Spot-Confirmed Shock Follow-Through: trade Hyperliquid alt perps long/short after a >=1.5-sigma 24h move that Binance SPOT confirms (same sign, spot volume surge); fixed 48h hold, intrabar stop (6 alts, 4H, cross-venue)

Cross-venue, long/short event strategy on 4-HOUR bars. Traded legs: the Hyperliquid perps SUIUSD, APTUSD, INJUSD, TIAUSD, SEIUSD and OPUSD. Signal-only legs: the same coins' Binance spot pairs (xxxUSDT.BINANCE_SPOT), which are never traded. Mechanism, building on distilled lesson L203 (spot-confirmed perp follow-through survived twice): a sharp move with heavy cash-spot participation reflects unlevered demand or supply, which tends to persist. A move on the perp alone is leveraged positioning, which tends to fade. Alignment: a decision at timestamp T is made only after both the HL perp bar and the Binance spot bar closing at T have arrived; if either is missing, that coin is skipped for that bar. Execution: a market order at the signal bar's close, which is the platform's faithful equivalent of a next-bar-open fill on these 24/7 venues. Stops are intrabar protective stops (reduce-only stop-market), not exchange-resting orders. The categorical parameter require_spot_confirmation gives the matched test without spot confirmation on the same shock events. The universe has no prior runs in the corpus, and every coin has >=2y of HL 4H history plus a Binance spot listing over that span. Fee arithmetic: the all-in round trip is ~0.13% (0.09% HL fees plus ~0.04% impact). A 1.5-sigma 24h shock on these alts (daily vol 4.5-6%) is about a 7-9% move. If 10-15% of it continues, that is a 0.8-1.2% gross mean and about 0.5% net per trade, roughly 4x the cost. Expected firing rate: ~25-40 trades per coin-year, about 300-400 trades in total.

negative_expectancy: The default operationalization of spot-confirmed shock follow-through on six Hyperliquid alt perps (SUI/APT/INJ/TIA/SEI/OP, 4H) loses money on a reliable sample. Over 444 closed legs from 2024-03 to 2026-10, avg_trade_return_pct is -0.925% against the +0.15% Hyperliquid floor. PF is 0.679, Sharpe -1.41 (95% CI -2.71 to -0.20), total return -29.9%, and max drawdown 33.9% against the 20% contract constraint. The loss holds in all three vol regimes and on both long and short legs. 34% of legs exit at the 8% stop, and the signal IC is -0.005 (t=-0.08). The per-trade loss is about 7x the round-trip cost: no plausible setting of shock_z, shock_bars, hold_bars or spot_volume_ratio can bridge that gap, and a negative-IC signal gives the optimizer nothing to tune toward. QA concern borne out on the confirmed arm. The require_spot_confirmation=False control was not run in this backtest, so this review does not show whether spot confirmation adds information. Scope: this rejects only this six-alt 4H follow-through operationalization. It does not reject L203 spot-confirmed follow-through on its original instruments. The apparent mean-reversion is a symptom, not a verified edge; per L193 it should not be flipped on the basis of this run.

Cross-venue long/short event strategy on 4H bars. It trades the Hyperliquid perps for SUI, APT, INJ, TIA, SEI and OP. The same coins' Binance spot USDT pairs are signal-only legs and are never traded. For each coin, a decision at bar-close T is made only after both the HL perp bar and the Binance spot bar closing at T have arrived; otherwise that coin is skipped for T. Shock z = the sum of the last shock_bars perp log returns (default 6 = 24h), divided by the std of the last vol_lookback_bars (180) one-bar returns times sqrt(shock_bars). When |z| >= shock_z, the strategy follows the move (long if up, short if down) if Binance spot confirms it. Confirmation means two things: the spot log return over the same window has the same sign, and mean spot volume over the window is at least spot_volume_ratio times the per-bar mean of the preceding 180 bars. With require_spot_confirmation=False, the same shock events are traded without the spot check (the matched control). Execution is a market order at the signal bar's close; on these 24/7 venues that is the platform's faithful equivalent of a next-bar-open fill. Each position has two exits: a fixed time exit after hold_bars bars (default 12 = 48h) counted from the fill, and an intrabar reduce-only stop-market at stop_pct=8% from the average entry, placed when the position opens. After a coin's position closes, that coin cannot be re-entered for cooldown_bars=6 bars. Sizing: each leg is leg_notional_pct=20% of equity x leverage (1.0), with at most max_concurrent_legs=4 legs open, so gross exposure stays at or below 80% of equity. The loss at the stop is 8% x 20% = 1.6% of equity, which satisfies risk route (a).

This implements the hypothesis as written. It detects the shock on the traded HL perp, takes confirmation from unlevered Binance spot participation (a same-sign move plus a volume surge), and trades the continuation with a fixed 48h hold and an intrabar 8% protective stop. Every optimisation-plan parameter name, fixed value and tunable (shock_z, hold_bars, shock_bars, spot_volume_ratio, require_spot_confirmation) is used verbatim, and none is clamped. Engineering notes: (1) Every coin, including the primary SUI perp, goes through the same _process_coin routine and enters via _submit_entry_instrument. So should_enter returns None, should_exit returns False, and calculate_signal returns SUI's continuous shock z. (2) Each coin's alignment is checked explicitly (perp bar ts == spot bar ts == T), and each T is processed once per coin. The base template's cross-leg barrier also holds the primary path until all 12 legs are aligned. So in the rare case where the SUI perp bar arrives last while some other coin's leg is missing at T, SUI is skipped for that T. (3) Indicator buffers advance only on aligned (perp, spot) bar pairs, so a missing leg bar makes the next return span two bars. (4) Per-coin state (last processed ts, last exit ts for the cooldown, latest z) is kept as scalar attributes, and the hold is anchored to position.ts_opened, so a restart changes no decision. (5) The volume baseline excludes the shock window itself. The usable history starts around January 2024: Binance spot listed SEI and TIA in Q3-Q4 2023, and the HL catalog starts about January 2024.

The unoptimised smoke run (default shock_z 1.5, shock_bars 6, hold 12, spot_volume_ratio 1.5, confirmation on) lost money across a usable sample: 444 closed legs over 945 days, avg_trade_return_pct -0.91% against the +0.15% floor (~0.13% all-in round trip), PF 0.68, win rate 40%, Sharpe -1.39 and max DD 33.8% against the contract's 20% constraint. Longs and shorts lose about equally, and the loss shows up in all three vol regimes. This looks more like 24h shocks on these alts mean-reverting than continuing. The 444 legs fall within the hypothesis's expected 300-400 range, so this is not a firing-rate artefact. What the analyst should check first on the full backtest: (1) the require_spot_confirmation=False control on the same events. If the confirmed and unconfirmed arms both lose by similar amounts, spot confirmation adds no information and the premise (L203 follow-through) does not carry over to this universe. (2) How many exits are 8% stop-outs versus time exits. (3) Whether a mean-reversion sign appears consistently (do not flip the strategy on the basis of this run alone, per L193).

negative_expectancy: The default operationalization of spot-confirmed shock follow-through on six Hyperliquid alt perps (SUI/APT/INJ/TIA/SEI/OP, 4H) loses money on a reliable sample. Over 444 closed legs from 2024-03 to 2026-10, avg_trade_return_pct is -0.925% against the +0.15% Hyperliquid floor. PF is 0.679, Sharpe -1.41 (95% CI -2.71 to -0.20), total return -29.9%, and max drawdown 33.9% against the 20% contract constraint. The loss holds in all three vol regimes and on both long and short legs. 34% of legs exit at the 8% stop, and the signal IC is -0.005 (t=-0.08). The per-trade loss is about 7x the round-trip cost: no plausible setting of shock_z, shock_bars, hold_bars or spot_volume_ratio can bridge that gap, and a negative-IC signal gives the optimizer nothing to tune toward. QA concern borne out on the confirmed arm. The require_spot_confirmation=False control was not run in this backtest, so this review does not show whether spot confirmation adds information. Scope: this rejects only this six-alt 4H follow-through operationalization. It does not reject spot-confirmed follow-through on its original instruments. The apparent mean-reversion is a symptom, not a verified edge; per a past lesson it should not be flipped on the basis of this run.