EthDailyDonchianBreakoutLongFlat

ETH USD-M Daily Donchian Breakout, LONG/FLAT Only (Single-Instrument BINANCE Perp, Pure OHLCV, Promoted-Book-Identical Fixed Single-Entry Sizing, NO Dynamic Resizing, NO Long-Short Flips, Leverage 1.0, Engine-Bug-Avoiding by Construction, Low-Parameter)

A LONG-ONLY (LONG/FLAT, never short), SINGLE-INSTRUMENT, DAILY Donchian-channel breakout trend strategy on ETHUSDT.BINANCE (USD-M perpetual), deliberately built as the cleanest construction that the current engine can plausibly evaluate. Context: the factory's backtest engine currently has confirmed, strategy-unfixable defects — (1) a USD-M perp base-pipeline OVER-EXPOSURE bug proven immune to hard caps/dual flat-guards (SOL vol-targeted fresh run still hit 345%); (2) spot-CASH held-position MTM broken (unbacktestable); (3) multi-instrument aggregation broken; (4) all census signals (liquidations/OI/account-ratio/taker) data-walled. CRUCIALLY, the over-exposure cluster ALL shared two traits the PROMOTED books do NOT: dynamic per-bar re-sizing (vol-/funding-scaling) and/or frequent long->short FLIPS; the promoted BTC/ETH/SOL TS-momentum books (which sized correctly enough to promote) use simple fixed single-entry sizing and hold. This hypothesis therefore copies the PROMOTED construction exactly and strips every over-exposure trigger: LONG/FLAT only (never holds a short -> can never stack opposing positions on a flip), ONE fixed-fractional position size computed ONCE at entry (NO dynamic resizing while held), leverage 1.0, and explicit close-to-FLAT before any re-entry. It uses OHLCV ONLY (the deep wall-free feed), trades the validated daily breakout/trend edge with-trend on ETH (a validated core trender, promoted TS-momentum sibling), and is single-instrument/single-venue (no aggregation bug) and perp not spot (no spot-MTM bug). It is the ETH sibling of the pending BTC long/flat Donchian, extending the maximally-evaluable construction across the two cleanest trenders. PRE-REGISTERED ACCEPTANCE / ENGINE-DEFECT DETECTOR: avg_position_pct must be ~<=100% (one position, no stacking) and no single-day return beyond ~+/-15-20% (ETH at 1x); if violated, it is the engine over-exposure defect (escalate/abandon), not the strategy. Low parameter count.

Implements the hypothesis by copying the promoted TS-momentum book's sizing/holding discipline and stripping every trait shared by the engine's over-exposure defect cluster. (1) LONG/FLAT only: should_enter returns BUY or None, never SELL, so the strategy can never hold a short and can never stack opposing legs on a flip. (2) Fixed single-entry sizing: position_size() returns a fixed fraction of equity and is only invoked by the base class while flat, so there is no dynamic per-bar re-sizing while held and no signal-magnitude scaling — the exact trigger the over-exposure runs shared. (3) leverage 1.0 with notional hard-capped at equity, so avg_position_pct should sit ~<=100% and no single-day return should exceed ETH's 1x daily move (~+/-15-20%) — matching the pre-registered engine-defect detector. (4) OHLCV-only (deep wall-free feed), single-instrument/single-venue perpetual (no multi-instrument aggregation bug, no spot-CASH MTM bug). (5) Explicit close-to-FLAT before re-entry is guaranteed by the base class refusing entry while a position is open. The edge is the validated daily Donchian breakout/trend-continuation on ETH (a promoted core trender), with a 100-day SMA regime filter to trade only with-trend and an ATR + inner-Donchian exit to cut losers. Signal is continuous (close position within the entry channel, normalized to half-width) so it varies each bar and thresholds cleanly at +1 for a fresh breakout.

Failed deflated Sharpe: DSR=0.0299 (vs 0.95 bar), with the optimized Sharpe 0.751 BELOW the expected-max best-of-225 luck bar of 1.389 (is_significant=false, sharpe_ci_low -0.503 straddling 0, PBO 0.5754) — after multiple-testing correction the selected config is statistically indistinguishable from best-of-N noise. The high probabilistic_sharpe (0.9867) is the classic PSR-vs-DSR trap (PSR ignores the 225-trial count DSR corrects for). The passing walk-forward (is_overfitted=false, avg OOS 1.31, windows [0.50, 1.28, 2.16]) and holdout (ratio 1.30) measure consistency, not significance, and cannot override the deflation failure. The edge is 2021-concentrated and outlier-carried (optimized 2021 +287%, return_kurtosis 467, skew 20, tail_ratio 0.028) — the headline rests on a handful of bull-market days. Not iterate (attempt 1 of 2): sensitivity is a flat plateau with 0 cliffs at a uniformly modest ~0.5-0.7 Sharpe, so the best-of-225 selected Sharpe sits below the luck bar with no robust region to tune toward — a second sweep re-selects another sub-significant config; the optimizer already had nothing but return-chasing to grab, breaching the frac<=1.0 design (position_frac 1.19). Not revise_hypothesis: the trend/TS-momentum mechanism is already validated on promoted BTC/ETH siblings and ETH is not a dead target — this is ETH's single-asset daily trend being too weak and too outlier-dependent to clear deflation (asset-selectivity), matching abandoned siblings EthDailyMultiWeekTrendContinuationLong (DSR 0.0) and EthDailyTrendFollowingLongATRTrail (DSR 0.78). FAILURE PATTERN: a clean, low-parameter, non-overfit single-asset ETH daily Donchian long/flat trend-follower with a passing walk-forward AND passing holdout still fails promotion because its genuine but modest, 2021-outlier-carried edge cannot clear the 225-trial luck bar (optimized Sharpe 0.751 vs 1.389, DSR 0.03, CI through 0, PBO 0.58). A non-overfit walk-forward and a favorable holdout ratio do NOT rescue an edge that fails deflated Sharpe; the PSR=0.99/DSR=0.03 split is the diagnostic tell.

LONG/FLAT-only daily Donchian-channel breakout trend strategy on ETHUSDT.BINANCE (USD-M perpetual), OHLCV-only, single-instrument/single-venue. Enters long when the daily close breaks above the prior 20-day Donchian high while price is above the 100-day SMA regime filter. Exits to FLAT on any of: a 3xATR stop from entry, a downtrend regime flip, or price losing the inner 10-day Donchian low. Position size is a single fixed fraction (95%) of account equity computed once at entry, leverage 1.0, notional hard-capped at equity. Never shorts and never re-sizes while a position is held.

Clean, low-parameter single-instrument OHLCV construction; no engine explosion — equity curve is smooth/plausible (100k -> ~1.07M peak -> 709k), no impossible single-bar jumps

31 long trades over 6.5 years is an appropriate sample for a daily 20-day Donchian breakout; entry_diag shows 31 signaled -> 31 submitted, 0 dropped

Trades faithfully implement the hypothesis: 0 shorts, LONG/FLAT only, entries are uptrend breakouts (regime + Donchian)

Genuine trend edge present: profit_factor 1.44, positive alpha 0.107, Sortino 3.74, Omega 1.95, metrics_reliable=true

Return is heavily concentrated in 2021 (+646%) with weak/negative other years (2022 -41%, 2025 -2.6%, 2026 -10.9%); skew 23, kurtosis 612, tail_ratio 0.015 => outlier-carried by one bull-market trade — high risk of failing deflated Sharpe/holdout post-optimization

Sharpe only 0.52 with sharpe_ci_low -1.09 (CI straddles 0); max_drawdown 46% with ~1406-day underwater duration

avg_position_pct 114.8% mildly exceeds the pre-registered ~100% guide — most likely MTM/equity drift during winning holds, but the optimizer should confirm notional never levers past equity (verify no base-pipeline over-exposure creeps in)

long/flat only, 0 shorts

long/flat

Clean, low-parameter construction; walk-forward not overfit (avg IS 1.49 -> avg OOS 1.31, all three OOS windows positive) and holdout passed (ratio 1.30)

No engine over-exposure explosion: avg_position_pct ~115% is benign MTM drift (identical in base at frac 0.95 and optimized), not leverage — the pre-registered defect detector did not meaningfully fire

Trades faithfully implement the hypothesis (17-31 long trades, 0 shorts, breakout entries)

Fails deflated Sharpe decisively: DSR 0.0299 (vs 0.95), optimized Sharpe 0.751 BELOW the 225-trial expected-max luck bar 1.389, is_significant=false, sharpe_ci_low -0.503 (CI through 0), PBO 0.5754

PSR 0.9867 vs DSR 0.0299 is the classic multiple-testing trap; passing walk-forward and holdout measure consistency, not significance, and cannot override the deflation failure

Edge is 2021-concentrated / outlier-carried (optimized 2021 +287%, kurtosis 467, skew 20, tail_ratio 0.028) — headline rests on a few bull-market days

Sensitivity is a flat plateau (0 cliffs) at ~0.5-0.7 Sharpe: no robust region above the luck bar to tune toward; optimizer breached the frac<=1.0 design (position_frac 1.19) chasing return

EthDailyDonchianBreakoutLongFlat was engineered as the maximally-evaluable construction — long/flat only, fixed single-entry sizing, leverage 1.0, pure OHLCV, single instrument — explicitly to sidestep the engine's confirmed over-exposure and aggregation defects, and it succeeded on that front: no explosion, a plausible equity curve, and 31 faithful long-only breakout trades. The mechanism showed a real trend edge (base return +609.8%, profit factor 1.44, positive alpha), and optimization pushed return to +1002.95% with profit factor 4.95 while the walk-forward stayed non-overfit and the holdout passed. But the edge was concentrated in the 2021 bull (kurtosis 467, skew 20, tail_ratio 0.028), so after multiple-testing correction it failed deflated Sharpe decisively — DSR 0.0299, optimized Sharpe 0.75 below the 1.389 luck bar, CI through zero, PBO 0.58 — and on attempt 1, with only a flat 0.5–0.7 Sharpe plateau to tune toward, the analyst abandoned it rather than iterate, matching the pattern of its abandoned ETH trend siblings.

A clean, low-parameter, non-overfit single-asset trend-follower with passing walk-forward and holdout can still fail promotion when its genuine but modest, outlier-carried edge cannot clear the multiple-testing luck bar — the PSR 0.99 / DSR 0.03 split is the diagnostic tell that consistency does not equal statistical significance.

The analyst abandoned it at the post-optimization ANALYZING stage on attempt 1: despite a non-overfit walk-forward (avg OOS 1.31, all three windows positive) and a passing holdout (ratio 1.30), it failed the deflated Sharpe gate decisively (DSR 0.0299 vs 0.95, optimized Sharpe 0.75 below the 225-trial luck bar of 1.389, is_significant=false, sharpe_ci_low -0.503, PBO 0.5754).

A long/flat-only, single-instrument daily Donchian-channel breakout trend-follower on ETHUSDT.BINANCE perp using pure OHLCV, fixed single-entry sizing at leverage 1.0 and no shorts or dynamic resizing, deliberately built as the cleanest construction the engine could evaluate to avoid its known over-exposure defects.

The base backtest returned +609.8% over 31 long trades (0 shorts) with profit factor 1.44, Sharpe 0.52 (CI low -1.09), Sortino 3.74, positive alpha 0.107 and a 46.3% max drawdown; optimization lifted total return to +1002.95% and profit factor to 4.95 but Sharpe only to 0.75, and the edge was heavily concentrated in 2021 (optimized 2021 +287%, kurtosis 467, skew 20).