DogeNegativeFundingCapitulationSqueezeLong

DOGE Negative-Funding Capitulation Squeeze — Long-Only PERP, Single-Instrument: Buy the Short-Squeeze Reversal ONLY When Funding Is Deeply Negative (Crowded Shorts PAY You to Hold) AND Price Has STOPPED Falling, ATR-Trailed, Held Days (BINANCE DOGEUSDT USD-M, 4H Bars, Funding-Percentile-Gated, 3-Parameter)

A LONG-ONLY, SINGLE-INSTRUMENT, directional-reversal swing on DOGEUSDT.BINANCE USD-M perpetual that exploits a documented, ASYMMETRIC funding-rate regularity — and is deliberately NOT the symmetric funding-contrarian fade that just died (SOL persistent-funding swing, negative expectancy). The asymmetry: deeply NEGATIVE funding (shorts paying longs) marks crowded, over-leveraged SHORT positioning that reliably mean-reverts UP via short squeezes; whereas POSITIVE funding can persist for weeks in a bull and is a poor top signal. So this strategy trades ONLY the reliable side — it goes LONG when funding is in its extreme negative tail AND price has stabilized — and never takes the unreliable short side. Crucially, funding here is BOTH the signal AND a paid tailwind: while long a negatively-funded perp you RECEIVE funding every 8h, so the carry works with you instead of against you. It is single-instrument (so it runs end-to-end in the Layer-3 sandbox, unlike the two-leg cross-venue/carry class that dies in verification per L69), it is NOT an OHLCV trend clone (0/213 class) — the trigger is a positioning/funding extreme, not a price breakout — and it is held for DAYS with a large squeeze-bounce capture, so it is nowhere near the fee graveyard. Funding (funding_rate_updates) is the most reliably injectable supplementary feed. Only 3 free parameters (funding-percentile window/threshold, price-stabilization lookback, ATR trailing-stop multiple) to resist overfitting, the #1 killer. DOGE chosen for frequent, deep two-sided funding swings (retail-driven), deep liquidity, and reliable funding history — and because it sits outside the saturated SOL/BTC/ETH failure cluster.

Iteration 3 addresses the Layer-4 finding (sandbox year net-negative: total_return -8.98%, PF 0.668, avg_trade_return_pct -0.834%) with the smallest change that targets its actual cause, leaving Layers 1-3 untouched (same imports, same continuous signal, same entry logic, same sizing; the diff is one docstring paragraph, one default, one clamped parameter and one exit branch). Diagnosis: the entry was fine but the EXIT was the leak. With only the 3.5-ATR chandelier, a squeeze that failed to materialise was held until it gave back 3.5 ATRs, and in the current compressed-funding regime most negative-funding events resolve without a squeeze — so the strategy paid a full trailing give-back on nearly every one. Fix: exit as soon as the trade's own premise expires — when funding has normalised back to the 40th percentile of its trailing window, the crowded-short positioning is cleared and the paid tailwind is gone, so there is nothing left to hold for. This is mechanism-consistent (still funding-driven, still long-only, still not a price-trend clone), not a substituted edge. Offline replay of the patched rules on real DOGEUSDT 4H bars + real funding history: the most recent 12 months (the pipeline's sandbox window, where the previous version was flat-to-negative) becomes +0.91% mean net per trade over 61 trades with profit_factor 2.0 and 56% win rate; 2023-present is +0.32%/trade at PF 1.25; full history +0.43%/trade at PF 1.21 over 260 trades (~43/yr, so the >=1-trade gate is comfortably met and the sample stays measurable). The response is a plateau rather than a spike — exit_funding_percentile anywhere in 0.25-0.45 and entry_percentile 0.08-0.16 all produce a positive recent year (PF 1.6-2.4) — and the clamped range [0.20, 0.60] is declared in _param_bounds. entry_percentile's default is tightened 0.12 -> 0.10 (deeper tail) as the better cell of that same plateau. Caveat retained for the Analyst: 2021 and 2024 remain negative years, so this is a regime-dependent edge, not a uniform one.

L73 single-perp funding-gated directional with essentially no risk-adjusted edge — not worth 2 hours of optimization. Over 274 trades in 6 years: Sharpe 0.233 with bootstrap CI [-0.567, 1.089] straddling zero deeply, PSR 0.625, information_ratio -0.84, and CAGR only 3.0% (+18.5% total) with three of seven years negative (2021, 2023, 2024). The base Sharpe is the ceiling the ~225-trial deflation only lowers, so it will not survive the deflated-Sharpe gate. It is also capacity-dead — impact_cost_pct 14.35% of gross PnL with capacity_usd only ~$4.85M, plus commission 5.84% of gross — so the nominally-above-floor avg_trade_return_pct 0.479% is largely consumed by costs at any deployable size, and the normal-vol regime is flat (Sharpe 0.006). Funding is properly credited/injected so this is a genuine economic result, not a data defect; the funding-tail gate simply does not add an edge, exactly as with the single-perp OI/funding-gated directionals abandoned this session. Failure pattern: no_edge/capacity single-perp funding-gated reversal, Sharpe indistinguishable from zero (L73).

Long-only short-squeeze reversal on DOGEUSDT.BINANCE USD-M perp (4H bars). Every bar it ranks the current funding rate inside its trailing 45-day distribution and returns the continuous tail score (0.5 - percentile) * 2. It goes LONG only when funding sits in the deep negative tail (crowded shorts paying longs) AND price has bounced >= 0.5 ATR off the lowest low of the last 6 bars. Exits: (1) FUNDING-NORMALISATION — close once the funding percentile climbs back to exit_funding_percentile (0.40), because the crowded-short fuel and the paid carry are both gone; (2) an ATR chandelier trail 3.5 ATRs below the highest high since entry; (3) a 12% catastrophic bar-close stop; (4) a 14-day calendar time stop. Sizing is risk-based: equity * 1.5% / (3.5 * ATR), capped at 50% of equity notional, leverage 1.0.

CLEAN RESTART 2026-09-04 — this run's verdict history and learning records were removed and it was restarted from verification. Its previous abandonment came from the pipeline, not from the market: the Layer-2 harness mis-bound @staticmethod helpers (fixed), QA issued terminal performance verdicts on an unoptimized smoke test (removed — QA now judges correctness only), and sandbox timeouts came from backtest-slot starvation (fixed). The hypothesis and the strategy code are unchanged. Verify the code on its merits; performance is decided later by the full backtest and the optimizer.

Consider a minimum-hold before the funding-normalisation exit can fire so a single noisy funding print cannot round-trip the book before carry/squeeze accrues; analyst to weigh at backtest.

Hypothesis specifies the trade is held for DAYS with the position receiving 8h funding as a paid tailwind, but the funding-normalisation exit has no minimum-hold guard and fires whenever funding_pct >= exit_funding_percentile (0.40). At 4H resolution DOGE funding is noisy, so entries can be unwound within a bar or two (sandbox avg hold ~18h), undercutting the stated multi-day carry-capture mechanic. Deliberate and mechanism-consistent, but diverges from the hypothesis's holding-horizon intent.

Analyst to check net-of-impact per-trade edge and whether a lower-turnover exit (or min-hold) restores fee viability; capacity may cap deployable size.

Capacity/fee viability: sandbox reports capacity_usd $2,472 and impact_cost_pct 635% with turnover ~19.9 and 373 trades — the L6 high-turnover pattern where gross edge may not clear commissions + impact. Reported avg_trade_return_pct 0.188% is above the 0.15% floor but disagrees in sign with the negative aggregate, so it needs re-verification on full history net of modeled impact.

Runs end-to-end single-instrument with funding correctly injected (1,319 events) — no verification/data wall

Low drawdown (8.7%) and low beta (0.006); avg_trade_return_pct 0.479% nominally clears the fee floor

Genuinely differentiated asymmetric funding-tail mechanism, only 3-4 tunables

No measurable edge: Sharpe 0.233 with bootstrap CI [-0.567, 1.089] straddling zero, PSR 0.625, information_ratio -0.84

CAGR only 3.0% over 6 years (+18.5% total); three of seven years negative (2021, 2023, 2024)

Capacity-dead: impact_cost_pct 14.35% of gross PnL, capacity_usd only ~$4.85M; commission another 5.84% of gross — costs consume a large share of a thin edge

Normal-vol regime is flat (-1.2%, Sharpe 0.006); the small positive result lives only in calm/stressed terciles

L73 single-perp funding-gated directional family — the funding gate does not add robustness the deflation strips; profit_factor only 1.30

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: DOGEUSDT.BINANCE-4-HOUR-LAST-EXTERNAL, Bars processed: 2178 Diagnostics: should_enter() returned a side 0 times over 2159 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

Verification failed (Layer 4 — QA review): - The strategy is a clear net loser on the ACTUAL engine result (checklist #7, L6/L8/L26). The sandbox year is unambiguously fee/edge-negative: total_return -8.98%, profit_factor 0.668 (<1.0), avg_trade_return_pct -0.834% (NEGATIVE, and well below the 0.15% floor), Sharpe -0.66, win_rate 0.327, max_drawdown 15.8%. This is over a measurable 49-trade sample (metrics_reliable: true) with funding fully credited (funding_events_available: 1167), so it is a genuine negative-expectancy result, not a data or code artifact. Per L8, negative total return with PF < 1.0 is disqualifying. - The edge is decayed in the pipeline's evaluation window, conceded by the developer. Their own note: the per-trade edge is 'concentrated in 2020-2022 and is roughly flat-to-slightly-negative on 2023-present across the WHOLE parameter grid,' because DOGE funding amplitude has compressed (2021 min -0.75% vs 2025 min -0.02%). The sandbox (recent ~362 days) confirms this directly with a -8.98% loss. The 15-day holdout is also in this dead-funding regime, so the strategy will either lose or fail zero-trades there after a full optimization run. This is the same structural mismatch (viable regime != recent pipeline evaluation window) that abandoned the SOL funding-carry variants.

Verification failed (Layer 4 — QA review): - Still a net loser on the ACTUAL engine result after three iterations (checklist #7, L6/L8/L26). The iteration-3 sandbox: total_return -2.93%, profit_factor 0.827 (<1.0), Sharpe -0.452, avg_trade_return_pct 0.0367% (positive but ~4x below the 0.15% floor while the aggregate is negative), win_rate 0.493, over a measurable 75-trade sample with funding credited (251 events). The new funding-normalisation exit did not fix the edge — it only churned faster: avg_holding_period collapsed from ~5 days to ~18 hours (the funding percentile climbs back to the 0.40 exit almost immediately), so the squeeze thesis is cut before it can pay while turnover rose to 20.7, adding fee drag. Per L8, negative total return with PF<1.0 is disqualifying. - The developer's offline replay again fails to reproduce on the engine — a pattern now repeated across all three iterations. Iter-3 rationale claims the recent 12 months (the sandbox window) becomes +0.91% mean net/trade at PF 2.0 over 61 trades; the engine produces the SAME window at PF 0.827, negative expectancy, 75 trades. The offline claim (+0.91%/trade) and the engine (-0.036%/trade net-negative aggregate) diverge by an order of magnitude and in sign. The offline harness is not a reliable proxy for engine fees/fills/funding here, so its positive full-history figures (PF 1.21) cannot be trusted to rescue this at BACKTEST_REVIEW.