Two months of paper against the backtests — one bot killed by slippage, one validated within 4%, and a look-ahead caught by its own absurdity

The expiry-day straddle: +₹5,718 over eight expiry days, ₹715 a day against a backtest of ₹689 — within 4%, the best live validation we own. The mirror seller: +₹13,171 over 35 paper trades, on its backtest path. The NIFTY and BankNifty overnight bots were small negatives in a chop regime, with BankNifty's short side working and its long side bleeding.

The daily 9:20 straddle: −₹12,054 over 43 paper days, killed by measured stop-loss slippage — 47 stop-outs averaged +2.98 points of slippage with a maximum of 17.14, about ₹195 a day of tax against a backtest edge of ₹120 a day. The paper phase did exactly its job: it measured the one number the backtest assumed, and the number was fatal. Daily straddle live-arming was switched off permanently; the expiry-only version survives because its edge per day is ten times larger.

It is a toll, not a shield. Over twelve years, recentring every 50 points lost ₹2.81L, every 100 points lost ₹50k, never adjusting made ₹7k, against +₹3.49L for the simple 25% leg stop. Adjustment shrinks the worst day (−₹42k to −₹25k) but pays for it with about five leg crossings a day. The asymmetry of a per-leg stop is the whole edge.

The first delta-neutral run showed +₹30.4L with a ₹9k maximum drawdown. Absurd on its face, and it was: decisions made on a ten-minute bar's close were being executed at that bar's open — a look-ahead through bar timestamps. The fix was a ten-minute execution delay; the result collapsed to the numbers above. Three habits caught it: an absurdity check, a pre-registered expectation, and a baseline to anchor against.