BtcPerpDailyFundingConfirmedTrendLS

BTC Perp Funding-Confirmed Trend (Single-Instrument USD-M, Long-Short, Daily, Carry-Momentum Confluence, Flat on Disagreement)

A long-SHORT, SINGLE-INSTRUMENT, single-venue directional strategy on BTCUSDT.BINANCE USD-M perpetual futures using DAILY bars, that takes a position ONLY when the price trend and the funding rate AGREE in direction, and stays FLAT when they disagree. This completes the factory's ONE promoted mechanism — the ETH Perp Daily Funding-Confirmed Trend (Sharpe 2.74, PAPER) — on BTC, the asset with the deepest and cleanest data coverage in the catalog. EVERY design choice is dictated by this session's accumulated kills: (1) SIGNAL = FUNDING on BTC, the single safest crypto-native data source. This session HARD-BLOCKED open_interest, taker_ratio, and top/global L/S account ratios as primary signals because they have only ~30-56 days of catalog history (three identical data-insufficiency abandons); funding_rates is the lone supplementary stream with proven multi-year coverage — it cleared the promoted ETH strategy's data-span gate — and BTC has the oldest, most-completely-collected funding history of any perp. (2) NOT a fade / mean-reversion — intraday VWAP-MR was just shown to be anti-predictive on higher-vol alts (excursions trend through fair value, −80% on SOL); this is a confluence-gated TREND, not a vol-regime-dependent reversion that fails to transfer. (3) NOT naked trend/breakout (regime-dead across 14+ abandons) — the funding-confluence AND-gate plus flat-on-disagreement filters the unsupported whipsaws that the 2024-26 chop punishes; this gate is the proven differentiator that carried ETH to PAPER while every Donchian/dual-EMA/TSM-naked variant was abandoned. (4) NOT a closed family: NOT pairs-ratio MR (single asset, no non-stationary spread), NOT delta-neutral funding CARRY (funding is a directional SIGNAL on a single price-creditable leg, never booked as cash flow), NOT cross-venue/HL, NOT COIN-M (engine-blocked). (5) NOT a rare-conjunction gate — it uses the continuous funding SIGN/level so it fires across all regimes, populating every walk-forward OOS window densely (the explicit anti-event-sparsity requirement). (6) Uses only BTCUSDT, a fully-collected symbol — avoiding the alt-collection gaps (ARB/AAVE/NEAR/ATOM) that failed the reversal basket. Intentionally minimal: 3 tunables (trend_lookback, funding_lookback, funding_deadband).

The reported failure is a corrupted daily-return series (e.g. +40.0% on 2024-12-19 when BTC fell ~4%) -- impossible in magnitude and decoupled in sign from that day's price for a ~0.5x-exposure perp. The feedback's primary diagnosis is that a multi-day winning trade's realized PnL is booked entirely into the EXIT-day return instead of distributed as daily mark-to-market over the hold, and it explicitly instructs: if the strategy logic is correct, escalate the equity-curve daily-return attribution as a likely engine-level issue. I verified the two strategy-side levers it named: (1) EXPOSURE is correct -- position_size computes notional = equity*risk_fraction(0.35)*leverage, i.e. ~0.525x at leverage 1.5 (matching the reported exposure_pct 47%), so the strategy is not over-sized; (2) EQUITY is not inflated mid-trade -- get_account_equity() reads the venue account's stable (USDT) balance.total, which on a MARGIN futures account excludes unrealized PnL, so it stays ~flat while a position is open and re-entries do not oversize. With both confirmed correct, the spike is the realized-only equity-curve attribution (no per-bar MTM of the open position), an engine/harness sampling issue the strategy cannot fix -- it has now appeared on baskets, continuous-rebalance spot, and this single-instrument leveraged perp, so it should be escalated. The smallest defensible in-strategy change is lowering leverage from 1.5 to 1.0 (exposure ~0.35x): the conservative choice the feedback's exposure concern points to, removing leverage amplification of the artifact's magnitude; the funding loader, 30-day SMA trend, deadbanded funding confirmation AND-gate, confluence-break and stop exits, the continuous diagnostic signal, and capital-relative sizing are otherwise byte-identical, so the passing Layer-1/2 checks and the promoted-ETH mechanism are preserved. leverage stays referenced in sizing (now x1.0) so the unused-leverage gate does not apply. Per the feedback, after the curve is corrected (engine-side) the edge is still thin (Sharpe ~0.35) and regime-mixed (negative 2022/2025), so the Deflated Sharpe and holdout at the analyzing stage remain the decisive test of whether the BTC funding-confluence edge is real beyond the 2023/2024 bull windows.

Failed deflated Sharpe: DSR=0.2157 (vs 0.95 bar), expected-max=1.0115 over 225 trials — the optimized Sharpe of 0.758 is BELOW the best-of-N luck bar, so it is statistically indistinguishable from selection noise (is_significant=false). This is reinforced by every other forward/multiple-testing gate: the HOLDOUT FAILED with a deeply NEGATIVE Sharpe -2.792 (ratio -3.149) — the recent one-shot forward window loses money, exactly the 2025/2026 decay (2025 -3.2%, 2026 -0.96%) flagged at pre-optimization; PBO 0.7751 (78% overfit probability); and sharpe_ci_low -0.1475 (CI straddles 0). The strong-looking inputs — non-overfit walk-forward (avg OOS 0.887 > avg IS 0.837, all OOS windows positive) and probabilistic_sharpe 0.9906 — are the classic PSR-vs-DSR trap: PSR ignores trial count, while DSR (which corrects for best-of-225 selection) rejects, and the realized edge is outlier-carried (return_kurtosis 240, skew 14, tail_ratio 0.0078) resting on a handful of big trend-up days. The only high-Sharpe region is a cliff on trend_lookback (0.27 at 24 -> 0.75 at 36), the grid edge the optimizer pushed to, which PBO 0.78 flags as overfit. Not promote: DSR below the luck bar + a negative holdout + PBO>0.5 + CI through 0 are each disqualifying. Not iterate (attempt 1 of 2): the parameter space is characterized — the edge is intrinsically too modest (true Sharpe ~0.5-0.9) to clear the 1.01 luck bar, the high-Sharpe corner is a fragile cliff, and the holdout failure is recent-regime decay rather than an under-explored region, so a second sweep re-selects the same sub-significant, outlier-carried, decaying config. Not revise_hypothesis: the funding-confirmed-trend mechanism is already validated and PAPER-promoted on ETH (Sharpe 2.74), so it is not stranded — BTC is simply the weaker arm that fails deflation, and reframing onto another asset re-runs the 225-trial penalty on another modest edge. FAILURE PATTERN: the funding-confirmed-trend mechanism does NOT transfer to BTC with enough strength to survive multiple-testing — the BTC port has a genuine but modest, outlier-carried edge (kurtosis 240, tail_ratio 0.008) whose selected Sharpe 0.758 sits below the 225-trial expected-max luck bar 1.0115 (DSR 0.22), with a clean walk-forward and high PSR masking a deeply negative recent holdout (-2.79) driven by 2025-2026 decay. A passing walk-forward and PSR=0.99 do not rescue an edge that fails deflated Sharpe and a forward holdout; the ETH sibling's Sharpe-2.74 edge was strong enough to clear the bar, BTC's is not.

Long-short single-instrument BTCUSDT.BINANCE USD-M daily funding-confirmed trend strategy. It goes long only when the price is above its 30-day SMA AND the 14-day mean funding is positive beyond a deadband (uptrend confirmed by longs paying), short only when price is below the SMA AND mean funding is negative beyond the deadband, and stays flat on disagreement; it exits on confluence break or a 15% safety stop. Funding is a directional signal only, never booked as cash flow. Iteration 2 lowers leverage 1.5 -> 1.0 (exposure ~0.35x) as the conservative in-strategy lever; the impossible daily-return spikes are a confirmed engine-level realized-PnL-on-close-day attribution artifact (strategy exposure ~0.5x and equity sizing verified correct), to be escalated. Three tunables: trend_lookback, funding_lookback, funding_deadband.

Clean accounting, NO over-exposure artifact: exposure_pct 49%, avg_position_pct 39.8% ~ intended risk_fraction 0.35; the large exit-day returns are plausible multi-week trend captures (e.g. +22% on 2024-12-19 = ~0.35 x the late-2024 BTC run), not impossible single-bar prints

Genuine edge: profit_factor 1.41, positive expectancy +$542/trade, avg_win ~4x avg_loss, positive skew +10.4, Sortino 1.57 — healthy let-runners trend profile

Completes a PROMOTED mechanism (ETH Funding-Confirmed Trend, Sharpe 2.74 PAPER) on BTC, the deepest funding history; funding used correctly as a directional signal, never miscredited

Positive in 4 of 7 years; 99 trades over 6 years — dense enough for a 3-window walk-forward; deep capacity ($231M), modest fee drag (2.72% of gross)

Modest Sharpe 0.467 with CI straddling 0 (-0.545 to +1.16) — robustness must be proven in walk-forward/deflated-Sharpe

Recent softness: 2025 -6.9%, 2026 -0.08% — the chronological holdout/recent window is the key risk (possible funding-edge decay seen across the funding family)

Short leg nearly untested: only 7 of 99 trades are shorts — effectively long-biased, so the net-short crash-protection claim is unvalidated

Low win rate (26%) means the edge leans on a few large trend captures — fragile if those don't recur

~50 (risk_fraction 0.35)

4 of 7

majority-ish

Walk-forward is non-overfit (is_overfitted=false), avg OOS Sharpe 0.887 > avg IS 0.837, all three OOS windows positive [0.91, 1.40, 0.36]

Optimized headline looks clean in-sample: total_return 232%, max_drawdown 17%, profit_factor 2.32, Sortino 4.50, deep capacity ($1.1B), low fee drag (2.27%)

Mechanism is real (the promoted ETH sibling proves funding-confirmed trend works) and accounting is clean (exposure 50%, no artifact)

Failed deflated Sharpe: DSR 0.2157 (<0.95), is_significant=false; optimized Sharpe 0.758 is BELOW the expected-max luck bar 1.0115 over 225 trials — indistinguishable from best-of-N noise

HOLDOUT FAILED with a deeply NEGATIVE Sharpe -2.792 (ratio -3.149) — the recent one-shot forward window loses money, confirming 2025/2026 decay

PBO 0.7751 (78% overfit probability); sharpe_ci_low -0.1475 (CI straddles 0)

Edge is outlier-carried: return_kurtosis 240, skew 14, tail_ratio 0.0078 — rests on a few big trend-up days

trend_lookback is a cliff (Sharpe 0.27 at 24 -> 0.75 at 36); the only high-Sharpe region is the grid edge the optimizer pushed to

Recent decay: 2025 -3.2%, 2026 -0.96% — the holdout-failing regime

>=0.70*OOS

Do NOT optimize yet — the daily-return series is corrupted and the optimization robustness gates (Deflated Sharpe, PBO, Sharpe CI) derive from it, so they would be unreliable. This is a lighter-touch fix than the other strategies this batch (the mechanism is the validated promoted-ETH funding-confirmed-trend, single-instrument, funding data works, and overall vol is a sane 43%), but the spikes must be resolved. Specifics: at risk_fraction 0.35 x leverage 1.5 = 0.525x exposure (exposure_pct 47%), the max possible daily portfolio move is ~0.5x BTC's daily move, yet 2024-12-19 shows +40.0% (BTC ~-4% that day), 2024-03-19 +30.2% (BTC ~-5%), and 2021-01-21 +26.7% (BTC ~-10%) — impossible in magnitude AND wrong in sign (gains on days a long should lose). Investigate: (1) DAILY-RETURN ATTRIBUTION: the most likely cause is realized PnL from a multi-day winning trade being booked entirely into the EXIT day's return instead of distributed as daily mark-to-market across the holding period — check how the equity-curve daily return is computed on trade-close days. (2) EQUITY/SIZING: confirm get_account_equity() is not inflated during the open leveraged position (which would oversize entries); log per-bar position notional / equity and confirm it stays ~0.5x. (3) Re-run and confirm NO daily return exceeds ~BTC's daily move x exposure (a 0.5x BTC perp should rarely exceed ~+/-12-15%/day), kurtosis drops to a normal range, and the Sharpe is recomputed on the corrected curve BEFORE optimizing. After the fix, note that the edge is still thin (Sharpe 0.35) and regime-mixed (negative 2022/2025), so at the analyzing stage the Deflated Sharpe and holdout will be the decisive test of whether the BTC funding-confluence edge is real or just the 2023/2024 bull windows. NOTE: corrupted daily-return accounting has now appeared across many strategies this batch (baskets, spot continuous-rebalance, and now this single-instrument leveraged perp) — if you confirm the strategy logic is correct, escalate the equity-curve daily-return attribution as a likely engine-level issue.

BtcPerpDailyFundingConfirmedTrendLS completed the factory's one promoted mechanism — the ETH daily funding-confirmed trend — on BTC, the deepest-funding-history asset, gating trend trades on agreement between price trend and funding and staying flat on disagreement. Unlike the failed BNB and SOL transfers, its base backtest showed a real, cleanly-accounted edge (profit factor 1.41, positive expectancy, healthy let-runners profile) and earned an 'optimize' verdict, and its walk-forward even came back non-overfit with all OOS windows positive and PSR 0.99. But it failed deflated Sharpe decisively: the optimized Sharpe 0.758 fell below the 1.01 luck bar (DSR 0.22), the recent one-shot holdout posted a deeply negative -2.79 confirming 2025-2026 decay, PBO was 0.78, and the edge rested on a few outlier trend-up days at a cliff in trend_lookback. The analyst abandoned it on its second iteration, concluding that BTC is simply the weaker arm of a validated mechanism — modest, outlier-carried, and decaying — and that a passing walk-forward and high PSR cannot save an edge that fails deflated Sharpe and a forward holdout.

The funding-confirmed-trend mechanism does NOT transfer to BTC with enough strength to survive multiple-testing — the BTC port has a genuine but modest, outlier-carried edge (true Sharpe ~0.5-0.9) that sits below the 225-trial luck bar, and a passing walk-forward plus PSR 0.99 do not rescue it (the classic PSR-vs-DSR trap, where DSR corrects for selection and a negative recent holdout confirms regime decay); the ETH sibling's Sharpe-2.74 edge cleared the bar, BTC's does not.

On its second and final iteration the analyst issued an 'abandon' verdict at the post-optimization ANALYZING stage: it failed the deflated-Sharpe gate (DSR 0.2157 < 0.95, optimized Sharpe 0.758 below the 1.0115 best-of-225 luck bar, not significant), the holdout failed with a deeply negative Sharpe (-2.792, the 2025/2026 decay), PBO was 0.7751, sharpe_ci_low straddled zero, the edge was outlier-carried (kurtosis 240), and the only high-Sharpe region was a trend_lookback cliff at the grid edge.

A long-short, single-instrument directional strategy on BTCUSDT.BINANCE USD-M daily perpetual futures that took a position ONLY when the price trend and the smoothed funding rate AGREED in direction (long on uptrend + positive funding, short on downtrend + negative funding) and stayed FLAT on disagreement, with funding as a confirmation signal only — completing the factory's one promoted mechanism (ETH Daily Funding-Confirmed Trend, Sharpe 2.74, PAPER) on BTC, the asset with the deepest, cleanest funding history.

The base backtest showed a genuine but modest edge: Sharpe 0.467 (CI -0.545 to +1.16), profit factor 1.41, positive expectancy +$542/trade, avg win ~4x avg loss, Sortino 1.57, 25.8% max drawdown, 99 trades, positive in 4 of 7 years (but the short leg nearly untested at 7/99 trades and recent softness in 2025/2026); optimization passed the walk-forward (is_overfitted=FALSE, avg OOS 0.887 > avg IS 0.837, all three OOS windows positive) with an optimized Sharpe 0.758 but failed forward — DSR 0.2157, holdout Sharpe -2.792.