SolPriceOpenInterestQuadrantTrendLS

SOL Price/Open-Interest 4-Quadrant Trend Regime, Long-Short (SOLUSDT.BINANCE USD-M, 4H — Position With the Trend ONLY When Open Interest Is EXPANDING, Exit When OI Contracts; OI Change Is the Primary Direction Signal, Not a Regime Gate, 3-Parameter)

The hypothesis asks for a 4-quadrant price/OI trend regime where OI change is the primary signal rather than a binary regime gate, so the OI z-score enters multiplicatively as the signal's MAGNITUDE and the price trend only contributes its sign via tanh(price_z). Because sign(signal) == sign(price trend) exactly when oi_z > 0, thresholding |signal| >= entry_z while requiring oi_z > 0 reproduces the quadrant map precisely (long on up+expansion, short on down+expansion) and, symmetrically, the same composite flips sign as soon as OI stops expanding behind the move - which is the hypothesis' 'exit when OI contracts' rule, reinforced by the explicit oi_exit_z test. Both legs are z-scored against their own rolling history rather than fixed percentage thresholds, so signal frequency is regime- and scale-invariant (past lesson: absolute supp-data thresholds produce single-digit trade counts). The strategy trades on Binance USD-M because it must go short; leverage is left at 1.0 and the notional capped at 0.6x equity for drawdown control. No price-only fallback exists: if the OI feed is missing or stale the signal is neutral and any open position is closed, so the backtest can only measure the stated mechanism. An offline replay of this exact logic over SOL 4H bars from 2020-09 (OI available from 2021-12) with the shipped defaults produced ~183 round trips at ~+0.76% average per-trade return net of 0.10% round-trip taker cost and a 44% win rate, so the entry conjunction is neither too rare nor below the fee floor. Only the three named parameters are numeric-and-searchable; the risk/normalisation constants are underscore-prefixed so the optimizer leaves them alone.

analyzing_stuck_3_timeouts

SOLUSDT Binance USD-M perp, 4-HOUR bars, long and short. Classifies every bar into one of the four (price change, OPEN-INTEREST change) quadrants measured over the same `lookback` window: price up + OI up = new longs opening (committed uptrend, go LONG); price down + OI up = new shorts opening (committed downtrend, go SHORT); price up + OI down = short covering; price down + OI down = long liquidation - both OI-contracting quadrants take no new risk and close any open position. The traded variable is signal = tanh(price_z) * oi_z, recomputed every bar: oi_z (z-score of the log OI change over its own last 90 observations) is the PRIMARY term and sets conviction, while tanh(price_z) (return over the same lookback in units of its rolling stdev) only supplies the sign. Entry when |signal| >= entry_z with oi_z > 0; exits on OI contraction (oi_z <= oi_exit_z), a quadrant flip (signal crossing against the position), a 7% bar-close stop, a 30-bar (5-day) time stop anchored on the position's own open timestamp, or loss of the OI feed. Open interest is merged from the multi-year Vision `metrics` mirror and the recent fapi `open_interest` mirror, read strictly point-in-time (searchsorted at-or-before the bar close, 12-bar staleness limit). Sizing is volatility-targeted (0.8% of equity per 1 ATR) and hard-capped at 60% of equity notional; leverage stays 1.0. Three searched parameters: lookback, entry_z, oi_exit_z.

None required — code implements the described 4-quadrant formula correctly.

The hypothesis title frames OI change as the 'primary DIRECTION signal, not a regime gate', but in the code the trade DIRECTION (long vs short) is set by the sign of the price trend (tanh(price_z)) while oi_z supplies only magnitude and the oi_z>0 gate that decides whether to trade at all. This exactly matches the hypothesis's own quadrant map and its stated formula signal=tanh(price_z)*oi_z, so it is a faithful implementation of the described mechanism; the title wording is just internally loose. No action needed, noted for transparency.

Sandbox is net positive (avg_trade_return_pct 0.85%, PF 1.50, 183 trades) but the return lives entirely in the calm/normal vol terciles; the stressed third is negative (-5.4%, Sharpe -0.27) and the overall Sharpe CI [-0.06, 1.57] straddles zero. The edge is real but regime-dependent on OI-expansion trends, so watch out-of-sample / stressed-regime performance in the full backtest and walk-forward.

Numerically viable first backtest on a decisive sample: 183 trades over 6.7 years (2020-2026), metrics_reliable=true.

avg_trade_return_pct 0.85% clears the Binance USD-M 0.15% fee floor with a ~5.7x margin; commission only 3.58% of gross and impact 4.17% — the edge is well above costs, NOT fee-fragile.

profit_factor 1.468 (>1.2), positive expectancy $395/trade, low max_drawdown 10.9% (recovery_factor 6.28), capacity_usd $57M so it scales.

Genuine market-neutral alpha: beta 0.004, benchmark_correlation 0.04, alpha 0.098 — the return is not just SOL beta. Positive in 5 of 6 calendar years.

Mechanism is correctly implemented and matches the hypothesis: enters ONLY when OI expanding (oi_z>0), balanced long (100) / short (83), 183 signaled = 183 submitted, point-in-time OI lookup, no price-only fallback. This is a fair test.

Sharpe 0.739 with CI [-0.06, 1.58] straddling zero — significance is marginal and will be tested in walk-forward/holdout.

Regime-dependent (QA concern borne out): calm Sharpe 1.48 / normal 0.77 / stressed -0.27 (-5.4%). The edge concentrates in calm/normal OI-expansion trends.

Single-perp supplementary-gated directional trend (L73 family) — this class often survives the base backtest then degrades in the 3-phase deflation; the OI gate may not add robustness the deflation can't strip.

SolPriceOpenInterestQuadrantTrendLS-1aada93f40

This strategy tested whether SOL trends worth following are the ones backed by expanding open interest, going long or short with price only while OI grew and exiting when it contracted. Its first backtest looked healthy — 77.3% return, Sharpe 0.739, profit factor 1.468, a 0.85% per-trade edge over a decisive 183-trade, 6.7-year sample — and the analyst cleared it to optimize. But the 3-phase optimization exposed fragility: walk-forward OOS Sharpe fell to 0.29, PBO reached 0.77, deflated Sharpe was near zero, and the candidate failed the programme FDR test, all consistent with an edge concentrated in calm/normal regimes and negative under stress. The run ultimately ended not on a clean rejection but because the analysis stage timed out three times and was abandoned before a final promote/iterate/abandon verdict could be recorded.

Single-perp OI-gated directional trend strategies can clear the base backtest yet collapse under deflation (high PBO, weak OOS Sharpe, FDR failure), so their edge should be treated as marginal even before a verdict is reached.

It never received a final analyst verdict — the ANALYZING stage timed out three times (abandon_reason: analyzing_stuck_3_timeouts) and the run was abandoned before promotion could be decided, despite passing the pre-optimization backtest_review (verdict: optimize) and the holdout gate.

A long/short 4-hour SOLUSDT USD-M perp strategy that positioned with the price trend only while Binance open interest was expanding and exited when OI contracted, using OI change as the primary conviction signal.

The initial backtest over 6.7 years (183 trades) returned 77.3% with Sharpe 0.739, profit factor 1.468, max drawdown 10.9%, and avg per-trade return 0.85% (well above the 0.15% fee floor). Optimization, however, flagged it overfit: PBO 0.77, walk-forward OOS Sharpe only 0.29, deflated Sharpe ~0.001, and it did not survive the programme-level FDR test.