result is trustworthy, not noise

Balanced two-sided book (69 long / 77 short) and favorable trade geometry (avg_win $992 > avg_loss $582)

Net loser: profit_factor 0.81, total_return -10.3%, Sharpe -0.21 (CI [-1.17, 0.89] straddles zero)

avg_trade_return_pct -0.282% — negative per-trade edge, below the trading-cost floor

Deteriorating every year: 2024 +6.2%, 2025 -4.4%, 2026 -12.4%; rolling Sharpe ends near -8

32% win rate means the Efficiency-Ratio gate — the entire distinguishing mechanism — failed to select regimes with a real directional edge

OHLCV momentum/trend class with PF < 1.2 (zero survivors historically)

This strategy aimed to survive where most momentum systems die by adding a noise filter rather than a new entry primitive: the Kaufman Efficiency Ratio would keep it flat in choppy AVAX regimes and only trade when price moved in a clean directional line, targeting the two biggest factory killers (overfitting and edge-below-fees). But across a trustworthy 146-trade, 872-day sample it lost money — profit factor 0.81, Sharpe -0.21, and a negative per-trade return below the trading-cost floor — with performance deteriorating every year. The 32% win rate showed the efficiency gate, the entire thesis, failed to select regimes with a genuine directional edge. The analyst abandoned it at the backtest-review gate before any optimization, judging that no parameter tuning could flip a negative expectancy this persistent, so it never reached the 3-phase optimizer, risk review, or paper trading.

A regime-selection filter like the Efficiency Ratio gate does not create edge on its own — a 32% win rate proves the gate failed to isolate regimes where the direction call was actually correct, and favorable win/loss geometry cannot rescue a negative-expectancy momentum system in the zero-survivor pure-OHLCV trend class.

The analyst abandoned it at the pre-optimization backtest-review gate with an 'abandon' verdict, concluding there was no edge to optimize — a negative per-trade expectancy holding across 146 trades and every recent year that tuning could not flip, only overfit.

A long-short, single-instrument pure-OHLCV trend-following strategy on AVAXUSD.HYPERLIQUID 4H perpetuals that used the sign and magnitude of the Kaufman Efficiency Ratio to take directional momentum positions only in clean, low-noise trending regimes and sit flat during chop, with an ATR chandelier trail and just three tunable parameters.

Over an 872-day backtest with a decisive 146-trade sample (69 long / 77 short), it was a net loser: total return -10.3%, profit factor 0.81, Sharpe -0.21 (CI [-1.17, 0.89]), expectancy -$75/trade, and avg_trade_return_pct -0.282% (negative, below the fee floor). Despite favorable trade geometry (avg_win $992 > avg_loss $582), the 32% win rate and monotonic yearly decline (2024 +6.2%, 2025 -4.4%, 2026 -12.4%) showed no real edge.