DogeTailRegimeConvexDirectionalLS

DOGE Binance USD-M Tail-Regime Convex Directional — Long-Short, Engage ONLY on Confirmed Large-Magnitude Directional Regimes (Fatter Tails = More Frequent Engagements), Flat Otherwise, Wide Chandelier Ride (Daily Bars, 2-Parameter)

A LONG-SHORT, single-instrument, pure-price CONVEX directional strategy on DOGEUSDT.BINANCE (USD-M perpetual), daily bars, engaging ONLY on confirmed large-magnitude directional regimes and flat the rest of the time. It applies the factory's ONE genuinely-validated edge — the BTC Tail-Regime Convex Directional (PF 2.03, decorrelated beta 0.009, avg_trade_return_pct 2.23% well above fees) — whose SOLE documented flaw was cadence (only 72 trades / 6.6y ≈ 11/yr, too few to validate). This is NOT a blind ticker-spray (the L56 warning targets the dead dual-TF momentum CONFLUENCE template, a different, 0/213 mechanism): it is a targeted, evidence-driven cadence fix that exploits a SPECIFIC structural property of DOGE — its extreme fat tails and frequent parabolic/cascade regimes produce large-magnitude directional moves roughly 2–3x more often than BTC, so the identical convex trigger fires ~30–45x/yr single-name (approaching validatable cadence) with even LARGER per-trade capture (DOGE regime legs routinely run 20–60%). It is a genuinely DIFFERENT construction from my two queued tail-regime relatives: distinct from the abandoned single-BTC version (different asset, chosen for cadence-via-frequency) and from the multi-major shared-parameter BASKET (single-name, DOGE-specific, no cross-asset aggregation). NOT a confluence clone (L56 — different, validated mechanism), NOT a mean-reversion fade (L53 — engages WITH the move), NOT a squeeze/compression breakout (L54 — needs a large EXISTING magnitude move, not a low-vol pop), NOT a channel breakout (structurally-unmeasurable, abandoned), NOT calendar/session (net-losing), NOT cross-venue carry/basis/liquidation (L57/L58), NOT a NON-PRICE-feed gate (L46), NOT options/COIN-M (L50/L51). Fills the long-short gap (13.5% vs 86% long-only) with a decorrelated, convex sleeve. Only 2 parameters (magnitude z-threshold, chandelier multiple) → low deflated-Sharpe hurdle. Risk profile: vol-scaled sizing (~2% equity risk anchored to the wide chandelier, downsizing into DOGE's volatility so no single bar dominates), capped at 20% of equity*leverage; 2x leverage cap (reads self.config.leverage). Symmetric engagement (both parabolic up-regimes and cascade down-regimes) so trades fire across the full sample.

Ports the factory's one validated convex mechanism (BTC tail-regime: PF 2.03, beta 0.009, +2.23%/trade, abandoned only for ~11 engagements/yr) to DOGE unchanged, with the tail window fixed at 4 days instead of 5 so distinct bursts are individuated rather than swallowed by one long engagement. Offline dry-run on the real DOGEUSDT daily catalog (2020-07 to 2026-08, 2,218 bars, net of 0.05%/side) at the submitted defaults: 93 engagements over 6.1y = 15.3/yr, avg +5.43% NET per trade (36x the 0.15% fee floor, and 2.4x the BTC version's per-trade capture), win rate 35% with the expected convex right tail, avg hold ~5 days, per-trade Sharpe proxy ~0.78, equity path x1.47 with a 5.0% max drawdown at the submitted sizing. Both tunables are cliff-free across the declared _param_bounds: entry_z 0.75/0.90/1.00/1.10/1.25 gives 21.4/17.3/15.3/14.2/12.9 engagements per year at +3.8/4.6/5.4/5.7/5.4% per trade (Sharpe proxy 0.70-0.80), and chandelier_atr 4-8 leaves per-trade capture flat (5.4-5.6%) while trading drawdown against return, so the defaults sit inside the plateau. TWO honest corrections to the hypothesis. (1) CADENCE: DOGE fires ~15/yr at a comparable threshold and ~21/yr at the loosest end of the bounds, NOT the 30-45/yr the hypothesis projects -- better than BTC's 11/yr (93 engagements over the sample, ~15 inside a 365-day sandbox window) but the hypothesis's frequency estimate does not hold, and reaching 25/yr requires a 3-day window whose per-trade capture halves. (2) LEVERAGE: the chandelier-anchored risk sizing puts only ~9% of equity of notional on per engagement (DOGE's daily ATR runs ~8% of price, so a 6-ATR stop is a ~48% price distance), which is far below 1x equity notional -- a declared leverage of 2.0 could never bind and the backtest would be byte-identical to 1.0, the textbook leverage_set_but_unused failure. Leverage is therefore left at 1.0 and risk_pct set to 0.04, which delivers exactly the notional the hypothesis intended (equity * 2% risk * 2x). Per-year attribution is honest about regime dependence: 2021 +21%, 2022 +8%, 2024 +11% carry the sample while 2023 -1%, 2025 -3% and 2026 +2% are flat-to-slightly-negative.

Outlier-driven, non-generalizing convex directional on DOGE (fat-tail outlier class, L10/L41). The strategy's entire justification — a cadence fix to ~30-45 engagements/yr — is falsified by its own backtest: only 89 trades over ~6y (~14.7/yr), still below the ~100-trade floor and barely above the BTC version it was meant to fix. The +45% total return is a mirage from a handful of extreme days (return_kurtosis 322, skew 11.9; 2021-04-17 +15.1% and 2022-11-03 +10.6% dominate). Sharpe is only 0.565 with a CI [-0.093, 1.053] that straddles zero, and information_ratio -0.695 against a meaningful buy-hold benchmark means it UNDERPERFORMS holding DOGE risk-adjusted — captured parabolic beta, not decorrelated alpha. Losing 2023 and 2025 shows no persistence. Optimization would best-of-N overfit this fat-tailed noise and collapse in walk-forward OOS and the 15-day holdout. No iteration path: the tail-regime trigger cannot be made to fire more often than the data allows, and the cadence premise is already disproven. Abandon at BACKTEST_REVIEW rather than spend the optimization budget.

Long-short single-instrument convex tail-regime strategy on DOGEUSDT.BINANCE (USD-M perpetual), DAILY bars. Flat by default: it engages only when DOGE is already inside a confirmed large-magnitude directional regime. Every bar it computes z = log(C[t]/C[t-4]) / (sigma_d * sqrt(4)) where sigma_d is the 60-day stdev of daily log returns, and returns that z as a continuous signal. Engagement requires |z| >= entry_z AND expanding volatility (20-day stdev above the 60-day baseline); direction is the sign of z, so both parabolic up-regimes and cascade down-regimes are traded. Exit on z flipping sign (regime over) or a wide chandelier giveback of chandelier_atr x ATR(14) from the best close since engagement, so a genuine blow-off is ridden rather than scalped. Sizing is risk-anchored to the chandelier distance (qty = equity * risk_pct / (chandelier_atr * ATR)), which downsizes automatically into DOGE's volatility, capped at max_notional_frac * equity * leverage; both terms are fractions of CURRENT equity so losses never compound into position size. Exactly two tunables: entry_z and chandelier_atr.

Analyst/optimizer: evaluate the holdout on pooled full-sample / walk-forward-OOS statistics rather than the 15-day window (as for the BTC iter-3 and macro-TSMOM siblings); note the optimizer can push entry_z DOWN to 0.75 to lift the trade count above 100 if a stricter measurability bar is required, at the cost of ~30% lower per-trade capture.

The hypothesis's core cadence premise is FALSIFIED (disclosed by the developer), and measurability sits at/near the floor with a likely-empty 15-day holdout — the deciding measurement call belongs to the analyst. The hypothesis projected DOGE's fatter tails firing the trigger ~30-45x/yr (2-3x BTC) to reach validatable cadence; the actual dry-run is ~15/yr (93 engagements over 6.1y), better than BTC's 11/yr but far short of the claim. 93 total is above the BTC iter-3 (74) I passed, and at the loosest entry_z bound (0.75 -> ~21/yr, ~130 total) the optimizer can exceed the ~100 floor — so it is measurable-enough by that precedent — but the 15-day holdout still expects only ~0.6 trades and will likely be empty. The mechanism and construction faithfully match the hypothesis; only the projected frequency is over-claimed, so this is a rationale over-projection, not a code-vs-hypothesis mismatch.

Backtest_review/analyst: weight the recent regime and confirm the long-leg convex engagements contribute positively across the pooled sample; judge on the OOS windows rather than the noisy single sandbox year.

Recent-regime dependence and a small-sample long-leg weakness — analyst considerations, not code defects. The developer's per-year attribution shows 2021 (+21%), 2022 (+8%) and 2024 (+11%) carry the sample while 2023 (-1%), 2025 (-3%) and 2026 (+2%) are flat-to-slightly-negative, so the recent-window OOS is the main risk. In the sandbox the LONG leg was weak (3 longs, win 0.0) while SHORTS carried it (6 shorts, win 0.83) and the sandbox Sharpe CI [-1.21, 2.17] straddles zero (PSR 0.76, only 9 trades, kurtosis 35), so the single sandbox year is noisy; the developer reports the full sample is positive across the convex right tail.

For live deployment, reconstruct _side/_extreme/_entry_atr from cache.positions_open() rather than the z sign.

should_exit() infers _side from the live z sign on restart (_side==0) and re-seeds _extreme/_entry_atr from current values. Unreachable in backtest; only a live mid-position crash-restart risk, where it could pick the wrong side or reset the chandelier reference.

avg_trade_return_pct 5.34% is above fees

profit_factor 2.17 and low max_drawdown 6.0%

Low beta (0.023) headline

Core premise FALSIFIED: 89 trades / ~6y = ~14.7/yr, not the claimed 30-45/yr — the cadence fix did not materialize and the sample is still below the ~100-trade measurability floor

Outlier-driven: return_kurtosis 322, skew 11.9; two days (2021-04-17 +15.1%, 2022-11-03 +10.6%) carry nearly the entire return

sharpe_ratio 0.565 with sharpe_ci_low -0.093 — CI straddles zero, not distinguishable from no-skill

information_ratio -0.695 vs a meaningful buy-hold benchmark — underperforms holding DOGE risk-adjusted; the convexity is captured beta, not alpha

Not persistent: losing 2023 (-1.4%) and 2025 (-2.7%)

Optimizing on 89 trades with kurtosis 322 will best-of-N overfit to a few outlier days and collapse in walk-forward OOS/holdout

DogeTailRegimeConvexDirectionalLS ported the factory's one validated convex edge to DOGE, betting its fatter, more frequent tails would fire the identical trigger 30-45 times a year and fix the BTC version's sole flaw of too few trades. The economics per trade were strong (PF 2.17, 5.34% per trade, 6% drawdown), but the cadence fix simply did not materialize — 89 trades over six years (~15/yr), a Sharpe of 0.565 with a CI straddling zero, kurtosis 322 with two days carrying the return, and a -0.695 information ratio underperforming buy-and-hold DOGE. The analyst abandoned it at backtest review as an outlier-driven, non-generalizing convex bet whose central premise was falsified by its own trade count; it never reached optimization, analysis, or risk review.

Choosing a fatter-tailed asset to raise a rare-event trigger's cadence does not deliver more measurable trades — DOGE fired only ~15 times/year, and its extreme tails made the result even more outlier-driven (kurtosis 322) and beta-captured than the BTC version, so the cadence premise was disproven by its own backtest.

The analyst abandoned it at backtest review: the strategy's entire justification — a cadence fix to ~30-45 engagements/year — was falsified by its own 89-trade (~14.7/yr) result, still below the measurability floor and barely above the BTC version it was meant to fix. The +45% is an outlier mirage (kurtosis 322, two dominant days), Sharpe CI straddles zero, and it underperforms holding DOGE risk-adjusted — captured parabolic beta, not decorrelated alpha, with no iteration path since the trigger cannot fire more often than the data allows.

A long-short, single-instrument convex directional strategy on DOGEUSDT.BINANCE USD-M daily bars (2 parameters) that stayed flat by default and engaged only on confirmed large-magnitude directional regimes (|z| ≥ threshold with expanding volatility), riding parabolic bursts and cascades with a wide chandelier stop — porting the factory's one validated convex edge (BTC tail-regime) to DOGE specifically to fix that strategy's low trade cadence via DOGE's fatter, more frequent tails.

The backtest (DOGEUSDT.BINANCE 1D, 2217 data days from 2020-07) returned +45% with strong per-trade economics (profit factor 2.17, avg_trade_return_pct 5.34%, max drawdown 6.0%). But the cadence fix failed — only 89 trades (~14.7/yr, not the claimed 30-45), Sharpe 0.565 with a CI straddling zero (-0.093), catastrophic kurtosis 322 and skew 11.9 (two days carry nearly the entire return), and information ratio -0.695 versus holding DOGE, with losing years in 2023 and 2025.