DogeOpenInterestConfirmedTrendContinuationLS

DOGE Open-Interest-Confirmed Trend Continuation, Long-Short (Single-Instrument DOGEUSDT.BINANCE USD-M, 4H Primary + 1D Trend Confirm, Trade Directional Trend ONLY When Open Interest Is EXPANDING With the Move = Fresh Capital, Flat/Exit When OI Contracts = Short-Covering Exhaustion, ATR-Trailing, 3-Parameter)

A LONG-SHORT, single-instrument directional TREND-CONTINUATION strategy on DOGEUSDT.BINANCE (USD-M perpetual), 4H primary bars with a 1D trend confirm. It sits squarely inside the ONLY archetype family this factory has ever promoted (BTC/ETH/ADA Dual-Timeframe Momentum Confluence, long-short, 4H+1D, low-frequency, ATR-trailing) but its confirmation signal is NOT price/volume (L91: the volume-confirmation edge is ETH-specific and does not generalize) and NOT a regime/vol/efficiency gate (L92: those only cut trade count). Instead it uses the OPEN-INTEREST feed to measure the QUALITY of the flow behind the move: in perps, price rising WITH open interest rising means new leveraged longs are entering (genuine, self-financing trend that continues), whereas price rising WITH open interest FALLING is short-covering — buyers closing shorts, no fresh conviction — which mean-reverts. Symmetric for the short side (price down + OI up = fresh shorts = continuation; price down + OI down = long-liquidation exhaustion = fade risk). This is a documented crypto-native microstructure effect that is mechanically distinct from volume: two bars with identical volume can have opposite OI deltas and opposite continuation odds. DOGE is chosen deliberately: it is extremely liquid (tight spreads, deep book, ~0.10% RT taker) yet massively under-tested vs the BTC/ETH/SOL concentration, and its retail-speculative, leverage-driven trends produce the largest, cleanest OI-expansion signatures of any liquid major (OI can double inside a multi-day trend), giving this mechanism its highest signal-to-noise. FEE ARITHMETIC: entries fire only on 4H trends confirmed by the 1D direction AND OI expansion, so cadence is LOW (est. ~150-260 trades over ~5.5 years of DOGE 4H history, ~1 trade/wk). A confirmed DOGE 4H trend leg with expanding OI typically runs 3-6%+ before the ATR trail stops it; capturing even ~35-45% of a 3.5% average favorable leg = ~1.2-1.6% gross per winning trade, and the OI-contraction exit cuts the short-covering fakeouts that produce the small losers. Blended expected per-trade capture ~0.45-0.6% of notional vs 0.10% round-trip taker cost = 4.5-6x the fee bar, well clear of the fee_edge graveyard that kills fast sub-fee mechanisms. Only 3 parameters (momentum lookback, OI-expansion lookback, ATR-trail multiple) to resist overfit — the same low-parameter shape as the promoted winners. Position sizing: volatility-scaled, ~2-3% equity risk per trade to the ATR stop, single position, max ~25% of equity notional per trade at 1-2x leverage (futures used because the mechanism is genuinely long+short and OI/leverage-based, per venue-routing rules).

Implements the hypothesis exactly: the promoted dual-timeframe confluence skeleton (4H primary + 1D confirm, long-short, ATR-trailed, 3 tunables) with the confirmation slot filled by the OPEN-INTEREST feed rather than price/volume or a vol/regime gate. calculate_signal() returns the continuous ATR-normalised momentum every bar (natural units, ~+-5); the OI-expansion and daily-direction gates live in should_enter(), and OI contraction is a first-class exit. Multi-year OI depth comes from merging `metrics` (open_interest column, 2021-12 onward, ~475k 5-minute points thinned to 30-minute spacing at load) with `open_interest` (the recent-months mirror, which alone only reaches back ~6 months); the merged series is walked with a forward-only pointer so per-bar cost is O(1) and the run is nowhere near the 300s budget. Offline dry-run over the full 13,492-bar DOGE 4H history with the shipped defaults: 366 trades (2021-12 to 2026-08, ~78/yr), avg gross per-trade return +0.66% of notional (~0.56% net of the 0.10% round-trip taker cost, 5-6x the fee bar), 36.6% win rate with a trend-follower's right tail, avg hold 8 bars (~32h) -- so neither a zero-trade nor a sub-fee outcome. Venue: BINANCE USD-M futures (genuinely long+short, and the OI/funding microstructure only exists on the perp). Leverage is left at 1.0 deliberately: the notional cap is 30% of equity, so a leverage>1 request would be inert in the engine (position never exceeds 1x equity notional) -- the sizing formula still reads self.config.leverage so raising the cap and the leverage together stays coherent. Three tunables only (mom_lookback, oi_lookback, trail_atr_mult) with their code clamps mirrored in _param_bounds.

analyzing_stuck_3_timeouts

Long-short 4H trend continuation on DOGEUSDT.BINANCE (USD-M perp) that only takes a directional position when OPEN INTEREST is expanding with the move. Direction comes from an ATR-normalised 18-bar 4H momentum read that must agree with a 20-day 1D trend confirm; the trade is only taken when open interest has grown more than 1% over the last 12 bars (~2 days) -- fresh leveraged capital entering behind the move. Exits: ATR trailing stop (3x entry ATR give-back from the best close since entry), OI CONTRACTION beyond 2% (short-covering / long-liquidation exhaustion -- the fuel behind the move leaving), a 1D trend flip, or a 4H trend flip. Open interest is read point-in-time from the merged `metrics` (multi-year 5-minute Vision series) and `open_interest` (recent OI-history mirror) supplementary feeds via a forward-only pointer; when no fresh reading exists the strategy stays flat rather than degrading into a price-only trend follower. Sizing is risk-first: the trailing-stop distance risks ~1.5% of equity per trade, capped at 30% of equity notional.

No change required if base warmup and this override are compatible; otherwise rename the internal warmup helper to avoid shadowing.

min_bars_required and _submit_exit both override/shadow framework methods (static-analysis base_shadow warnings). The overrides look deliberate — min_bars_required is used as an internal warmup gate in should_enter, and _submit_exit correctly calls super() before resetting trailing state — so behavior is intact, but confirm the base template does not also rely on its own min_bars_required for warmup gating (double-gating would only delay first entry, not corrupt logic).

None needed given sandbox coverage; if a future venue/feed renames the OI column (e.g. snake_case sum_open_interest) the merge would silently yield fewer samples.

OI columns are matched on _OI_FIELDS=('open_interest','sumOpenInterest') and _row_oi excludes *Value (USD) columns to avoid a disguised price delta — good. Correctness depends on the Binance Vision `metrics` feed exposing one of those exact keys. The sandbox traded 368 times across the full 2021–2026 span, which confirms the merged metrics+open_interest series is actually populating and being read point-in-time, so this is informational only.

Sandbox is positive (PF 1.25, avg_trade_return_pct 0.60%, 368 trades) and clears the fee bar comfortably, but Sharpe 0.45 with CI [-0.37, 1.20] straddles zero and edge is concentrated in the 'normal' vol tercile (calm +2.65%, stressed +2.73% vs normal +28.8%). Worth checking on full history whether the OI-expansion edge is regime-dependent rather than broad-based; the tariff_shock_2025 window (-7.9%, Sharpe -6.8) shows the ATR trail can lag on fast reversals.

Real edge above costs: avg_trade_return_pct 0.60% is ~4x the 0.15% Binance USD-M fee floor; expectancy +$98/trade, commission only 4.86% of gross, impact 7.5% (capacity $17.7M) — this is NOT a fee-fragile mechanism.

Reliable, adequately-sized sample: 368 trades over 6.2 years (metrics_reliable=true), so optimization has real data to fit rather than noise.

Trades implement the stated hypothesis: 368 signaled = 368 submitted, balanced long/short (177/191), the OI supplementary feed is actually consumed and gates entries — no mechanism mismatch, no broken signal, no fallback to price-only trend.

Acceptable risk: max_drawdown 15.4%, headline is realized (end_unrealized_pct only 2.6%, not an MTM artifact), positive across annual buckets in 2022/2024/2025.

Sits in the one Binance-USD-M dual-TF-confluence archetype family that has produced promotions (L82), with the low-parameter (3-tunable) shape that resists overfit.

Modest risk-adjusted performance: Sharpe 0.45 with bootstrap CI [-0.37, 1.20] straddling zero, Sortino 0.46, PF 1.234 — the edge is real but thin.

Regime-concentrated edge (QA concern borne out on full history): normal-vol tercile +28.8% (Sharpe 0.99) vs calm +2.65% and stressed +2.72%; tariff_shock_2025 window -7.9% (Sharpe -6.8), ATR trail lags fast reversals.

Single-perp OI-gated directional family (L73) with PF < 1.7 has historically died overfit in the 3-phase deflation — this is a genuine risk the optimization/holdout must resolve, not a reason to pre-abandon.

DogeOpenInterestConfirmedTrendContin-e2d5aff703

This strategy filled the factory's only promoted archetype — long-short 4H+1D confluence, ATR-trailed, three tunables — with an open-interest confirmation signal on the deliberately under-tested but liquid DOGEUSDT perpetual, trading trends only while OI expanded with the move. The backtest showed a genuine, fee-clearing edge (Sharpe 0.45, 0.60% per trade, PF 1.23 over 368 trades), enough for the analyst to approve optimization, and optimization lifted the numbers and passed the holdout gate. However, the deeper robustness checks were unflattering (walk-forward overfitted, PBO 0.48, deflated Sharpe near zero, did not survive programme FDR), and before the analyst could render a final verdict on that tension the ANALYZING stage timed out three times. The run was therefore abandoned procedurally after a single iteration, its overfitting question left unresolved.

A slow single-perp OI-gated directional strategy can show a real above-fee edge yet remain statistically indistinguishable from programme noise (fails FDR/deflated Sharpe); it also needs to reach a completed analyst decision, since a stalled/timed-out analysis ends the run before its overfitting risk is properly adjudicated.

It cleared the pre-optimization backtest-review gate (verdict: optimize) and completed the 3-phase optimization, but was abandoned in the ANALYZING stage for a procedural reason — the analyst timed out three times (analyzing_stuck_3_timeouts) — so no final promote/iterate/abandon merit verdict was ever recorded.

A long-short dual-timeframe (4H + 1D) trend-continuation strategy on DOGEUSDT.BINANCE that used the open-interest feed as its confirmation slot — trading a confirmed directional trend only when open interest was expanding with the move (fresh leveraged capital) and exiting on OI contraction (short-covering/liquidation exhaustion).

The initial backtest over ~6.2 years produced Sharpe 0.45 (CI [-0.37, 1.20]), total return 38.79%, profit factor 1.234, max drawdown 15.36% across 368 balanced long/short trades, with avg per-trade return 0.60% of notional (~4x the 0.15% fee floor). Optimization improved it to Sharpe 0.61, return 54.9%, PF 1.36 and passed the holdout gate (holdout Sharpe 1.91 on 71 trades), but walk-forward flagged it overfitted with PBO 0.48, deflated Sharpe ~0.005, and it did not survive the programme-level FDR.