The honest read

The technical core does not have a demonstrable edge here. A ~38% win rate against a roughly symmetric target and stop is what you’d expect if the entry timing were close to random. Structuring the trades as debit spreads instead was tested and trimmed the average loss but also capped the winners, for a slightly worse net — so the bot keeps it simple with single options. The equity curve is a volatile round-trip to roughly flat, with drawdowns that would end most real accounts.

Walk-forward, no lookahead. Recomputes only the technical-analysis directional score exactly as the live bot would, on a trailing ~1y window, sampling every 2nd trading day. Strike chosen by Black-Scholes delta-targeting with trailing realized volatility as an IV proxy; the option is repriced with Black-Scholes each day and the trade exits on an intraday target/stop touch (High/Low) or at expiration. Does NOT include the live bot's news-sentiment, ML, or market-regime layers - those depend on present-tense data with no clean historical replay on free data. Treat as a directional sanity check of the technical signal + option-structuring logic, not a forecast of live performance. P&L excludes commissions and slippage.

This omits the sentiment, ML, and regime layers, so the live bot applies more filters and trades less often — but a backtest this weak is a red flag, not a green light. Past simulated performance does not predict live results. Not financial advice.