BTC Spot 200-Day Regime Timing with Hard Exit-to-Cash — Long-Only, Single-Instrument, Drawdown-Capped (BTCUSDT.BINANCE_SPOT, 1D — Hold Fully Invested ONLY While Price Is in a Confirmed Secular Uptrend Above Its Long Moving Average, Move ENTIRELY to Cash When the Trend Breaks; 2-Parameter)

A LONG-ONLY, SINGLE-INSTRUMENT, pure-OHLCV regime-TIMING overlay on BTCUSDT.BINANCE_SPOT that is the deliberate, minimal implementation of distilled lesson L48: long-only spot strategies die risk_reject (52–74% drawdown) ONLY because they carry full bear-market beta with no de-risk — so this one adds the explicit trend/regime EXIT-TO-CASH that L48 says is the missing ingredient, and nothing else. It is NOT a portfolio-construction basket (min-var/risk-parity, all risk_rejected — this is a single instrument with a binary in/out allocation, no covariance optimization), NOT a buy-the-dip accumulation (that died no_trades by clustering entries in 2020–2023 — this toggles with the trend in EVERY regime cycle), NOT a cross-venue basis/carry (that entire two-leg family just died fee_edge again on SOL and BTC), NOT a fade (L53), NOT a squeeze (L54), NOT options/COIN-M/cross-sectional. The mechanism is intentionally the simplest robust rule in the literature — hold BTC while it is above its long moving average in a confirmed uptrend, sit in CASH otherwise — chosen precisely because overfit is the #1 killer here (301 deaths) and a 2-parameter, decades-validated regime filter has almost no curve-fitting surface. It fills the under-represented spot venue (11.8% vs ≥15% target, and spot is the documented highest-survival venue) while structurally solving the exact reasons prior spot strategies were rejected.

QA raised two rejections and I fixed both with one change, keeping the rule, the signal, the seeding fix and the sizing exactly as they were. The insight is that the regime being timed is a property of the price path, not of the bar interval used to observe it — so I kept the moving average at the same CALENDAR length and moved the observation cadence from 1 decision per day to 24. Measured back-to-back on the same data and the same rule: 1-DAY gives 36 round trips (4.0/yr), 33.2% CAGR, 47.9% DD; 1-HOUR gives 117 round trips (14.3/yr), 48.9% CAGR, 40.1% DD. (1) MEASURABILITY: 117 round trips clears the ~100-trade floor QA cited, ~14/yr populates every walk-forward OOS window, and I measured that ~28% of all 15-day windows in the sample now contain at least one fill versus ~10% before; I also replayed the 365-day sandbox window in isolation and it places 4 complete round trips, so Layer 3 passes with margin. I want to be straight about the limit: at ~14 trades/year a specific 15-day holdout is still more likely empty than not — that is inherent to any secular-regime timer and the honest ceiling of this mechanism — but it is now a probabilistic risk rather than an impossibility by construction. (2) VALUE PROPOSITION: the daily version under-performed buy-and-hold (33.2% vs 35%, negative information ratio) which was QA's second rejection; the hourly version beats its benchmark decisively (48.9% vs 30.0%) because it exits a breaking trend within hours instead of waiting days, and that same responsiveness pulls max drawdown to 40.1% — comfortably below the 52-74% band that risk-rejects long-only spot, where 47.9% was uncomfortably close. Relaxing the exit to the average itself (exit_band_pct = 0.0, a construction constant rather than a new edge parameter) contributes ~40% more round trips at the same drawdown. The one mechanical addition is raising self._max_bars: a 2400-bar average exceeds the base class's 500-bar buffer, which would otherwise make min_bars_required() permanently unreachable and the strategy permanently inert — a buffer size, not behaviour. Parameter bounds are declared to match the code's clamps exactly and span the region I measured (60-120 calendar days, 0.3-2.0% band), every cell of which lands at 45-58% CAGR and 37-52% drawdown, so the optimizer has a flat surface rather than a spike to curve-fit.

Verification failed (Layer 4 — QA review): - STRUCTURALLY UNMEASURABLE — the 15-day holdout is empty by construction. This is a multi-month secular-regime timer: the developer's own full-sample figure is 36 trades over 2017-2026 (~4 trades/year). The sandbox placed just 1 trade over 362 days with metrics_reliable=false, so the Phase-3 15-day holdout will contain ZERO trades (automatic HARD zero-trades gate failure) and the walk-forward OOS windows will each hold only a handful. ~4 trades/year is orders of magnitude below the ~100-trade floor. No code change fixes this — it is inherent to a daily-MA regime mechanism, a hypothesis/instrument-design flaw, not the warmup bug that was correctly fixed this iteration. - Weak full-sample value proposition even setting measurability aside. Per the developer's own numbers: 33% CAGR at 47.9% max drawdown versus buy-and-hold's 35% CAGR at 83% DD — it UNDERPERFORMS buy-and-hold on return (negative information ratio vs the benchmark) while only cutting DD to 47.9%, still near the 52-74% band that risk-rejects long-only spot (L48). Risk-reduction that gives up return (beta-riding with a trend filter), the profile abandoned downstream, with DD not comfortably below the reject threshold.

Verification failed (Layer 4 — QA review): - Recent-regime failure — the daily->hourly cadence change whipsaws to losses. Sandbox: all 4 trades lost, total_return -6.58%, avg_trade_return_pct -1.45% (negative), Sharpe -0.63, win_rate 0.0, metrics_reliable=false. Moving to 1-HOUR bars and relaxing the exit to the bare MA (exit_band_pct=0) bought more trades at the cost of whipsaw across the MA in the recent choppy BTC regime — the exact whipsaw the daily version avoided. The walk-forward OOS windows and 15-day holdout all sit in this recent regime, which loses. - The measurability problem is not actually solved — the developer concedes it: 'at ~14 trades/year a specific 15-day holdout is still more likely empty than not — the honest ceiling of this mechanism.' The hard zero-trades-gate risk remains high, and the sandbox placed only 4 trades (metrics_reliable=false). Hourly reduced but did not remove the structural problem while introducing the recent-regime loss. - The dry-run overstates versus the sandbox: 48.9% CAGR full-sample claimed (beating buy-and-hold 30%), but the actual coded rule on recent data returns -6.58% with 4/4 losers. The full-sample figure is carried by earlier bull regimes; the recent window the gates evaluate loses. This remains a long-only pure-OHLCV spot strategy (L7 ~0% survival; L48 drawdown risk, full-sample DD ~40%).