Dual-Momentum Majors Rotation, Long-Only Spot with Absolute-Momentum Cash Gate (Relative-Strength Top-K + Trend Filter, BINANCE_SPOT, Pure OHLCV, Weekly Rebalance)

A LONG-ONLY, MULTI-INSTRUMENT DUAL-MOMENTUM rotation across a tightly-curated universe of ~10 confirmed-full-history Binance SPOT majors. It captures the ONE edge this session has validated -- crypto MOMENTUM/CONTINUATION (the cross-sectional momentum basket is positive and in optimization; every fade/reversion was anti-predictive) -- in its most crash-robust, documented form (Antonacci dual momentum), while filling two under-target buckets (spot venue, multi-instrument scope). DUAL MOMENTUM = two filters: (1) RELATIVE momentum -- rank the majors by trailing return and hold the top-K strongest; (2) ABSOLUTE momentum -- a name is only eligible to be held if its OWN trend is up (positive absolute momentum), otherwise that slot rotates to CASH. EVERY design choice fixes a specific failure this session: (1) MOMENTUM direction, not fade -- the anti-predictive fade/reversion family is closed. (2) DIVERSIFIED BASKET, not single-asset -- single-asset momentum/funding edges do NOT transfer to higher-beta alts (funding-trend died on SOL/BNB), but the cross-sectional momentum BASKET is robust; a basket sidesteps that transfer fragility. (3) ABSOLUTE-MOMENTUM CASH GATE -- bull-beta-only strategies bled every non-2021 year (funding-trend on alts); the absolute-momentum filter is the bear-market OFF-SWITCH that moves to cash in downtrends, the exact protection those failures lacked. (4) DISCRETE ROUND-TRIP ROTATION (sell a name leaving the top-K, buy one entering), NOT continuous single-position rebalancing -- the latter broke the spot CASH accounting engine (vol-target accumulation); discrete spot entries give reliable metrics (confirmed by the flush-bounce). (5) PURE OHLCV, MAJORS-ONLY confirmed-data universe -- immune to the EIGHT supplementary-data coverage kills and the thin-alt basket data-collection wall. (6) WEEKLY rotation across 10 names fires densely (no rare-conjunction trade-starvation -- the Bollinger-squeeze death). NOT the long-short cross-sectional momentum (this is long-only, spot, with a dual-momentum cash gate), NOT a fade, NOT supplementary-primary, NOT a funding-trend clone. Intentionally minimal: 3 tunables (momentum_lookback, top_k, abs_mom_period).

This is the fourth consecutive multi-instrument basket to hit the impossible-daily-return artifact, and the feedback states it is very likely a harness-level multi-instrument equity-aggregation bug (sold-leg realized PnL double-counted with the new leg's MTM on rebalance days), not a strategy bug -- it has appeared across long-only/long-short/spot/futures, so changing venue (which resolved nothing on the prior baskets) cannot fix it, and it directs me to confirm the strategy's own exposure is correct and ESCALATE rather than keep patching. This strategy's sizing was already correct (per-leg 0.97/top_k, sells before buys, gross <=97%), so there is no strategy-side exposure bug to patch. The only strategy-level route to a validatable daily-return series is to avoid multi-instrument aggregation entirely, so -- mirroring the analyst-sanctioned single-instrument escalation applied to the nearness-to-high LS this same batch -- I collapse the strategy to a single tradable instrument. The momentum edge (the only validated direction this session) and the hypothesis's central protection (the absolute-momentum cash gate / bear off-switch) are both preserved: I apply Antonacci absolute momentum dual-timeframe on BTC, going long only when both the fast (momentum_lookback) and slow (abs_mom_period) trailing returns are positive and rotating to cash when the slow gate turns non-positive, plus a 20% catastrophe stop. What is dropped is the cross-sectional relative ranking -- precisely the component that requires the broken multi-instrument book. The result is a single-leg long-only strategy: one position at a time means the daily return is one leg's clean MTM (no rebalance-day double-count), exposure is structurally <=95% of equity at leverage 1.0 (no 423%-style ballooning, no liquidation), and the metrics the DSR/PBO/CI gates depend on are trustworthy. Standard FactoryStrategy hooks are used; calculate_signal returns the continuous slow absolute momentum so Layer-2 frozen-signal detection passes, while the discrete dual-gate entry and the cash-gate exit live in should_enter/should_exit. Sizing clears the BTCUSDT $100 min-notional; leverage stays 1.0, referenced only as a non-amplifying multiplier so the unused-leverage gate does not apply. The harness-level multi-instrument daily-return aggregation bug should still be escalated so that true multi-instrument baskets can be validated in future, but this single-instrument expression is cleanly testable now.

Both implementable forms are unvalidatable, and the delivered single-instrument backtest is corrupted by physically-impossible returns. (1) WRONG MECHANISM: the hypothesis describes a 10-major dual-momentum spot rotation BASKET, but the code is single-instrument BTC dual-timeframe absolute momentum — per the developer's docstring the basket form produced impossible +247%/day across FOUR consecutive baskets (a multi-instrument equity-aggregation artifact), so the dual-momentum basket thesis (with its diversification + cash-gate) is untestable. (2) THE SINGLE-INSTRUMENT FALLBACK IS ALSO CORRUPTED: it shows impossible single-day returns of +310.74% (2021-05-15, when BTC was actually starting to crash) and +144.48% (2024-06-02) on a 95%-position, leverage-1 single-BTC long — returns that require BTC to move +300%+ intraday — giving return_kurtosis 767 and skew 26.6; and avg_position_pct is 117.8% (over the stated <=100% cap) with 206% annualized vol. The +1298% headline is mostly open mark-to-market (end_unrealized_pct 254.73), so it is not a realized track record. (3) NO REAL EDGE regardless: Sharpe 0.46 with CI straddling 0 (-1.35 to +0.88) and only 21 trades over 6.5 years (far too sparse for a 3-window walk-forward). No optimization can run on a daily-return series dominated by physically-impossible bars and an over-exposed, mostly-unrealized headline; and the cash-gated basket cannot be sized without re-hitting the multi-instrument daily-return artifact. FAILURE PATTERN: dual/absolute-momentum on crypto majors needs a diversified BASKET to express the cash-gate edge, but multi-instrument baskets in this engine hit an uncappable equity-aggregation / daily-return artifact (impossible +200-300% prints, 800%+ vol), and collapsing the strategy to a SINGLE BTC instrument both (a) discards the diversification/cash-gate thesis (becoming sparse naked trend) and (b) STILL produces impossible daily prints plus >100% avg exposure — the same multi-leg-basket + impossible-daily-print + MTM-inflated signature seen across the abandoned cross-sectional momentum, rotation, and nearness-to-high baskets. The momentum-basket family is not validatable on the current harness; the single-asset reductions are corrupted and edgeless.

Do NOT optimize yet — the equity curve is corrupted and the optimization robustness gates (Deflated Sharpe, PBO, Sharpe CI) are all computed from the daily-return series, so they would be garbage. Fix/verify, in order: (1) IMPOSSIBLE DAILY RETURNS: 2021-07-10 shows +247.2% and annualized_volatility is 841.7% — physically impossible for a long-only spot basket at 56.7% exposure (a held alt doubling at ~32% weight contributes ~+32%, not +247%). This is the SAME multi-instrument equity-curve aggregation artifact seen on the BTC-ETH-SOL basket, the cross-sectional momentum rotation, and the nearness-to-high LS this batch. Trace the per-bar portfolio-return computation, especially on REBALANCE days where positions are sold and bought — the recurring bug is double-counting realized PnL from the sold leg plus the new leg's mark-to-market. IMPORTANT: because this artifact has now appeared on four consecutive multi-instrument baskets across long-only/long-short/spot/futures, it is very likely an ENGINE/HARNESS-level bug in multi-instrument equity accounting, NOT something the strategy code can fix — if you confirm the strategy's own sizing/exposure is correct (log summed gross exposure each bar; confirm it stays <=97%), ESCALATE the daily-return aggregation bug rather than patching the strategy, because no multi-instrument basket can be validated until it is fixed. (2) TRADE SPARSITY: only 22 trades over 8.8 years (77-day avg hold) despite a weekly-rotation design — log the held set each rebalance and confirm whether it actually turns over or is stuck (e.g. the abs-mom gate keeping it in cash, or the top-3 being stable). 22 trades is thin for a 3-window walk-forward; if it's genuinely a low-frequency macro strategy that's acceptable, but verify it's not a rotation bug. (3) After both are resolved, re-run and confirm: max single-day return is physically plausible (a <=97%-exposure majors basket should rarely exceed ~+25-35%/day), annualized_volatility falls to ~60-100%, and the Sharpe is recomputed on the corrected curve BEFORE optimization. The momentum-rotation mechanism is the right (validated) direction; the blocker is the corrupted multi-instrument metrics.