AvaxBreakoutThrowbackContinuationLS

AVAX Breakout-Throwback Continuation — Long-Short, Single-Instrument: Enter on the PULLBACK Retest of a Freshly Broken Range Level (Not the Break Itself), Tight Stop Below the Reclaimed Level, Measured-Move Target (BINANCE AVAXUSDT USD-M, 4H Bars, Pure OHLCV, 3-Parameter)

A LONG-SHORT, single-instrument directional-continuation strategy on AVAXUSDT.BINANCE USD-M perp that captures the surviving edge class of the factory (directional momentum on a liquid alt) via a DISTINCT, non-clone entry structure: the BREAKOUT THROWBACK. Instead of buying the raw breakout (the vol-squeeze expansion breakout is a proven net loser — breakouts fail too often and the entry stop is far), it WAITS for price to break a recent range extreme, then PULL BACK to retest the broken level (former resistance becoming support, or vice versa), and only enters when the level HOLDS and price resumes in the breakout direction. This deliberately sits outside every dead/blocked family: NOT dual-TF momentum-confluence (0/213, the EMA-agreement template just killed 3x), NOT raw breakout-continuation (vol-squeeze, net loser), NOT the CME-gap or liquidity-sweep reversals, NOT funding/OI-gated (L70/L71), NOT liquidation-feed (only 68d of data), NOT Polymarket (data uncollectable in sandbox), NOT options (timeout graveyard), NOT cross-venue/rotation (injection/0-125 deaths). It uses ONLY OHLCV on multi-year Binance history (no significance trap, no data-availability risk). It structurally answers the two dominant killers: (1) L19 DRAWDOWN BREACH — the throwback entry sits right at the reclaimed level, so the stop is TIGHT (just beyond it) and each loss is small, capping per-trade and portfolio drawdown, unlike the open-ended trend clones that hit 56-76% DD; (2) FEE GRAVEYARD — the tight stop plus a measured-move target gives a high reward:risk on a multi-percent 4H alt move, so per-trade capture is many times the 0.10% round trip, not a sub-fee scalp. Traded on a liquid ALT the recent queue hasn't used (AVAX, not BTC/ETH/SOL/ADA/XRP), long-short (fills the starved 14.3% bucket), 3 tunable parameters.

Implements the hypothesis's distinguishing structure literally - the entry is the retest, never the break - and stays clear of every family the brief rules out: no dual-timeframe momentum confluence, no raw breakout continuation, no gap or sweep reversal, no funding/OI gate, no liquidation feed, no cross-venue leg, no basket, no options, no supplementary data at all. (1) The two named killers are answered structurally, and I measured both on the real AVAXUSDT 4H series (13,013 bars, 2020-09 to 2026-08) by driving the actual strategy class through a bar-by-bar harness rather than trusting the design on paper. Drawdown: because the reclaim must happen within retest_atr_tol of the level, the stop distance averages 4.4% of price and every loss is bounded by risk-first sizing across that exact distance, with average deployed notional at 26% of equity. Fees: average per-trade result is about +1.24% net of a 0.10% round trip, roughly twelve times the cost floor, because the payoff comes from a measured-move target on a multi-percent 4H alt move rather than a scalp. (2) The harness also confirms the state machine actually works end to end: 314 setups armed, 243 of them converting to entries, exits split 170 stop / 43 measured-move target / 30 time stop - the intended shape of a tight-stop, high-reward-to-risk continuation book, with a 28% win rate carried by the target tail. That is about 40 trades a year, so the sandbox year will produce a healthy sample rather than risking a zero-trade abandon. (3) Honest statistics: with 243 trades and a per-trade standard deviation near 10%, the standard error on the mean is 0.67%, so +1.24% is roughly 1.8 sigma - clearly positive in sign and stable across the whole parameter neighbourhood I checked (every lookback 35/45/55 x stop 0.5/0.75/1.0 combination came out positive, +0.58% to +1.45%), but not yet decisive. I deliberately set the defaults to the CENTRE of that plateau (lookback 45, tol 1.0, stop 0.75) rather than the sweep argmax, since the spread between neighbours is inside noise and picking the maximum would be in-sample fitting. (4) One robustness detail worth flagging: should_enter optimistically records the trade state, but the base template can still drop the order on the size or min-notional guard, which would leave the machine believing it holds a position and silently stop arming new setups; _sync_position_state checks the cache one bar after entry and releases the machine if no position materialised. (5) Per-bar cost is O(1) - bounded deques for the Donchian extremes and the ATR plus a few comparisons, nothing rescanning history - so 4H bars over six years are far inside the wall-clock budget. leverage stays 1.0 and is read in position_size so the notional cap is exact rather than inert.

ohlcv_breakout_continuation_subfloor_sharpe_regime_concentration: single-instrument pure-OHLCV directional-continuation (breakout-throwback) on AVAX — the OHLCV trend/breakout family with 0/213 survival (L9/L41/L63). Base Sharpe 0.204 with CI [-0.44, 0.92] straddling zero is statistically indistinguishable from no-skill and far below the 0.5 OOS floor; information_ratio -0.51 (negative alpha vs buy-hold). The result is entirely front-loaded in the 2021-2022 bull/high-vol years (annual +30.7, +9.7) while 2023 (-6.8), 2025 (-4.4) and 2026 (-4.8) are all net losers — regime-concentration, not a persistent edge (rolling Sharpe deeply negative through 2023 and most of 2025). avg_trade_return_pct 1.45% clears the fee floor so it is not a fee death, but per L63 base Sharpe is the ceiling for a directional OHLCV book and tuning 3 params over ~225 trials would curve-fit the bull-year concentration into a DSR/holdout death, not reach the promotion floor. Not worth the 2-hour optimization run.

Long-short, single-instrument continuation strategy on AVAXUSDT.BINANCE USD-M 4H bars that trades the THROWBACK, not the breakout. A close beyond the prior lookback-bar range extreme arms a setup, recording the broken level and the range height as a measured move. The setup only becomes a trade if, within max_wait_bars, price pulls back to within retest_atr_tol x ATR of that level and then closes back through it on the breakout side while still near it - former resistance holding as support, or the mirror for shorts. Entry is therefore at the level, so the protective stop sits just beyond it at stop_atr_mult x ATR (measured on real AVAX history that is about 4.4% of price), the target is the full measured move, and a time stop closes anything still open after max_hold_bars. A decisive close back through the level before the reclaim marks a failed break and disarms the setup. Both stop and target are tested intrabar against the bar's low/high. Sizing risks a fixed risk_pct of equity across the ACTUAL distance to the stop, capped at 35% of equity notional. Pure OHLCV on multi-year history, one instrument, three tunable parameters.

The real-engine sandbox disagrees with the developer's harness, to the downside. Harness: +1.24%/trade over 243 full-history trades (~1.8σ). Real Layer-3 engine over the recent ~362 days is NEGATIVE: total_return -5.84%, PF 0.71 (<1.0), Sharpe -0.60, win_rate 0.219, expectancy -$182/trade. RR-implied breakeven win rate ~28.5% (avg_win $1992 / avg_loss $791); full history ~28% (barely breakeven), sandbox year 21.9% (losing). A marginal, win-rate-sensitive edge — not a demonstrated refutation, not a confirmed edge.

The 'tight stop' is not venue-managed. The breach is detected intrabar (bar_low <= stop_px) but the exit is a should_exit market order filling at the bar close, so on 4H bars a spike/gap that closes beyond the level realizes a loss larger than the sized risk (largest_loss -$1785 ~1.8% vs risk_pct 1%). Undermines the L19 bounding claim and likely explains part of the harness/engine gap (the harness probably credited clean fills at stop_px). Not a bug, but realized risk exceeds sized risk. Consider use_bracket_orders.

avg_trade_return_pct (+0.064%) is misleading: below the 0.15% floor and opposite in sign to the USD outcome (avg_trade_pnl_usd -$182, total_return -5.84%) — losers carried larger notional. Also L7: pure-OHLCV directional continuation on a liquid alt is a 0/213 family and the raw breakout variant is a known net loser; the throwback is distinct but in that broad family. Judge on full-history real-engine metrics.

warning

Pure OHLCV on multi-year liquid AVAX data — no data-availability or injection risk, well-measured 241-trade sample

avg_trade_return_pct 1.45% clears the 0.15% futures fee floor comfortably — not a fee-fragility death

Risk-first sizing keeps max_drawdown moderate (15.9%), avoiding the L19 >50% breach of open-ended trend clones

Sharpe 0.204 with CI [-0.44, 0.92] straddling zero — no risk-adjusted edge; far below the 0.5 OOS floor and ~1.0 ceiling for a directional OHLCV book (L63)

information_ratio -0.51 — negative alpha vs AVAX buy-and-hold

Edge is entirely concentrated in 2021-2022 (annual +30.7, +9.7); 2023 (-6.8), 2025 (-4.4), 2026 (-4.8) are all net losers — regime-concentration, not persistence

OHLCV single-name directional-continuation family (0/213 survival, L9/L41); throwback framing doesn't change the class outcome

profit_factor 1.12, win_rate 29.9%, kurtosis 13.6 — thin, outlier-dependent

AvaxBreakoutThrowbackContinuationLS was a thoughtful attempt to capture the factory's surviving directional-momentum class through a distinct, non-clone entry — the breakout throwback, entering the retest of a freshly broken level rather than the break itself — explicitly designed to fix the two dominant killers with a tight stop (holding drawdown to 15.9%) and a measured-move target (per-trade capture of 1.45%, well above fees). On those two dimensions it succeeded, but the underlying edge was absent: over 241 trades and ~5.9 years the Sharpe was just 0.204 with a confidence interval straddling zero, alpha versus AVAX buy-and-hold was negative, and all the profit came from 2021-2022 while 2023, 2025, and 2026 were net losers. The analyst abandoned it at the backtest-review gate, concluding it was another single-name OHLCV directional-continuation clone (0/213 survival) whose regime-concentrated, sub-floor Sharpe was the ceiling — so optimization would only curve-fit the bull years into a DSR/holdout death, not worth the run.

Successfully engineering away the known failure modes (tight stops fixed the drawdown, a measured-move target cleared the fee floor) still does not manufacture a directional edge — a pure-OHLCV single-name continuation strategy whose profits are concentrated in one bull-market regime has no persistent, statistically-distinguishable alpha, and a clever entry structure does not change the dead class's outcome.

The analyst issued an 'abandon' verdict at the pre-optimization backtest-review gate: although the design successfully avoided the drawdown breach and fee-death failure modes, its base Sharpe of 0.204 had a CI straddling zero (far below the 0.5 OOS floor), its alpha versus buy-and-hold was negative, and its edge was regime-concentrated in the 2021-2022 bull years rather than persistent. As a single-name OHLCV directional-continuation clone (0/213 survival) whose base Sharpe is the ceiling, tuning three parameters would only curve-fit the bull-year concentration into a holdout death. It never reached optimization, post-opt analyst review, or risk review.

A long-short, single-instrument directional-continuation strategy on AVAXUSDT Binance USD-M perp (4H, pure OHLCV) that used a breakout-throwback entry structure — waiting for price to break a recent range extreme, pull back to retest the broken level, and only entering when the level held and price resumed — with a tight stop just beyond the reclaimed level, a measured-move target, and 3 tunable parameters, deliberately shaped to answer the factory's two dominant killers (open-ended drawdown and sub-fee scalping).

Over a well-measured 241-trade sample across ~5.9 years (133 long / 108 short, 29.9% win rate) it returned 31.85% total with profit factor 1.12, avg_trade_return_pct 1.45% (clearing the fee floor), and a moderate 15.9% max drawdown — but Sharpe was only 0.204 (95% CI [-0.44, 0.92]), information ratio -0.51, and the edge was concentrated in 2021-2022 (+30.7%, +9.7%) while 2023 (-6.8%), 2025 (-4.4%), and 2026 (-4.8%) were all net losers.