Posted in My Daily Journal · 28 October 2026
A day of end-to-end replays — the mirror seller and the pivot futures system re-run line by line on real prices and matched against their own journals — followed by a 22-cell option-buying hunt, a put-call-ratio surprise, a rule for which years count, and a ranking of every system on the fleet.
Do an end-to-end backtest of the mirror seller — its logic and its implementation, not just the idea.
The live code was replayed line by line on real option prices from January 2020 to October 2026: 1,572 trades, net about ₹8.5 lakh on one lot (₹541 per trade), win rate 50%, average win ₹4,162 against average loss −₹3,044, maximum drawdown −₹89k, profit-to-drawdown 9.5, positive in all seven years. Then the replay was matched against the bot's own paper journal: 47 of 55 trades lined up within five minutes with the same strike, and the bot's actual fills were about ₹300 a trade better than the replay's conservative convention. Two corrections came out of the exercise and were recorded: the first run had the expiry-day roll an hour early, caught only by the journal comparison, and a long-standing "collapse the 09:15 bar" habit turned out to be unnecessary once the two bar sources were compared (median difference 0.00 points). The structure to understand: 67% of trades are held overnight, 94 expiry-day rolls contribute ₹5.4 lakh of the ₹8.5 lakh, and 271 gap-morning exits cost −₹5.0 lakh. Code equals backtest; the edge is real on real prices; the risk is the overnight gap.
Replay the pivot futures system the same way, and recheck the stop-order logic.
Here the fill convention is the result. The same signals over 2019–2026 earn about ₹3.6 lakh when filled the way the bot actually works (poll the price, send a market order), ₹7.6 lakh with resting stop orders at the pivot levels, and ₹30 lakh if every stop is assumed to fill exactly at its level — which no order can do on a morning that gaps 200 points through it. Applying costs one at a time showed what separated the old numbers from the realistic ones: the April 2026 STT rise to 0.05% on futures alone removes about 6.9 index points per round trip, roughly ₹90k a year per lot — more than the whole net of the stop-order version. With corrected bars, 2024–26 comes to −₹22k as coded and +₹1.72 lakh with resting stops, the latter sitting inside ±2 points of slippage with a −₹2.6 lakh drawdown. Decision: no code change; at current STT the futures expression of this signal is dead, and the signals live on in the option books.
Find a buying strategy as successful as the mirror seller — go out of the box.
Twenty-two pre-registered cells: open-interest walls, straddle expansion, market breadth, Donchian channels, expiry-day ranges, IV regimes, call/put volume imbalance, multi-day continuation, and seven ideas generated by independent researchers. Zero passed and twenty-one failed outright. The one lead was an instrument, not a signal: the pivot system's own entries bought as a deep in-the-money (ATM ∓ 500) current-month option — 1,526 trades, ₹671 per trade, win rate 37%, average win ₹9,151 against −₹4,287, confidence interval [219, 1,140], profit-to-drawdown 4.4, maximum drawdown −₹2.30 lakh; it missed the bar only on its worst trade (−₹35k, more than three times the average win). The mechanism is simple: the same trades as the futures bot, but charges of about ₹126 a trade instead of ₹926, because STT falls on the premium rather than the notional. It is sensitive to the deep-ITM spread — at 0.5% per side it drops to ₹359 with two red years — so a quote logger and a paper-only version were started before anything else.
Check the popular 9:20 short straddle on the second data set.
1,663 sessions since 2020, sell the at-the-money pair at 09:20, 25% stop per leg, square off 15:15. Whether a stop counts on a touch or on a one-minute close changes the answer by ₹220 a day (−₹120 versus +₹102), and under both conventions the only pre-registered split that passes is expiry day (+₹271 to +₹580 per day). On non-expiry days in 2023–26 the structure lost ₹252 a day with a confidence interval excluding zero; one leg was stopped on 66–72% of days. The bot's own 72 paper days (+₹351 a day, one good October) cannot distinguish that from zero. Verdict: no edge outside expiry day, which a separate expiry-day system already owns.
Look for option-buying strategies driven by option data itself — OI, skew, put-call ratio — not index systems that merely use an option.
Fourteen cells; thirteen failed on both data sources. The surprise was the put-call ratio of open interest at 09:45 across ATM ± 300: when it was 0.70 or lower (about 20% of sessions), buying a put ITM-150 at 09:46 with the house exits made ₹623 a trade on 330 trades, win rate 40%, both halves positive (₹699 / ₹500), interval [107, 1,168], profit-to-drawdown 3.8, six of seven years — against −₹37 a day for buying the same put every day. The effect is monotonic by bucket and replicates in sign on BankNifty (₹233 versus −₹285 unconditional) without passing there. Checks: the OI is genuinely intraday (about 123 updates per contract per day), the previous day's ratio predicts far less (₹211), and the threshold curve is a broad band from 0.70 to 0.80 rather than a spike. It went to paper as pre-registered — and later the same evening, under the two-window rule below, its modern-window figure (−₹203 on 64 trades) downgraded it from pass to lead.
Which history should we trust — 2015 onwards, 2025–26, or only 2026?
The structural breaks are real: weekly expiries from February 2019 (median days-to-expiry fell from 14 to 2), high volatility in 2020–22, compression in 2023–25, the October 2024 regulatory package (lot size 25 → 75, one weekly expiry per exchange, retail option contracts down 83%), Tuesday expiry from September 2025, the STT change in April 2026. The intraday chain by regime earned ₹502 per trade in 2015–20, ₹880 in 2021–22, −₹298 in 2023, ₹600 in 2024, ₹124 in 2025 and ₹199 in 2026 — the modern figure is ₹100–160, not the ₹447 a sizing plan had assumed, and year-to-year correlation of the per-trade result is 0.18. Power matters more than preference: the per-trade standard deviation is ₹3,680, so one year of ~127 trades resolves ±₹640 — a single year cannot tell ₹200 from zero or from ₹800. The rule adopted: robustness on the full weekly era and on both data sources, but expectation and size from the window since November 2024, with sign agreement there required before money. Regraded under it, the mirror seller (₹725 a trade since late 2024, interval [195, 1,255]) is the only system stronger in the modern era than in its history; the chain is robust but near zero (₹45, interval [−415, 524]) since late 2024.
Backtest the one bot we never backtested — buy today, sell tomorrow on NIFTY futures — then rank all fifteen systems for real money.
Over 2015–2026 the rule (up day above the 20-day average, long at the close, sell at the open) made ₹964 a night on 925 nights, 65% winners, eleven of twelve years positive. Since November 2024 that falls to ₹401 a night with an interval of [−852, 1,654], and 2026 is negative. Two details decided it: the index's 09:15 print reflects the pre-open auction and sits 6.3 points above the 09:16 open, and the fourteen recorded nights show real fills landing at the 09:16 level, so the honest modern figure is −₹244 a night; and real futures charges are about ₹830 a night against the ₹150 the journal had assumed. Not fundable; it stays on paper with honest charges, and its dormant live scheduler was removed. The ranking put the mirror seller first (the only modern-window pass), the intraday chain second at small size as the execution book, a new BankNifty mirror seller (₹736 a trade over seven years, profit-to-drawdown 9.7, ₹570 since late 2024) third pending thirty paper trades, and the deep-ITM pivot buyer fourth pending its paper record. Nothing below the top six had a case for money.