The Williams buyer on its own, a corporate-action trap in the spot series, and 230,000 stock-option trades that say no

Standalone it makes 1,691 trades since 2015 at ₹197 per trade, profit-to-drawdown 4.42 and 11 of 12 years positive, so it is a real but thin edge. Its best route — a gap that fills and then breaks the other way — earns ₹575 per trade and is positive in 10 of 12 years, against ₹60 and ₹81 for the two plain routes; puts earn ₹222 against ₹167 for calls. Its one tuning parameter sits on a plateau around 0.5, and the whole leg has been flat since 2023, which is the pattern everything in this book shows.

The spot series from the charting feed is adjusted for corporate actions, while the option vendor's strikes are the ones that actually traded, and on 20% of trade-days the two disagreed about where the money was. Mismatched longs showed ₹3,597 per trade against ₹930 on clean days, a fictitious gain. The fix is to back the spot out of put-call parity on the option prices themselves, and once that was in place a look-ahead in the Williams-plus-filter variant also surfaced: its ₹443 per trade became ₹21 when the filter was only allowed to see what was known at entry.

Two measurements that know nothing of each other agree. A Roll effective-spread estimator on the historical premiums gives a median half-spread of 0.52%, and a live sampler quoting 12 large caps on its first day measured 0.47%. The door-plus-filter stock variant earned ₹790 per trade but on 110 days with a day-clustered t of 1.1, which is not evidence, and the same filter idea applied to the live NIFTY bot raised the per-trade figure by 5% while cutting net profit by 19% once the spread was charged, so it was not adopted.

All twelve were negative on 113,640 real-premium trades. Fading a 5-EMA stretch lost ₹511 per trade with a 28% win rate; VWAP reversion lost ₹602, because a stock 1.5% from VWAP tends to keep going; buying event straddles lost ₹2,676 per trade; selling after events had a worst day of about −₹182k; expiry-day stock straddles carry a 4.8% half-spread; a daily strangle earns ₹124 only at zero spread, and the real spread is two to five times the theta it collects. The one hypothesis that survives is a gap-down fade with calls at +₹1,505 over 104 days, t 1.3, filed rather than traded. Across roughly 230,000 real-premium trades, stock options are not the next edge.

No, and the reason is instructive. The window happened to contain 53% up days, yet the door's single best cell is calls in a bear regime at ₹1,729 per trade, and the day-ORB variant is structurally put-leaning at ₹998 per trade and 10 of 12 years. Every put-biased policy tested made the book worse, cutting profit-to-drawdown from 4.44 to 3.32, and "puts only" would have kept just 1 of the live book's 6 trades. Doubling put size when the index is below its 20-day average goes on the hypothesis list, not into the code.

Not one we could find. Forcing an intraday buy on every session costs ₹114–126 per day per lot; a call seller on quiet days comes to about +₹21 a day after slippage and dilutes the door rather than adding to it, and its first coding contained a look-ahead that had to be removed. About 40% of sessions have no trade, and on those days no trade is the correct trade.