BtcOpenInterestAccelerationTrendConfirmationLong4H

BTC Open Interest Acceleration Long-Only Trend Confirmation (Single-Asset BINANCE 4H, OI/Price-Divergence Position-Flow Signal)

Single-asset long-only strategy using a GENUINELY NEW mechanism class never tested in the portfolio: Open Interest acceleration as a forward-looking trend-confirmation signal. The analyst explicitly demanded 'genuinely new hypothesis classes' and stopping 'hypotheses that are structurally identical to abandoned strategies. The mechanism is what matters, not the framing.' This proposal uses OPEN INTEREST (perp OI from supplementary_data, distinct from any signal currently in the portfolio) combined with the classical futures-market signature 'rising price + rising OI = new long positions deploying = bullish continuation' vs 'rising price + falling OI = short cover = bearish exhaustion' (Schwager 1989 'Complete Guide to the Futures Markets' chapter on OI/price/volume divergence; this is foundational futures-market analysis but has never been systematically tested in the crypto factory). Mechanism: When BTC perp OI accelerates UP simultaneously with price closing above its 20-bar SMA, fresh leveraged longs are deploying capital — a flow signal that PRECEDES sustained price drift (not lags it like simple momentum). Go LONG, hold while OI continues to expand (capital still flowing in), exit when OI rolls over (capital exit signal — leading indicator of trend exhaustion). This is NOT volume-breakout (already in pipeline 5+ variants), NOT EMA crossover (already in pipeline), NOT cross-venue (parked by analyst), NOT cross-sectional (failed), NOT funding-extreme (dead regime), NOT a classical chart pattern, NOT positioning-extreme contrarian (just submitted). It's a flow-confirmation TREND strategy using OI rate-of-change — fundamentally different mechanism. Single asset (BTC), single venue (BINANCE USD-M), 4H bars with multi-year history (safe data), long-only (avoids short-leg crypto disasters per analyst's recommendation). Uses open_interest from supplementary_data which IS actually collected by the factory (unlike liquidations, unlike HL funding). Four parameters total — minimal overfitting surface.

Implements the hypothesis's genuinely-new mechanism class: Open Interest acceleration as a forward-looking trend-confirmation/flow signal, distinct from volume-breakout, EMA-crossover, funding, cross-venue, or contrarian-positioning strategies already tested. The classical futures signature 'rising price + rising OI = bullish continuation' is encoded as: signal = OI rate-of-change (continuous, varying every bar) gated against price>SMA for entry, with exit on OI rollover (the leading capital-exit signal central to the thesis) plus trend-break and stop. Single asset (BTC), single venue (BINANCE USD-M where OI is intrinsic and taker fees 0.05% are cheaper than spot), 4H bars with multi-year safe history, long-only to avoid short-leg crypto blowups. Uses open_interest which the factory actually collects. Four core signal params (sma_period, oi_lookback, oi_entry_threshold, oi_exit_threshold) keep the overfitting surface minimal. Holds across many 4H bars so the trend capture comfortably clears the ~0.10% round-trip taker cost.

Negative-expectancy mechanism plus an insufficient effective sample. In the only window where the open-interest gating data exists (all 33 trades are in 2026), the strategy loses money: profit_factor 0.66, total_return -8.13%, expectancy -$247.58/trade, avg_trade_return_pct -0.124% (negative), Sharpe -0.34 with CI [-0.937, 0.282], information_ratio -0.76. The sign of the edge is wrong — the OI/price-divergence long signal underperforms — and no tuning of 4 parameters can flip a negative-expectancy long-only book positive. Compounding this, the OI supplementary feed only covers ~6 recent months, so despite 6.6 years of price bars the effective sample is ~33 trades — far too thin for a 3-window walk-forward plus holdout. This is the supplementary-gated single-perp data wall (L73), same as the recently-abandoned taker-flow, funding-divergence and CoinM-liquidation cases. Not worth 2 hours of optimization.

Long-only BTC 4H trend-confirmation strategy on BINANCE USD-M futures driven by perpetual Open Interest rate-of-change. calculate_signal returns a continuous OI ROC (%) over oi_lookback bars (default 6 bars = 24h). Entry: OI ROC above oi_entry_threshold (capital deploying) AND price above its 20-bar SMA (rising price + rising OI = fresh longs). Exit: OI rolls over below oi_exit_threshold (capital exiting, leading exhaustion signal), price drops below the SMA (trend break), or the protective stop-loss triggers. OI is parsed from supplementary_data['open_interest'] with robust key/timestamp normalisation and carry-forward alignment with a staleness guard; when OI is unavailable the signal is neutral and the strategy does not trade (no price-only fallback). Sizing is equity-relative (position_fraction of equity, leverage 1.0).

CLEAN RESTART 2026-09-04 — this run's verdict history and learning records were removed and it was restarted from verification. Its previous abandonment came from the pipeline, not from the market: the Layer-2 harness mis-bound @staticmethod helpers (fixed), QA issued terminal performance verdicts on an unoptimized smoke test (removed — QA now judges correctness only), and sandbox timeouts came from backtest-slot starvation (fixed). The hypothesis and the strategy code are unchanged. Verify the code on its merits; performance is decided later by the full backtest and the optimizer.

At optimization, widen the search cautiously (33-trade base overfits easily), weight deflated/OOS Sharpe and PBO heavily; if trade count can't rise toward ~100 without loosening into a non-OI signal, abandon rather than tune.

The entry gate (OI ROC > 1% over 24h AND close > SMA20) fires only 33 times over the full 2398-day sandbox (~5/yr) — below the ~100-trade threshold for reliably separating edge from noise, and it's the full history so the backtest won't add many more. Not a code defect and not data-absence (OI present, signal triggers), but it makes WF windows and holdout statistically thin.

Distinct order-flow mechanism (OI rate-of-change) not otherwise in the portfolio; clean, no-fallback gating on real OI data

Execution is correct: 33 signaled, 33 submitted, no drops; long-only, drawdown contained at 10.9%

Net loser: total_return -8.13%, profit_factor 0.66, expectancy -$247.58/trade, avg_trade_return_pct -0.124% (negative)

Negative risk-adjusted metrics: Sharpe -0.34 (CI [-0.937, 0.282]), Sortino -0.09, information_ratio -0.76

All 33 trades fall in 2026 — the OI feed lacks deep history, so the effective sample is ~6 months despite 6.6 years of price bars

That window cannot support walk-forward + holdout; supplementary-gated single-perp data wall (L73)

return_kurtosis 174.6 — a single +5.67% day (2026-08-20) is the only thing propping up an otherwise uniformly negative tape