ATOM 4H Supertrend Trend-Following Long (10, 3.0)

Hypothesis

Net loser with a catastrophic drawdown — not worth optimizing. Over 6.6 years and 159 trades the strategy loses money (total_return -12.8%, profit_factor 0.88, expectancy -$129/trade, win_rate 9.4%) and its max_drawdown is 50.75% with CI to 96.8% — past the hard-abandon threshold regardless of headline (L19). Sharpe 0.28 has a CI straddling zero, so there is no significant risk-adjusted edge, and the equity path is carried by a handful of outlier up-days (kurtosis 20; +61.5% on 2021-08-26, +43.9% on 2023-11-14) with the strong 2026 being two big months against an otherwise losing record. This is the single-name OHLCV trend-follower zero-survivor class, and the ATOM Parabolic-SAR sibling on the same symbol was already abandoned for the identical net-loss/>50%-DD profile; the 'approved DOT Supertrend' framing does not transfer because on ATOM the same mechanism has negative expectancy. Tuning the ATR period/multiplier cannot flip a PF<1, 50%+-drawdown trend-follower positive — the optimizer would only curve-fit the outlier rallies and fail deflated-Sharpe/holdout.

Implementation

Long-only ATOMUSDT.BINANCE USD-M perp 4H canonical Supertrend(10, 3.0) trend-follower, pure OHLCV. Wilder ATR(10) builds bands at hl2 ± 3.0·ATR; the active band ratchets (lower band rises in uptrends, upper band falls in downtrends) and direction flips when close crosses it (strict inequalities). It enters long when Supertrend flips short→long while flat, and exits when it flips long→short or a 10% hard stop triggers. Sized to 90% of equity, leverage 1.0. Two economic parameters (atr_period, atr_multiplier) plus a stop and a sizing knob; zero supplementary-data dependency.

Verification Results

CLEAN RESTART 2026-09-04 — this run's verdict history and learning records were removed and it was restarted from verification. Its previous abandonment came from the pipeline, not from the market: the Layer-2 harness mis-bound @staticmethod helpers (fixed), QA issued terminal performance verdicts on an unoptimized smoke test (removed — QA now judges correctness only), and sandbox timeouts came from backtest-slot starvation (fixed). The hypothesis and the strategy code are unchanged. Verify the code on its merits; performance is decided later by the full backtest and the optimizer.

Confirm ATOM catalog price/size precision and volume are correct first. If correct, add an ADV-relative notional cap to position_size; if not, the capacity/impact numbers are a data artifact and no sizing change is warranted.

position_size commits 90% of equity to a single alt perp per trade. Code is correct (equity*fraction/price, no leverage multiply, never requests >90%; the 101.6% avg_position_pct is a mark-to-market artifact of the drawdown path vs shrinking equity). But that concentration feeds the extreme impact 397% / capacity $6,326. If ATOM volume metadata is sound (low capacity real), cap notional to a fraction of rolling ADV; if broken (likely given the prior Layer-3 precision mismatch), fix the instrument data instead.

Backtest Review

Sharpe

0.28

Total return

-12.79%

Max drawdown

50.75%

Trades

159

Win rate

9.4%

Profit factor

0.88

Deep clean data (14,397 bars, 6.6 years), zero supplementary dependency; clean execution (159 signaled, 159 submitted, no drops)

avg_trade_return_pct 0.34% clears the Binance USD-M fee floor; strong recent 2026 (+46%)

Net loser over full history: total_return -12.8%, profit_factor 0.88, expectancy -$129/trade

max_drawdown 50.75% with CI to 96.8% — past the 50% hard-abandon line (L19)

Sharpe 0.279 with CI [-0.328, 0.916] straddling zero — no significant risk-adjusted edge

win_rate 9.4% and kurtosis 20 — equity carried by a few outlier up-days (2021-08-26 +61.5%, 2023-11-14 +43.9%); most months/years negative

Single-name OHLCV trend-follower — zero-survivor class; the ATOM Parabolic-SAR sibling was already abandoned with the same net-loss / >50%-DD profile

Iteration History

data_unavailable: backtest data for ATOMUSDT.BINANCE could not be collected after 5 retries: - bars (bar_type=ATOMUSDT.BINANCE-4-HOUR-LAST-EXTERNAL): got 0, needed 100, or 60+ days span The data layer (not the strategy) is the blocker — re-evaluate this hypothesis once collection is fixed / the symbol is backfilled.