Faithful port of berlinguyinca's Freqtrade AdxSmas (ADX14>25 + SMA3/SMA6 cross, 10% ROI, -25% stop) on NEARUSDT spot 1h

Faithful port of the Freqtrade strategy 'AdxSmas' by Gert Wohlgemuth (berlinguyinca), GPL-3.0, itself converted from sthewissen's Mynt AdxSmas.cs. Credit to the author; any performance the source claims is unverified and our own backtest with real fees decides. The Freqtrade Python file is the spec, and its rules and defaults are kept unchanged: timeframe 1h, TA-Lib ADX(14), SMA(3) as 'short', SMA(6) as 'long'. ENTER LONG when ADX > 25 and short crosses above long. EXIT on signal when ADX < 25 and long crosses above short. minimal_roi {'0': 0.10} means a +10% take-profit at any trade age. stoploss = -0.25. Long-only (Freqtrade default can_short=False). The Mynt C# original uses SMA(3)/SMA(10) and an asymmetric cross test; the Python version is ported, not the C# one. Only these were adapted for our platform: (1) Venue and instrument: NEARUSDT.BINANCE_SPOT. The strategy is long-only and unleveraged, so it belongs on spot. NEAR was chosen because it has multi-year 1h spot history and little use in the corpus, which avoids the over-mined BTC/ETH and the LINK/DOT/ATOM series used in the latest Freqtrade ports. The source names no pair. (2) Execution: Freqtrade fills a signal at the next candle's open. On a 24/7 venue this becomes the registry equivalent, a market order at the signal bar's close. (3) The -25% stoploss and the +10% ROI exit are intrabar protective orders, as in Freqtrade backtesting, which checks both against candle low/high: a reduce-only stop-market at entry*0.75 and a take-profit at entry*1.10, both matched intrabar by the platform. Where one bar touches both, the engine's adaptive OHLC ordering applies, since same_bar_policy is not selectable. (4) Sizing: one position at a time (Freqtrade max_open_trades=1 with stake_amount 'unlimited'), with 95% of account equity per entry. The 5% cash buffer covers the 0.10% taker fee and rounding. No pyramiding: entry signals are ignored while a position is open. Fees: Binance spot 0.10% taker per side, ~0.20% round trip. Fee arithmetic is stated honestly: the SMA3/6 cross on 1h bars is a fast signal. A typical 1h NEAR trend leg after an ADX>25 cross is ~1.5-3%. Losing trades usually do not exit on the signal, because the exit also requires ADX<25, so they can run to the -25% stop. Wins cap at +10%. The expected average trade is therefore uncertain, and it may fall below the 0.25% spot floor. This is a faithful replication objective, not a predicted winner. Novelty (L202): no earlier run ports AdxSmas. Its distinguishing rule is the ADX-gated asymmetric entry/exit (trend strength >25 to enter, weakness <25 plus the reverse cross to exit). It differs from the failed LTC SMA crossovers (no ADX gate, no ROI/stop ladder) and from the Freqtrade EMA20+OBV and Supertrend ports.

negative_expectancy (fee-floor check: avg_trade_return_pct -0.467% vs the +0.25% BINANCE_SPOT floor). The raw fill-price return per trade is -0.014% before the ~0.20% round-trip fee, about -0.21% net before any impact, so fees are not the cause: the gross edge is about zero. Evidence: 745 long trades on NEARUSDT 1h spot (2020-10 to 2026-10), PF 0.80, win rate 43%, executed-signal IC -0.011. Mean gross per-trade return is negative in 2022, 2024 and 2025 and flat in 2023. QA's smoke-window concern (-0.57%/trade, PF 0.64) is borne out on the full history. Optimization cannot help: the contract fixes every parameter and the tunable set is empty. Scope: this rejects only the faithful Freqtrade AdxSmas port (ADX14>25, SMA3/6, +10% ROI, -25% stop, 95% equity) on NEARUSDT.BINANCE_SPOT 1h. It does not test the Mynt SMA3/10 original or other pairs. OPERATOR LEDGER NOTE (separate from the verdict): the account shows -100% / liquidated on 2021-12-18 on a 1x CASH account. Compounding about -0.21% net over 745 trades implies roughly -80%, not -100%. Trades also keep printing P&L through 2026 on a zero-equity curve (last trade qty 833, +$181). Other inconsistencies: avg_position_pct 131%, avg_holding_bars 0 against a 1d5h average hold, and $117k of folded impact against $65k of commission. The same pattern was flagged on BchOkxFmzSingleSuperTrend453 and FreqtradeSupertrendLinkSpot. It needs an engine/ledger review, but the price-level returns are negative regardless.

The Freqtrade Python file is the spec. ADX is an incremental, exact mirror of TA-Lib's ADX: Wilder DM/TR rules, (period-1)-bar seed sums, the first ADX as the mean of the first `period` DX values (first value on bar 2p-1), then Wilder smoothing. A zero TR or zero DI sum skips the update, as TA-Lib does. The SMAs come from a bounded deque. All indicator state is scalar attributes. calculate_signal returns the continuous SMA spread (% of SMA6). The entry/exit cross flags are computed on the same bar and read by should_enter/should_exit. Freqtrade's minimal_roi {0: 0.10} and stoploss -0.25 are the base-template bracket legs (stop_loss_pct/take_profit_pct). These mirror the frozen plan keys stoploss/roi_take_profit and carry the same values. should_exit implements only the signal exit, so nothing double-exits. Sizing is 95% of settled equity / close, and the base caps it to affordable cash including the taker fee. Fee concern noted: losing trades usually reach neither the ADX<25 reverse-cross exit nor the TP and can run to -25%, so the average trade may fall below the 0.25% spot floor. That is the faithful behaviour and is kept.