BtcVwapTrendChandelierBreakerLS

BTC Volume-Weighted Trend (Rolling-VWAP Sign) with Chandelier Stop-to-Flat and a Drawdown Circuit-Breaker — Corner-Resistant Entry + Hard Drawdown Cap, the Two Fixes for This Session's Trend Deaths (BINANCE USD-M, Daily Bars, Long-Short, 3-Parameter)

A LONG-SHORT, single-instrument BTC trend model deliberately engineered to thread the FOUR distinct death modes this session has now demonstrated for trend sleeves, using a signal I have not used (a rolling VOLUME-WEIGHTED trend anchor rather than price-only EMAs/momentum): (1) OVERFIT few-trade CORNER — the efficiency-gate, OI-gate and Donchian breakouts died because a threshold parameter could be pushed to an extreme that collapsed the book to 3–9 trades; this entry is a rolling-VWAP SIGN cross that fires regardless of parameter values (price is always above or below its VWAP), so no setting can starve trade count into a corner. (2) DRAWDOWN hard-abandon — the always-in weekly MA sleeve died at 62% max-drawdown; this model is NOT always-in: a chandelier trailing stop takes it to FLAT to cap drawdown, and a portfolio drawdown circuit-breaker de-grosses on losses. (3) OUTLIER dependence — the weekly sleeve's record rested on one +156% week (kurtosis 587); vol-targeted daily sizing spreads risk across many bars so no single week dominates. (4) FEE_EDGE — daily cadence with multi-week holds means per-trade capture dwarfs the ~0.10% fee. It stays on the ONE proven asset+edge (BTC directional trend) with a DISTINCT construction (volume-weighted anchor + stop-to-flat) from every pending sleeve (momentum-confluence, always-in MACD, pullback, EMA-crossover, rotation, TSMOM basket). Pure OHLCV+volume, computationally trivial (a rolling VWAP + ATR). Avoids the graveyards: NOT reversion/pairs/funding/options (dead/broken), NOT a rarity-gated breakout (corner-collapses), NOT always-in-through-drawdowns (DD abandon), NOT a symbol respray (L56 — BTC, proven asset, distinct signal). Exactly 3 tunable parameters: VWAP lookback, chandelier ATR multiple, and the drawdown-breaker threshold.

Implements each of the hypothesis's four anti-death-mode requirements literally. (1) Corner-resistance: the entry is a rolling-VWAP SIGN cross, not a rarity threshold — should_enter maps signal>0 to BUY and signal<0 to SELL with no magnitude gate, so trade count cannot be starved by pushing a parameter to an extreme (the optimizer's only levers are lookback length, stop width and the breaker level, none of which can make entries rare). (2) Drawdown: the strategy exits to FLAT on the chandelier stop, and the equity-peak circuit-breaker de-grosses by 0.5x while drawdown exceeds dd_breaker_pct, re-grossing at half recovery — plus a hard gross cap of 1x equity independent of the equity path (lesson L15: no compounding sizing blowups). (3) Outlier dependence: vol-targeted daily sizing (30% annualized target off a 20-day realized vol) shrinks exposure in high-vol regimes so no single week can dominate the record. (4) Fees: daily cadence with multi-week holds and a 3x-ATR trailing stop implies per-trade capture far above the ~0.10% round-trip taker cost. Venue is BINANCE USD-M futures because the strategy goes short; leverage is left at 1.0 and never referenced in sizing (the risk cap binds well below 1x notional, so any leverage>1 would be inert). Signal is returned in natural ATR units (clipped only at an absurd +/-50 to guarantee finiteness), thresholded at zero in should_enter. Exactly three parameters are declared tunable via _param_bounds, matching the clamps applied in __init__; the remaining values are fixed structural constants.

Well-engineered risk controls, but no statistically significant edge to optimize. Base Sharpe is only 0.595 with sharpe_ci_low -0.042 — the CI straddles zero, the weakest of the session's trend candidates and back in the failed-breakout-sibling range; a base this weak cannot clear the deflated-Sharpe 0.95 gate after best-of-N selection. The corner-resistant VWAP-sign entry successfully prevents a few-trade collapse and the chandelier + drawdown-breaker deliver a genuinely low max_drawdown (13.1%), but neither manufactures significance from an edge indistinguishable from noise. The hypothesis's claim to fix outlier dependence is contradicted by its own record: 2021's entire +27.0% is a single trade (2021-01-21 +26.9%), and win_rate is 0.22 with losing 2022 (-6.9%) and 2025 (-6.5%). This is a weaker sibling of edges already in the 3-phase run — the BTC pullback (Sharpe 0.86, CI-low 0.29) and diversified TSMOM basket (0.82, 0.19) — and it sits below the MACD always-in sibling (0.70, CI-low +0.09) that was the marginal accept. Sending a fourth, CI-on-zero variant of the same BTC-trend edge is spraying; the stronger constructions already carry it forward. Abandon at BACKTEST_REVIEW rather than spend the optimization budget. Failure pattern: vwap_trend_ci_on_zero (weak base Sharpe + negative CI-low, redundant with stronger queued siblings).

Long-short BTCUSDT.BINANCE daily trend model anchored on a rolling VOLUME-WEIGHTED average price. Signal = (close - rolling VWAP) / ATR, a continuous ATR-normalized distance computed every bar; its sign is the trend direction. Entry is the VWAP sign (long above, short below), so a tradeable state exists in every regime and no parameter setting can collapse trade count. The book is NOT always-in: a chandelier trailing stop (peak-since-entry minus chandelier_atr_mult x ATR for longs, mirrored for shorts) takes it to FLAT, and a VWAP sign flip also exits. After a stop-to-flat, the same direction is re-armed only once the signal crosses to the other side of VWAP, preventing immediate re-entry churn. Sizing is the minimum of (a) a fixed 2% of equity risked against the chandelier stop distance, (b) a 30% annualized vol target using a 20-day realized-vol estimate, and (c) a 1x gross-notional cap; a drawdown circuit-breaker halves both risk and gross whenever equity is more than dd_breaker_pct below its peak, restoring full size once half the drawdown is recovered. All state is incremental/bounded (deques, Wilder ATR), pure OHLCV+volume, no supplementary data.

At BACKTEST_REVIEW confirm the FULL-history trade count is >=100 and the full-history per-trade return clears the fee floor. If the full history comes in under ~100 trades, treat this like the failed daily-single-asset class and abandon (or lengthen to a finer timeframe), rather than sending an unmeasurable book into optimization.

Full-history trade count is UNCONFIRMED and is the decisive measurability item -- verify before optimizing. This is a DAILY single-asset sleeve, the exact configuration that has failed the ~100-trade floor three times this session (weekly MA 17 trades, HL daily 56, and the MACD/dual-EMA daily sleeves at 71/33-52 that were moved to 8H for this reason). Unlike every sibling this session, the developer did NOT provide a full-history dry-run (trade count + per-trade return); the only evidence is the sandbox's 21 trades over 362 days. That rate extrapolates to ~140 over the ~6.7-year history -- above the floor, and plausibly so because the chandelier stop-to-flat + re-arm generates ~2x the round trips of a pure sign-flip model (which is why daily can be measurable here where it was not for the pure-sign siblings). But it is an extrapolation, and the conspicuous omission of the dry-run every other sleeve supplied is a mild red flag.

Judge on the vol-targeted equity path / Sharpe over the full history rather than win rate; confirm the edge is not confined to one regime.

Right-tail dependence and thin recent-only evidence (BACKTEST_REVIEW judgment, not a block). Win rate is 24% with avg_win 2834 vs avg_loss 580 and kurtosis 11.9 -- the sleeve depends entirely on the trend right tail (by design for a stop-to-flat trend model), so any per-trade or win-rate floor applied without regard to payoff asymmetry will misjudge it. The sandbox (recent ~year) is positive (+4.9%, +0.66%/trade, Sharpe 0.52) which is a good sign, but no full-history per-year breakdown was provided, so recent-regime robustness is unverified.

Genuine risk engineering: chandelier stop-to-flat + drawdown circuit-breaker deliver max_drawdown 13.1% (CI to 28.8%); corner-resistant VWAP-sign entry (no gate) avoids the few-trade collapse that killed the breakout siblings

avg_trade_return_pct 1.58% clears the fee floor ~10x; profit_factor 1.40, capacity $257M, impact 2.0%, uncorrelated (beta 0.038)

sharpe_ci_low -0.042 — the CI straddles zero; base Sharpe 0.595 is the lowest of the session's trend candidates and cannot clear the deflated-Sharpe gate after best-of-N (same profile as the failed breakout siblings)

Outlier-concentrated despite the hypothesis's claim to fix it: 2021's entire +27.0% is one trade (2021-01-21 +26.9%); win_rate 0.22

Loses in 2 of 7 years (2022 -6.9%, 2025 -6.5%); more moving parts (stop, breaker, re-arm guard) than the cleaner pullback/MACD for a lower Sharpe

Weaker sibling of edges already in optimization: pullback (Sharpe 0.86, CI-low 0.29) and TSMOM basket (0.82, 0.19); MACD always-in (0.70, +0.09) was the marginal accept — this is below that line

BtcVwapTrendChandelierBreakerLS was a thoughtful engineering response to the session's four distinct trend deaths: a volume-weighted VWAP-sign entry that cannot be starved into an overfit corner, a chandelier stop-to-flat plus a drawdown circuit-breaker to escape the 62%-drawdown fate, and vol-targeted daily sizing to defuse outlier dependence. The controls demonstrably worked — max drawdown just 13.1%, 172 trades, PF 1.40, 1.58% per trade — but the edge underneath did not: base Sharpe 0.595 with a CI straddling zero (-0.042), a 0.22 win rate, two losing years, and 2021's whole return from a single trade despite the outlier-fix claim. The analyst abandoned it at backtest review as vwap_trend_ci_on_zero: sound risk design cannot manufacture significance, and it is a weaker, redundant sibling of the pullback, TSMOM, and MACD trend edges already in the 3-phase run. It never reached optimization, analysis, or risk review.

Excellent risk engineering (corner-resistant entry, stop-to-flat, drawdown breaker, vol targeting) can fix the failure modes of a trend edge — drawdown, corner-collapse, fee drag — but it cannot create statistical significance where the underlying edge is indistinguishable from noise; a base Sharpe with a CI on zero is a dead end, and a fourth CI-on-zero variant of an edge whose stronger constructions are already in optimization is redundant, not additive.

The analyst abandoned it at backtest review: the risk controls worked (low drawdown, corner-resistant entry that avoided the few-trade collapse) but manufactured no statistically significant edge — a base Sharpe of 0.595 with a CI straddling zero is the weakest of the session's trend candidates and cannot clear the deflated-Sharpe gate after best-of-N; its outlier-dependence claim was contradicted by 2021's single-trade year; and it is a weaker, redundant fourth variant of the same BTC-trend edge whose stronger siblings (pullback Sharpe 0.86, TSMOM basket 0.82, MACD always-in 0.70) already carry it forward — sending it on would be spraying.

A long-short, single-instrument BTC trend model on BTCUSDT.BINANCE USD-M daily bars, engineered against four demonstrated trend-death modes: entry on the sign of a rolling volume-weighted price anchor (VWAP) in ATR units — which fires in every regime so no threshold can collapse it into a few-trade overfit corner — with a chandelier trailing stop-to-flat plus an equity drawdown circuit-breaker to cap drawdown (not always-in), vol-targeted daily sizing to spread outlier risk, and daily cadence with multi-week holds to clear fees, on the one proven asset+edge with a distinct construction, using 3 tunable parameters.

The backtest (2411 daily bars, 2019-2026) returned +61.8% over 172 trades with profit factor 1.40, avg_trade_return_pct 1.58% (~10x the fee floor), and — validating its risk engineering — a low max drawdown of 13.1% (CI to 28.8%), uncorrelated (beta 0.038), $257M capacity. But base Sharpe was only 0.595 with sharpe_ci_low -0.042 (CI straddles zero), win rate just 0.22, it lost in 2 of 7 years (2022 -6.9%, 2025 -6.5%), and despite the outlier-fix claim 2021's entire +27.0% came from a single trade.