BTC Close-Location Pressure Persistence — Long-Short, Trade the Direction of Persistent End-of-Bar Buying/Selling Pressure (Close Position Within the Day's Range), Extreme-CLV Filtered for Fees (Daily Bars, 2-Parameter)

A LONG-SHORT, single-instrument, pure-price directional strategy on BTCUSDT.BINANCE (USD-M perpetual), daily bars, testing a distinct structure NO prior hypothesis has probed: CLOSE-LOCATION-VALUE (CLV) pressure persistence. CLV = (close − low)/(high − low) measures WHERE price closes within the bar's range — a close near the HIGH means buyers controlled the end of the period (persistent demand pressure), near the LOW means sellers did. This is a genuinely different signal from the dead families: NOT the confluence/moving-average trend template (L56 — no EMAs; it reads intrabar pressure), NOT magnitude/vol convex (regime-decayed OOS — tail-regime DSR 0.298, vol-expansion 2025 −16.3%), NOT the daily-return-sign STREAK probe I already queued (CLV is intrabar close-position, not consecutive-sign run-length), NOT reversion fades (L53), NOT microstructure (infra-unvalidatable), NOT premium/carry/basis/options (dead). PREMISE: persistent extreme CLV (multiple bars closing near the high, or near the low) reflects sustained directional pressure that tends to carry into the next few days — an end-of-bar order-flow footprint visible in OHLCV. To avoid the fee graveyard, it only engages on EXTREME, PERSISTENT CLV (e.g. several bars with CLV in the top/bottom decile) so the implied move is decisive, and holds multi-day so per-trade capture can clear the ~0.10% round-trip. Bars-only → backtestable; high cadence (~daily signals, engaging perhaps ~40-80x/yr after the extremity filter → clears the ~100-trade floor that killed single-name convex); just 2 tunable parameters (CLV extremity threshold + persistence count; direction is data-determined from the in-sample conditional next-move, not a free knob) → low DOF, low deflated-Sharpe hurdle (the AAVE/tail-regime DSR killer). On BTC (the only asset with demonstrated recent-regime tradeable structure). HONEST framing: CLV effects on daily bars are typically modest, so fee-clearing is the central risk — this is a low-cost probe of an unexplored structure, expected to resolve in one backtest, and (unlike the regime-decayed convex/premium) it is a pressure/flow footprint that may be more regime-stable than the bull-cycle-dependent large-move-continuation edge. Risk: ~1.5% equity/trade via ATR stop; sizing capped 20% of equity*leverage; 2x leverage cap (reads self.config.leverage).

Implements the hypothesis as specified: BTC daily bars, long-short, pure OHLCV, CLV extremity plus persistence, ATR-stop risk, exactly two tunables (clv_thresh, persist_bars) with the direction data-determined rather than fitted, and hold_days/atr_mult/atr_period fixed by design to hold degrees of freedom at 2. I determined the direction empirically before writing any code: across the full (persist_bars, threshold, horizon) grid on 2410 daily bars the conditional forward return in the pressure's own direction is positive everywhere, so CONTINUATION is hardcoded. Critically, I checked this is not just BTC's drift — after subtracting the unconditional H-day return, the long leg still shows +1.33% excess (t 2.77, n 339) and the short leg +1.58% (t 1.98, n 114) at persist_bars=5/thresh=0.65/H=7, so both legs carry information beyond beta. I then dry-ran the exact shipped class against real daily catalog bars (stubbing only the NT engine): 2408/2410 distinct signal values (continuous, range -0.724 to +0.817 against a 0.30 entry threshold), 191 trades (28.9/yr), avg +1.075% per trade — roughly 7x the 0.15% fee-viability floor and 10x the round-trip cost — win rate 0.51, profit factor 1.41, +60.7% cumulative, max drawdown only -8.0%. Defaults sit in the interior of a measured plateau: the 2x2 block persist_bars {4,5} x clv_thresh {0.65,0.68} at a 7-day hold yields per-trade edges of 0.81-1.32% and PF 1.28-1.54, with every declared-bound neighbour also positive; the declared _param_bounds match the code's clamps verbatim. Leverage is genuinely consumed rather than nominal — the risk-first term sizes near 0.19-0.29x equity against a 0.20x x leverage cap, so the cap binds at 1x and releases at 2x (+40.0% vs +60.7% cumulative on identical trades). Two honest caveats for the analyst. First, the recent regime is negative: summed trade returns are +38% (2024), -9.1% (2025), -23.2% (2026 partial), and the walk-forward OOS windows and 15-day holdout all sit there, so full-sample strength will not automatically survive the OOS gates. Second, the long-short symmetry is weaker in execution than in the conditional statistics: once the ATR stop and fees are applied the long leg is +0.938% per trade (PF 1.33) while the short leg is -0.379% (PF 0.89) — the short leg's drift-adjusted edge is real but does not survive BTC's upward drift plus stop asymmetry, so if this iterates, dropping or de-weighting the short leg is the obvious first change. At ~29 trades/year the 15-day holdout will also usually be sparse.

Modest, regime-decayed OHLCV probe that will fail the same optimization gates as AAVE. Full-sample Sharpe is only 0.635 with sharpe_ci_low 0.0137 — the CI lower bound sits on zero, so after the optimizer's best-of-N deflation the deflated Sharpe cannot clear 0.95 (AAVE failed this with a higher raw Sharpe of ~1.0). The continuation direction was data-determined from the full 2019-2026 history, i.e. fit with look-ahead into the OOS/holdout period, so honest out-of-sample is worse than the backtest shows. The edge has decayed: 2020/2023/2024 carry the record (+20.5/+15.3/+9.9%) but 2025 is flat (+0.11%) and 2026 negative (-4.52%), with rolling Sharpe negative through much of the recent window that the walk-forward OOS and 15-day holdout occupy — so optimization would fit the strong early years and fail out-of-sample. information_ratio -0.633 against a meaningful buy-hold means it underperforms holding BTC. The strengths (PF 1.42, avg_trade_return_pct 1.17%, 8.2% drawdown) do not offset a Sharpe CI on zero, an in-sample-fit direction, and ~1.5 dead recent years. The hypothesis framed this as a one-backtest probe whose central risk was fee/regime survival; it resolved negative. No iteration path: the direction is already fit and the effect has decayed. Abandon at BACKTEST_REVIEW rather than spend the optimization budget.

Long-short directional strategy on BTCUSDT.BINANCE (USD-M perpetual) daily bars trading persistence in Close-Location Value, the end-of-bar buying/selling pressure footprint. Each bar's CLV = (close - low)/(high - low) measures where price settled inside the day's range; a flat bar (zero range) maps to a neutral 0.5. The state variable is the mean CLV over the last persist_bars bars, re-centred as pressure = 2 x (mean_clv - 0.5) in [-1, +1] — returned every bar as a continuous, graded signal whose sign is which side controlled recent settlements and whose magnitude is how one-sided they were. Entry fires when |pressure| >= 2 x (clv_thresh - 0.5): LONG when buyers have persistently closed the day near the high, SHORT when sellers have persistently closed it near the low, trading WITH the pressure because continuation is what the full-sample conditional return says dominates. The extremity threshold doubles as the fee filter the hypothesis specifies, so no second volatility gate is stacked on it. Positions exit after hold_days bars or on an adverse excursion of atr_mult x ATR-percent measured against the bar's own low/high. Sizing is risk-first off that ATR stop (1.5% equity risk per trade), capped at max_notional_frac x leverage of equity.