SOL-AVAX L1 Pair Mean-Reversion on Hyperliquid 15-Minute (Short-Horizon Stat-Arb, Long-Short Cointegrated Pair)

Short-horizon market-neutral statistical arbitrage trading the SOL/AVAX ratio on Hyperliquid perpetuals. SOL and AVAX are both high-throughput L1 blockchain tokens with strongly cointegrated price action driven by shared macro factors (L1 narrative flows, DeFi TVL rotation, validator-staking economics, layer-1-vs-layer-2 sector sentiment). Their RATIO is significantly more mean-reverting than either price alone because shared market-beta cancels out. The strategy samples 15-minute bars on SOLUSD.HYPERLIQUID and AVAXUSD.HYPERLIQUID, computes the ratio, and trades z-score reversion around its 96-bar SMA (24 hours of context). When the ratio is > 2σ above its mean, ETH/BTC-style mean-reversion logic dictates AVAX is undervalued vs SOL → short SOL, long AVAX. Inverse for ratio < -2σ. This hypothesis simultaneously fills FOUR under-represented quotas: HYPERLIQUID (4.2% vs 20% target), short_1m_15m horizon (4.2% vs 10% target), long_short direction (9.5% vs 45% target), and pairs scope (14.1%). Critically distinct from EthBtcRatioMomentum already in pipeline: that trades the RATIO TREND (momentum, daily/4H, BTC vs ETH), this trades the RATIO REVERSION (mean-rev, 15m, two correlated L1 peers). Distinct from FundingExtremeReversion (uses funding rates, single asset). Distinct from the failed BTC daily 3-sigma return outlier mean-rev (different mechanism: that fades single-asset return tails which has no edge; this trades cointegrated-pair spread which has decades of academic support — Engle-Granger 1987, Gatev-Goetzmann-Rouwenhorst 2006). Simple parameter footprint (3 main: lookback=96, entry_z=2.0, exit_z=0.3) minimizes overfitting risk after recurring frozen-signal failures.

Implements the cointegrated-pair reversion with the repo's SyncedBuffer pairs helper and the proven centralized multi-leg pattern (same architecture as the landed SOL/AVAX 4H and LINK/ETH 1H pairs). SyncedBuffer guarantees the SOL/AVAX ratio is always computed on contemporaneous closes -- it commits an (a,b) sample only when both legs report the same ts_ns -- which at 15m resolution is essential because a one-bar misalignment manufactures a phantom ratio jump that dwarfs the real spread edge; the base cross-leg barrier (both legs 15m) ensures calculate_signal fires only when the AVAX bar for the current SOL timestamp is present. The strategy is symmetric two-sided: z >= entry_z shorts SOL/longs AVAX, z <= -entry_z does the inverse, and each side exits back through the exit_z band toward the mean (short side at z<=exit_z, long side at z>=-exit_z) or on a 96-bar time stop. Entry is dollar-neutral (each leg sized to equity*leg_frac/price, not equal qty) so P&L tracks the ratio reverting rather than either L1's direction, and atomic -- if either leg fails its qty/min-notional check, neither opens, preventing a naked directional bet. leg_frac=0.45 keeps gross ~90% within the leverage-1.0 margin ceiling (leverage not set >1, so the unused-leverage gate does not apply). min_bars_required=2 lets the buffer fill immediately while zscore_last returns None until 96 aligned samples accrue, gating warmup without dead bars. The returned z-score is a genuinely continuous per-bar signal so Layer-2 frozen-signal detection passes while entries stay threshold-gated. Fills the HYPERLIQUID, short_1m_15m, long_short, and pairs quota gaps; both legs are liquid HL majors with multi-year 15m history. The should_enter/should_exit/position_size hooks are inert because the base entry path manages only the primary instrument.

SOL-AVAX Hyperliquid 15m ratio-MR pairs fails on three fronts simultaneously. (1) DECISIVELY NEGATIVE EDGE: over the available window Sharpe is -3.30 with the ENTIRE bootstrap CI negative (-5.47 to -0.75), profit_factor 0.79, expectancy -$46/trade, probabilistic_sharpe 0.0013 — the SOL/AVAX ratio did not reliably revert (avg_loss $458 > avg_win $340, skew -3.39), consistent with the SOL-AVAX pair's prior de-cointegration abandon. (2) DATA-INSUFFICIENT: only ~2.3 months of HL 15m data (2026-03-16 to 2026-05-24, data_days 102), the same Hyperliquid-15m coverage wall that already killed the prior SOL-HL-15m transplant — far too little for a valid 3-window walk-forward + 15-day holdout. (3) DEGENERATE CAPACITY/IMPACT: capacity_usd only $38K with impact_cost_pct 162% (modeled market impact exceeds gross PnL) and commission 16.1% of gross — a 15-minute SOL/AVAX spread on Hyperliquid is too small and too illiquid to clear costs at any deployable size. Not optimize: a negative-edge, ~2-month, capacity-walled base cannot be validly optimized. Not iterate: unlike the ETH-HL-1H VWAP case (a promoted-sibling mechanism with a fixable exit and a borderline result), here the edge is decisively negative with a fully-negative CI, the spread doesn't clear HL fees+impact at $38K capacity, and the HL-15m data wall is a structural blocker already confirmed. FAILURE PATTERN: 15-minute cointegrated-pair MR on low-liquidity alt perps (SOL/AVAX on Hyperliquid) is triply blocked — (a) HL 15m has only ~2 months of history (unvalidatable), (b) the alt-pair ratio de-cointegrates / does not reliably revert (negative Sharpe with fully-negative CI), and (c) the per-trade spread is far too small to survive HL fees + market impact ($38K capacity, 162% impact). This is the same de-cointegrated/illiquid-pair signature as the abandoned SOL-AVAX, AAVE-COMP, and ETH/BTC pairs, compounded by 15m fee-fragility and the HL-15m data gap. The Research Lead should stop proposing HL-15m and low-liquidity-alt-pair stat-arb until both the HL sub-hourly data coverage and a genuinely cointegrated, liquid pair are confirmed.