XrpDailyTsmomFundingCrowdednessLS

XRP Daily Time-Series Momentum with Funding-Crowdedness Confirmation, Long-Short

Iteration 2 fixes the backtest gate's finding: the equity curve ended at -96.8% (the Sharpe/return incoherence is the symptom of a compounding blowup, and the equity curve is the truth). The cause was churn plus over-sizing, not the signal: (a) entry_threshold 0.50 on a momentum t-stat with an exit at exactly 0.0 flipped the book on every marginal zero-crossing, so a long whipsaw run stopped out repeatedly; (b) each of those stops cost risk_frac 0.05 of equity (notional = 0.05/0.09 = 0.55x equity, up to 2x conviction, capped at 0.75x), so a modest losing streak compounded to ruin. The smallest fix that addresses exactly that: raise entry_threshold to 1.00 (trades only genuine trend t-stats), give the exit a negative hysteresis band (-0.15, symmetric for shorts) so positions are not flipped by noise around zero, cut risk_frac to 0.02 and widen the bracket stop to 0.15 (XRP daily sigma is ~4-5%, so a 9% stop was ~2 daily sigma and was being tagged constantly) which brings gross notional to ~0.13x equity, and lower the caps (conviction 1.5, gross 0.50x equity, leverage 1.0). Signal computation, funding overlay, point-in-time supp_window access and imports are untouched, so Layers 1-2 stay green.

analyzing_stuck_3_timeouts

XRPUSDT daily time-series momentum (volatility-normalised 30-day return t-stat) with a funding-rate crowdedness overlay added as a continuous confirmation term: mild same-side funding adds conviction, extreme funding z-scores subtract and eventually flip it. Long and short on Binance USD-M perps at 1x, entries at |signal| > 1.0 with a symmetric hysteresis exit band, venue-managed bracket (15% stop / 60% take-profit) and risk-budgeted sizing capped at 50% of equity notional.

No action required; the override is intentional and correctly implemented.

min_bars_required() overrides a framework hook (Layer-1 base_shadow warning). This is a documented, deliberate override — the base method exists and the override returns a sane warmup (>= mom_lookback+2). No functional problem.

Sandbox smoke test (unoptimized, single param draw) shows avg_trade_return_pct 0.0759% — below the ~0.15% Binance USD-M round-trip floor — with PF 0.90, total_return -2.0%, win_rate 0.25 and a losing calm-vol tercile (-5.9%). The entry gate is a momentum t-stat (>1.0), not a price-move threshold, so check-8's fee-floor logic does not mechanically apply, and ~15-day daily-momentum holds leave room for >0.15% moves; but the marginal per-trade edge is worth watching. The funding-crowdedness overlay is additive (confirm = w*z*(1-|z|/extreme)) and only mildly moves the t-stat, so its incremental contribution over pure TSMOM should be scrutinized at the real backtest. Carry forward to the analyst for the performance decision.

Per-trade edge is genuinely above costs on the full-history sample: avg_trade_return_pct 2.17% (pct_notional) vs the 0.15% Binance USD-M floor. QA's smoke-test fee concern (0.0759% on the truncated, unoptimized window) is NOT borne out on the full backtest.

Positive gross edge, not fee-fragility: commission is only 1.0% of gross and profit_factor is 1.076 — the thin PF reflects a weak-but-real gross signal, not costs eating the edge.

Proper trend-follower payoff structure: win_rate 0.314 with avg_win/avg_loss ~2.35:1 — a low hit rate with a >1.5 payoff is expected for TSMOM, not a defect.

Trades match the hypothesis: 86 signaled = 86 submitted, 57 long / 29 short, ~16.7d avg hold consistent with 30-day daily momentum; no signal-silencing gate (funding overlay is additive, not a conjunctive gate).

Adequate sample and span for a daily strategy: 86 trades over 6.7 years, DD 24.2% (< 60%).

Thin and statistically noisy: Sharpe 0.319 with CI [-0.44, 1.01] straddling zero; PF 1.076 sits at the low end.

Deteriorating and regime-dependent: profitable in only 4 of 7 calendar years (loses 2023 -10.9%, 2025 -12.0%, 2026 -3.6%), with most gains from the Nov-2024 rally; positive-years share 57% is below the 60% validity floor.

~16.2% of the 28.4% total return is end-of-backtest unrealized (open position) — headline is partly paper.

This is the single-name OHLCV momentum clone family that frequently dies in the 3-phase deflation (L63/L93); optimization must demonstrate a robust parameter region, not curve-fit the 2024 window.

Backtest gate: - metrics unreliable: Incoherent: Sharpe 2.67 disagrees in sign with total_return -96.80% — the Sharpe is computed from the mtm_equity_curve series while the return comes from the equity curve. Judge on the equity curve.

Benjamini-Hochberg at q=0.10 over 254 programme candidates keeps 4. A candidate that does not survive here is not distinguishable from the programme's own noise, however good its individual statistics look.