Expiry day, the first live session, and what a minute-by-minute replay showed

For an overnight buyer, yes — the rule "nearest expiry with at least five days left" skips today's contract and buys next week's, so the position never holds an expiring option. For an intraday gap-and-go buyer, the 0-DTE contract is the normal instrument and the backtest includes every expiry day. A separate study of expiry days and the days before them (2020–26) found no effect on either system's results.

When the index reaches entry plus 2.33 times the entry box — an index level, not a rupee amount. This morning's box was wide (145 points, 0.62% of NIFTY), so the target sat 339 points above entry even though the premium was already up 73 points. The second lot's stop was the last pivot low, about 400 points below the market, and it ratchets only when a new pivot low confirms, which needs five 10-minute bars on each side. Roughly a third of signals reach the 2.33× target, usually over one to three sessions. A gap-up continuation morning is also the kind of day where the move is often spent at the open, so an unrealised gain at noon is a number, not yet money.

Because the average is carried entirely by 2020. Run as a proper engine-level rule, skipping entries on strong-day-plus-gap mornings gained ₹1.03L in 2020 and lost money in every period since mid-2023, cutting the recent window by 17%. Removing entries removes the tail you live on. Same lesson as every filter before it.

Everything that lets you replay the day against the model later: the bid and ask at the moment of every trigger, the fill price and broker order reference, the index level at the trigger, the signal-minute price (slippage), and a venue flag so paper and real never mix. Separately, quotes every minute on the contracts you trade and their neighbours — half-spread, depth at the touch and at the size you intend to trade later, discount to parity. The decision to scale is made on those numbers, not on the P&L.

When the NIFTY index prints the target level — entry plus 2.33 times the morning box — not when the rupee figure looks nice. At noon the index needed about 180 more points, three quarters of a percent. The bot reads the index every twenty seconds and sells one lot at market, with a price cap, within the loop after the touch. There is no clock on it: the target can come this afternoon or in two sessions, since the position holds overnight; about a third of these signals get there, usually within one to three sessions, and the rest leave on the trailing pivot stop. Selling a lot by hand to "lock" the gain is the one thing not to do — the system's return comes from the exact rule, and the big days are the ones where the second lot keeps going.

Nothing by the clock. The only thing that locks profit is the trailing stop, and it sits at the last confirmed pivot low. At entry that pivot was from the previous afternoon, 326 points below entry, so for the first three hours a full reversal would have been a loss. A new pivot low needs a dip and then five 10-minute bars with higher lows; the midday dip did exactly that, and at 12:06 the stop ratcheted to a level 93 points above entry. From then on a reversal would have exited in profit — and thirteen minutes later the index came within 18 points of that stop before turning up. At 15:26 a second pivot low lifted the stop again, to 138 points above entry. That staircase is the whole mechanism: the floor rises only when the market builds a new step, never because time passed or because the P&L reached a round number.

Both bots acted in the first loop after their rule fired, and every order filled in the same second it was placed. Market-protected market orders arrive at the exchange as limit orders capped two percent away; the cap was never the fill. The overnight buyer paid 0.55 above the quoted ask on its entry (0.2% of the premium), with enough depth at the touch for five lots at that moment. The intraday buyer filled 0.35 better than the ask going in and 0.6 under the bid coming out — its exit printed within half a point of the minute's low, which is what a market sell into a fast dip looks like. Measured against the quote the bot saw, the day's execution cost was about ₹0.34 per unit of premium at this size.

Against the backtest convention (fill at the next minute's close plus a tenth of a percent) the overnight entry was 17–21 points better, because the bot acts at the top of the minute while the replay waits for a bar to close; on a reversing morning that same habit cuts the other way, so it is a timing difference to track, not skill. The intraday trade cost about ₹9 per unit more than its replay — a third on the entry, two thirds on the exit.

The intraday trade itself: the gap was +0.85% on the 15:15 basis the bot uses; the opening-range break was decided by the last ticks of the 09:51 minute (the bot's close was one point over the trigger; the final candle later printed under it). The option peaked 17% above entry at 10:35, the 20%-from-peak trail sat just under the 11:15 dip, and the 11:31 close tripped it. After the exit the expiring contract went on to more than double by 15:14 — the trail did what it says, and the re-entry question stays on the December review, not in the code.

The closing auction was the day's real lesson. The index froze from 15:15 to 15:28 and the 15:29 auction print jumped 79 points, while options kept trading to 15:40 on a level about 70 points lower — the held call's last trade sat 11 below the intrinsic value the auction print implies. Two effects on a pivot system: the 15:26 pivot low was "confirmed" by flat artificial bars, and for three minutes the bot was reading a print that, had it sat beyond a stop or target, would have fired an exit priced off a different market. Today it fell 61 points short of the target. Under review for the weekend: ignore bars from 15:15 in pivot confirmation and take no stop or target decisions on index prints after 15:14 — a mechanical fix for a structural change in how the close is printed, not a strategy change.