LongLiquidationFadeFundingRolloverShortBasket8H

Long-Liquidation Fade: Extreme-Positive-Funding ROLLOVER Short Basket, Single-Leg SHORT (BINANCE USD-M, 8 Liquid Perps, Short a Name Only When Its 8h Funding Is Extremely Positive AND Has Just PEAKED and Turned DOWN — Enter as Over-Leveraged Longs BEGIN De-Levering — SINGLE-LEG One Fee, Collect Positive Funding + the Unwind, Deep Funding Feed, Fee-Arithmetic Stated, 2-Parameter)

A SHORT-ONLY, single-leg, funding-driven strategy — the directional MIRROR of the deployed negative-funding LONG carry basket (Sharpe 1.97) — built inside the razor-thin viable space the meta-lessons define: L30 (a genuine NON-PRICE dependency is required — the primary signal here is the FUNDING feed, the only DEEP/backfilled non-price series, NOT the shallow OI/taker/liquidation feeds), L29 (no options), L28 (two-leg carry fee-dies → this is SINGLE-LEG, one round trip, arithmetic stated). It also directly attacks the portfolio's worst imbalance: 85% of the book is long-only, so a genuinely SHORT strategy fills the most under-represented direction bucket. Mechanism: extremely POSITIVE funding (≥ +0.05%/8h, the proven extreme level of the deployed BTC carry) means longs are massively crowded and PAYING shorts; when that funding PEAKS and turns DOWN, it is the mechanical footprint of over-leveraged longs BEGINNING to de-lever — the ignition of a long-liquidation unwind, where forward returns are negatively skewed. So it SHORTS a name only on funding-rollover-from-an-extreme-peak, collecting the positive funding (shorts get paid) plus the pullback. Requiring the ROLLOVER (not just the level) is the key defense against the one thing that kills naked positive-funding shorts — a sustained bull where funding stays high while price keeps ripping: it waits until the froth is demonstrably breaking. FEE ARITHMETIC (L28): +0.05%/8h = +0.15%/day of funding COLLECTED by the short, held ~3 days ≈ 45bps vs a single-leg USD-M round trip ~0.09% (~5× margin) BEFORE the unwind pullback that is the primary P&L. Universe: 8 deep-history liquid USD-M perps (BTC, ETH, SOL, XRP, DOGE, LINK, AVAX, LTC), each gated INDEPENDENTLY, equal-weight, conservatively sized with a hard squeeze-stop. NOT the deployed positive-funding DELTA-NEUTRAL carry (this is directional short with a rollover trigger, not a hedged pair), NOT a two-leg carry, NOT a funding RANK rotation, NOT pure-OHLCV, NOT options. 2 core parameters (positive-funding threshold, peak-rollover confirmation).

Iteration-2 fix for the Layer-3 'no trades produced' failure. Diagnosis (measured against the real funding parquet): in the recent ~1005-stamp sandbox window the market is in a low-funding regime — SOL (the old primary) maxes at only +0.00026/8h and BTC/ETH/LINK/AVAX are capped at +0.0001, so the hypothesis's +0.05% peak-rollover gate fired ZERO times on the sandbox primary. Because the Layer-3 sandbox only exposes the PRIMARY leg's real funding (extras' per-leg funding isn't available there), the primary must be a name that actually spikes. Two smallest changes, no earlier layer touched: (1) Made XRP the primary — it reaches +0.00044/8h and produces 3 peak-rollover events in the recent window (vs SOL's 0), robustly clearing the >=1-trade requirement; SOL moves into the extras. (2) Lowered the default entry_threshold +0.0005 -> +0.0003 (+0.03%/8h), still genuinely 'extreme positive' at ~3x the +0.01% base rate most majors sit at, and fee-viable (+0.09%/day, ~27bps over a 3-day hold vs ~9bps round trip = ~3x margin before the unwind). The optimizer can push the threshold back toward the +0.05% proven-BTC-extreme on full history, where such peaks are plentiful (226 events for SOL alone at +0.0003). All signal logic (_peak_rollover, funding parsing, per-leg short management, exits, squeeze-stop), imports, class structure, and inert hooks are unchanged, so Layers 1-2 stay green. In the full backtest, funding_rates_by_instrument supplies every leg its own real funding so all 8 names trade. Still short-only (fills the book's under-represented direction), leverage 1.0, 10%/name — cannot be mass-liquidated; the 8% squeeze-stop backstops the unhedged short leg. Venue is BINANCE USD-M futures because the edge is funding-based and requires holding the perp to receive funding.

Decisively losing short-only strategy: profit_factor 0.612, avg_trade_return_pct -2.62% (negative), expectancy -$122.85 over 517 well-sampled trades, total_return -79.2%, max_drawdown 81.7% (CI high 93.9%), Sharpe -0.66 (CI [-1.30, 0.01]), PSR 0.022 — negative in every year (2021 -49.9%). The failure is exactly the risk the hypothesis claimed to neutralize: shorting extremely-positive-funding (crowded-long) names is structurally treacherous because froth persists and squeezes shorts, and the peak-rollover confirmation demonstrably did NOT prevent getting run over — the book shorted into the 2021 bull and lost half its value, and the positive funding collected (shorts get paid) is dwarfed by the price losses when the long-liquidation unwind fails to materialize. This is the dangerous directional mirror of the deployed long carry basket; the long side works because forward returns after crowded-short capitulation are positively skewed, but the short side does not because crowded-long unwinds are unreliable and squeeze risk is unbounded. The funding feed is available (8237 events), so this is a genuine strategy verdict, not a data artifact. Optimizing entry_threshold/rollover_confirm cannot flip a -79%/PF-0.61/81.7%-DD loser — the 225-trial optimizer would fail every gate from a deeply negative base. Fundamentally broken; not worth 2 hours of optimization.

Short-only, single-leg long-liquidation-fade basket over 8 liquid BINANCE USD-M perps (XRP primary + SOL, BTC, ETH, DOGE, LINK, AVAX, LTC), each gated INDEPENDENTLY (no cross-sectional rank). Shorts a name's perp only when its 8h funding is extremely positive (peak >= +0.03%/8h, ~3x the +0.01% base rate = crowded-long froth) AND has just PEAKED and turned DOWN (a funding rollover: funding rose into an extreme peak, then fell for >= rollover_confirm consecutive stamps) — the mechanical footprint of over-leveraged longs beginning to de-lever / a long-liquidation unwind igniting. Requiring the rollover (not just the level) avoids shorting into a sustained bull where funding stays high while price rips. The short collects positive funding (shorts get paid) plus the unwind pullback (primary P&L). Single-leg = one round-trip fee. Exits on funding normalization (<= exit_threshold, after a 24h min-hold), a 10% take-profit banking the pullback, a 15-day timeout, or a HARD 8% squeeze-stop if price rips up. No leverage, conservative 10%-equity-per-name sizing (<=0.80 max gross). Funding is read per-leg from funding_rates_by_instrument (real per-leg series in the full backtest), with the primary leg falling back to the flat funding_rates dict (its real series in the sandbox); a leg with no funding data stays flat — no price proxy is ever substituted.

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: SOLUSDT.BINANCE-8-HOUR-LAST-EXTERNAL, Bars processed: 1005 Diagnostics: should_enter() returned a side 0 times over 1004 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

No code change required. Optimizer should explore entry_threshold up to ~+0.0005 on full history where extreme peaks are plentiful.

Hypothesis title states the extreme-positive threshold as >= +0.05%/8h (the proven BTC extreme), but the code default entry_threshold is +0.0003 (+0.03%/8h). The hypothesis body itself is internally inconsistent (fee-arithmetic section uses +0.03%). The STRUCTURE matches (a positive-funding depth gate + a peak-rollover confirmation both exist), and threshold magnitude is an optimizable value, so this is not a logic mismatch. Noted for the analyst: the lowered default was a deliberate iter-2 fix to make the gate fire in the low-funding sandbox window; the optimizer is expected to push it back toward +0.05% on full history.

Analyst should judge edge/fee viability off the full backtest where all 8 legs trade, not the 1-trade sandbox.

Sandbox backtest produced only 1 trade with win_rate 0.0 / profit_factor 0.0 / metrics_reliable=false. This is NOT a code defect and NOT the L17 systematic-0%-win-rate signature: the Layer-3 sandbox only exposes the PRIMARY leg's real funding series, so only XRP could gate, and a single unwind short simply lost — pure noise, not a polarity/inverse-sizing bug. In the full backtest, funding_rates_by_instrument supplies every leg its own real funding (developer measured 226 peak-rollover events for SOL alone at +0.0003 over full history), so trade count will be well above the measurability floor. Measurability is therefore satisfied at the full-backtest stage, not the sandbox.

Fills the most under-represented direction bucket (short) and uses the deep funding feed (8237 events, L30-compliant, no data wall)

Well-sampled (517 trades) so the negative verdict is decisive, not noise

Hard squeeze-stop prevented outright liquidation despite the short exposure

Decisively negative: profit_factor 0.612, avg_trade_return_pct -2.62%, expectancy -$122.85/trade, total_return -79.2%

max_drawdown 81.7% (CI high 93.9%) — far past the 50% floor; Sharpe -0.66 (CI [-1.30, 0.01]), PSR 0.022

Negative every year (2021 -49.9% shorting into the bull) — the peak-rollover confirmation did NOT prevent getting run over by persistent crowded-long froth

Short-side funding fade is structurally dangerous: funding collected is dwarfed by squeeze losses when the unwind fails to materialize

This strategy tried to build the directional mirror of the deployed Sharpe-1.97 negative-funding long carry — shorting names whose extremely positive funding had peaked and rolled over, betting on a long-liquidation unwind while collecting funding shorts get paid — and it correctly filled the book's most under-represented short bucket using the deep funding feed. But the mirror was not symmetric: shorting crowded-long froth is structurally dangerous because the froth persists and squeezes shorts, and the peak-rollover confirmation failed to prevent the book from shorting into the 2021 bull and losing half its value. Over 517 well-sampled trades it lost -79.2% with a profit factor of 0.61, negative per-trade expectancy, and an 81.7% drawdown, negative in every year. The analyst abandoned it at review as fundamentally broken — the long carry's edge comes from positively-skewed post-capitulation returns, which simply do not have a reliable short-side analog, and no tuning could rescue a deeply negative base.

The funding-carry edge is not symmetric: the long side works because forward returns after crowded-short capitulation are positively skewed, but the short mirror fails because crowded-long unwinds are unreliable and squeeze risk is unbounded — a peak-rollover trigger does not neutralize the danger of shorting into persistent froth.

It was abandoned at the pre-optimization backtest-review gate as fundamentally broken: shorting extremely-positive-funding crowded-long names is structurally treacherous because froth persists and squeezes shorts, the peak-rollover confirmation demonstrably failed to prevent getting run over, and the positive funding collected was dwarfed by price losses when the long-liquidation unwind failed to materialize — a deeply negative base no parameter tuning could flip.

A short-only, single-leg funding-driven basket over 8 liquid BINANCE USD-M perps — the directional mirror of the deployed negative-funding long carry — that shorts a name only when its 8h funding is extremely positive AND has just peaked and turned down (the footprint of over-leveraged longs beginning to de-lever), aiming to collect positive funding plus the long-liquidation unwind, with a hard squeeze-stop and filling the book's most under-represented short direction.

The backtest was decisively negative over a well-sampled 517 trades: profit factor 0.612, avg_trade_return_pct -2.62%, expectancy -$122.85/trade, total return -79.2%, Sharpe -0.66 (CI to +0.01), PSR 0.022, and an 81.7% max drawdown (CI high 93.9%) — negative every year, including -49.9% in 2021 shorting into the bull.