BtcPerpLiquidationCascadeContinuationLS1H

BTC Liquidation-Cascade Momentum Continuation, Long-Short (Single-Instrument BTCUSDT.BINANCE Perp — NEW Information Source, CONTINUATION Polarity: Trade WITH a Large Forced-Liquidation Cascade for Short-Term Follow-Through, ATR-Trail + Time Stop, 3-Parameter)

A LONG-SHORT, EVENT-DRIVEN MOMENTUM strategy on the BTCUSDT.BINANCE USD-M perpetual that trades WITH a large forced-liquidation cascade for short-term continuation — a new-information-source (liquidations) idea in the CONTINUATION polarity, per L38, and aligned with the ONE edge that survives in this factory (momentum/trend), unlike every BTC fade this session (including my own liquidation-REVERSAL, which was tested and lost net-negative). This is the deliberate OPPOSITE polarity of that dead fade: instead of fading the overshoot, it rides it. Mechanism: when the liquidation feed shows a large forced-flow burst (z-score spike) in one direction, the exchange's market liquidations plus triggered stop-cascades and momentum-chasers EXTEND the move for a short window before it exhausts — a documented short-term ignition/continuation effect. It goes LONG on a short-liquidation squeeze (forced buying) that prints an up-bar, SHORT on a long-liquidation dump (forced selling) that prints a down-bar, and rides the follow-through with an ATR trail and a short time stop (continuation is brief). It uses WELL-COVERED Binance liquidation data (unlike the taker/L-S coverage ceilings and the options no-trades/no-edge failures), trades the liquid BTC perp (dev-safe, signal=execution, and the exact SOL/BTC liquidation mechanic that was whole-strategy-swapped before must be preserved), single instrument, long-short (fills the 13.4% bucket vs 86.6% long-only), 1H cadence for a measurable sample. 3 parameters.

Iteration 2 makes exactly the one change the Layer-4 review's diagnosis implies, and touches nothing that already passed. The review did not find a coding defect — it found that on 1-HOUR bars the strategy can only act at the close of the bar in which the cascade printed, i.e. up to 60 minutes after the forced flow, 'by which point the ignition is usually already exhausting', producing an inverted RR and a -7.3 return skew. A liquidation cascade is a minutes-scale ignition, so the cadence was mis-specified, not the polarity. I therefore moved the identical mechanism onto 5-MINUTE bars (max entry latency 60 min -> 5 min) and rescaled every bar-counted window so each represents the same wall-clock horizon or better: ignition/time stop 12 bars = 1 hour, standardization window 576 bars = 2 days, ATR 60 bars = 5 hours. The ATR trail widens to 3.0 ATRs because a 5-minute ATR on BTC is ~0.10-0.20% of price — at 1.5 ATRs a completed trade would not clear the ~0.10% round-trip taker cost, whereas 3 ATRs targets a 0.3-0.6% move. The notional cap tightens 0.35 -> 0.30 to trim the largest-loss tail the review flagged. Signal construction, the BUY/SHORT vs SELL/LONG side mapping, the confirmation filter, cross-only entry, the exit structure, sizing formula, imports and class structure are byte-for-byte unchanged, so Layers 1-3 stay green. One caveat, stated once as required: if this cadence change does not flip the sign, I agree with the reviewer that the liquidation-cascade mechanism is falsified in both polarities and should be abandoned rather than re-tuned again.

Decisively net-negative and coverage-crippled. profit_factor 0.306 (gross losses 3x gross wins), avg_trade_return_pct -0.058% (negative), expectancy -$60.6/trade, total_return -7.9%, Sharpe -1.147 with CI [-1.50, -0.87] entirely below zero, win_rate 26.7%, return_skew -20.7 / kurtosis 560 (a few cascade losses dominate). Because the liquidation feed is live-stream-only with no historical archive, every one of the 131 trades falls in a single ~3-month window (2026-04 to 2026-07; annual_returns has only 2026), so this is a tiny single-regime sample on top of a catastrophic loss. This is iteration 2: the developer already moved 1H→5m to fix 'entry latency' and profit_factor barely changed (0.28→0.31), which disproves the latency diagnosis — the continuation polarity has no tradeable follow-through edge net of costs, exactly like the liquidation-reversal sibling it was meant to invert (L18 liquidation fee_edge/no_edge family). There is no parameter region to optimize toward when PF is 0.31 and per-trade expectancy is negative over the only data the feed covers; tuning entry_z/trail/hold would overfit 131 loss-dominated trades. Abandon per the PF < 1.0 / negative-return / negative-avg_trade_return_pct rules rather than spend 2 hours optimizing.

Long/short BTCUSDT.BINANCE USD-M perpetual on 5-MINUTE bars, driven by the Binance forced-liquidation tape. Per-bar net forced flow (short-liquidation USD minus long-liquidation USD) is z-scored against a trailing 576-bar (2-day) window with the current bar excluded; the bar's own body (in ATR units) must agree in sign with the flow or the reading is damped and clipped below the entry threshold. A fresh cross above +entry_z goes LONG (short squeeze / forced buying), below -entry_z goes SHORT (long-liquidation dump / forced selling). Exits are a 3.0-ATR trailing stop from the best close since entry (detected intrabar, booked at the close) and a 12-bar (1-hour) time stop measured from the position's ts_opened. Sizing is gap-aware risk sizing capped at 0.30x equity notional. No price-only fallback: if the tape is missing or does not cover the window, no position is taken.

Verification failed (Layer 4 — QA review): - DECISIVELY NET-NEGATIVE -- the continuation polarity loses WORSE than the fade it was built to invert, so the liquidation-cascade mechanism has no edge in EITHER direction. Over 32 trades on the ~100-day covered window: total_return -2.74%, profit_factor 0.280, Sharpe -1.94 (CI [-2.86, -0.67] entirely below zero), win_rate 0.281, avg_trade_return_pct -0.120% (negative, below the fee floor), inverted RR (avg_win $118 vs avg_loss $165), largest_loss -$516 vs largest_win $270, return_skew -7.26, kurtosis 64.5, probabilistic_sharpe 0.0. The catastrophic left tail is the mechanism's real problem, not a bug: on 1H bars the strategy can only enter AFTER the cascade prints (at the bar close), by which point the ignition is usually already exhausting -- so riding the cascade buys the top of the squeeze and is reversed. The fade of this same feed lost (PF 0.465-0.544 across its iterations); the continuation loses harder (PF 0.280). Both polarities are now falsified.

Decisively falsified edge (PF 0.31, Sharpe -2.97, 131 trades) — verified NOT a polarity bug; abandon at backtest_review

Liquidation feed is live-stream-only; covered 131-trade sample is already sufficient to disconfirm

Careful implementation with correct side-mapping and a coverage guard that honestly refuses to trade when the tape is absent

profit_factor 0.306 (gross losses 3x gross wins) and total_return -7.9%

avg_trade_return_pct -0.058% (negative), expectancy -$60.6/trade, win_rate 26.7%

Sharpe -1.147 with CI [-1.50, -0.87] entirely below zero; probabilistic_sharpe 0.0; return_skew -20.7, kurtosis 560

Liquidation tape is live-only: ALL 131 trades fall in 2026-04 to 2026-07 (~3 months, one regime) — annual_returns has only 2026 — not a multi-year sample

Iteration 2 (1H→5m latency fix) barely moved PF (0.28→0.31): the continuation polarity is as net-negative as the dead reversal sibling — the fix addressed the wrong thing

BtcPerpLiquidationCascadeContinuationLS1H tried to ride large forced-liquidation cascades for brief momentum follow-through — the exact opposite polarity of a liquidation-reversal fade that had already been tested and lost. Iteration 2 moved the whole mechanism from 1H to 5-minute bars on the theory that entry latency (buying at the top of an already-exhausted squeeze) was the problem, but profit factor barely budged from 0.28 to 0.31, disproving that diagnosis. The result stayed catastrophic: profit factor 0.306, negative per-trade expectancy, a 26.7% win rate, and a Sharpe of -1.147 whose entire confidence interval sat below zero — with all 131 trades trapped in a single ~3-month window because the liquidation tape has no historical archive. The reviewer abandoned it at backtest-review under the PF<1.0 / negative-return rules, concluding both the continuation and reversal polarities of the liquidation mechanic lack any tradeable edge net of costs, and that a live-only feed leaves nothing to optimize toward.

Inverting a dead fade's polarity does not manufacture an edge — the liquidation-cascade continuation lost just as decisively as the reversal sibling it was meant to invert, and a live-stream-only feed with no historical archive can't produce a multi-year sample, so the mechanism was both edgeless and untestable across regimes.

It was abandoned at the pre-optimization BACKTEST_REVIEW gate as net-negative and coverage-crippled: profit factor 0.306 with negative per-trade expectancy over a tiny single-regime sample, and — decisively — the iteration-2 latency fix (moving 1H→5m) barely moved profit factor (0.28→0.31), disproving the latency diagnosis and showing the continuation polarity has no tradeable follow-through edge net of costs.

Trade the BTCUSDT.BINANCE perp long/short WITH a large forced-liquidation cascade for short-term momentum follow-through — going long on a short-liquidation squeeze that prints an up-bar and short on a long-liquidation dump that prints a down-bar — the deliberate opposite polarity of a dead liquidation-reversal fade, on 5-minute bars with an ATR trail and time stop.

It was decisively net-negative: 131 trades (67 long / 64 short), profit factor 0.306 (gross losses ~3x gross wins), total return -7.9%, avg per-trade return -0.058%, expectancy -$60.6/trade, win rate 26.7%, Sharpe -1.147 (CI [-1.50, -0.87] entirely below zero), with extreme tails (skew -20.7, kurtosis 560). Because the liquidation tape is live-only, all 131 trades fell in a single ~3-month window (2026-04 to 2026-07).