Replicating a paid options course rule by rule, building it as a bot, and what an independent review found before any money

Transcribe them, screenshot every configuration screen, and write down each strategy as a rule with every parameter: the bar size, the indicator settings, the entry conditions, the exit conditions, which option is bought, the stop, the trailing step, the time exit, and whether re-entry is on. The course's base strategies buy a call or a put on 3-minute index bars when a bar closes beyond the supertrend with RSI past 70 or 30, beyond the previous day's pivot, and beyond a Bollinger band; the advanced call tightens the band to three standard deviations over fifteen bars, adds a Williams %R filter, a 13% stop that trails 10% for every 15% the premium gains, and a 15:10 exit; the advanced put adds a 200-EMA, a 20-period RSI below 30, CCI below −200, Williams %R below −85, a narrow central pivot range and exits when RSI recovers above 45. One trade a day, the current week's option priced nearest ₹250. Our earlier attempt at "the Ganesan setup" had used 5-minute bars, a different supertrend multiplier and unlimited re-entries — close, but not the course.

On NIFTY, seven years of real weekly option quotes, one lot, our costs: the advanced call made about ₹2.3 lakh over 300 trades (₹750 a trade, 53% winners, worst drawdown ₹16k, six of seven years positive — this year slightly negative); the advanced put about ₹1.5 lakh over 213 trades (₹700 a trade, 46% winners, worst drawdown ₹17k, seven of seven years). Both reproduce the course's claims in shape, at a third to half of its quoted rupees once realistic costs are charged. The base strategies make similar money with far more trades and three times the drawdown, which is the course's own point. Every one of eighteen parameter neighbours was positive in both halves of the sample, and placebo versions — the same exits on random entry bars — made a fraction of the real rules' profit. That is what "the edge is in the entry, not the exits" looks like in numbers.

BankNifty's advanced call is real but three-quarters of its profit came before the weekly expiries were withdrawn; since then it earns a few thousand a year. Its put fails. On gold and silver, where usable intraday option data does not exist, the entry filters were run at the futures level: silver's long side works in the same evening pattern we already had, with fewer trades; gold fails; every short fails. The recommendation that came out of the replication was one instrument: NIFTY, both legs.

It buys both ways — a call on the breakout setup, a put on the breakdown setup — as two independent legs, one trade per leg per day, each exiting on its own rule, stop or time. It never sells an option it does not hold. It is strictly intraday: the call leg is flat by 15:10, the put leg by 15:20. The course's put exits at 15:28; that was pulled earlier because the broker force-closes intraday positions a few minutes before the close, and a bot needs a minute and a half of its own for an order to go through — only fifteen of the backtest's put trades reached the time exit, so the cost of the change is noise, and the re-run with the earlier exit came out marginally better.

Two ways. A synthetic test harness runs the bot's own code against a scripted exchange — fake clock, fake quotes, scripted order fates — through 112 checks: indicator arithmetic identical to the engine's to a millionth, the trailing stop's exact step boundaries, one trade per leg, both legs on one day, a stop jumped inside a minute, late candles, restarts mid-trade, holidays, a dead token. Then the decisive one: the same code replayed day by day on the real historical data, minute by minute, with a stub that serves quotes the way the engine saw them. Over 323 sessions, 99 of 100 trades were identical to the engine — same day, same entry minute, same strike, same prices to the paisa, same exit. The single difference was the engine's mistake, not the bot's: it had let a stop fire two minutes after its own time exit.

The reviewer was given only the code and a specification, told to read it as a trader who has run automated option buying with real money, and to break it. The paper logic passed. The live order path had two blockers. First: if the broker's positions feed was unreachable at the moment of an exit, the bot would have sold the quantity in its own book without confirming the lot was still held — the one way a long-only bot can end up short an option, and the reviewer proved it in the test harness. Second: a network timeout on a sell could store an error message in place of an order number, after which the leg had no stop for the rest of the day. Five more high findings followed: a halt that silently cleared itself overnight, a dashboard that read its arming flag only at startup, a restart schedule with a two-hour hole, an exit time too close to the broker's forced square-off, and an unconfirmed sell booked as sold. Every finding was fixed and each fix got its own test: a sell is placed only for a quantity the broker has just confirmed; an unconfirmed sell is parked and never followed by a second one until the first is known dead; every exit the bot owes is retried until it is done; a position the broker shows that the book does not is adopted and managed; a short in one of its symbols is bought back at once; a halt persists until a human clears it. The best thing the review did was not the list — it was showing that a pattern shared by older bots has the same two holes, which are now scheduled for the same patch.

Four things the reviewer put plainly. The money at risk per lot is the whole premium — around ₹16k — not the 13–15% stop, because a stop sampled once a minute can be jumped. Live profit will run below the backtest, since the backtest fills at the next minute's open and a real exit is a market order fired in the very minutes a stop is hit; a 15–25% haircut is the planning number. The call leg is mostly a gap-up continuation taken at 09:18 — 257 of its 300 entries were on the first bar of the day — which is a specific trade with a specific character. And the ₹250 target drifts with volatility: in a quiet market it is an in-the-money option, in a violent one it is far out of the money with a very different stop. None of that changes the backtest; all of it changes how you size.

Eight of ten closed above the previous day's close; the two that fell were 2016 and 2017, the worse by 0.63%. Inside the hour itself the index opened higher and then drifted down in seven of the eight sessions we have minute data for — by 2 to 68 points — and the widest full-session range in eight years was 108 points, under half a percent. For a deep in-the-money position with a stop several hundred points away, a session like that is noise, and a bot that only runs on weekdays correctly sleeps through it. The one quirk worth knowing: the next trading day's pivot levels come from that short session, so they sit unusually tight.

Tested both halves of it on seven and a half years of silver-mini futures, evening entries only. The course's entry filters keep the same edge per trade and throw away two-thirds of the trades — a third of the money. The course's stop and trail, laid over the installed rules and expressed as a percent of the futures price, moves the result by between minus ₹81k at a tight setting and plus ₹13k at a loose one — the loose settings are indistinguishable from no stop, because the supertrend flip and the session-end exit already cap the losers at a ₹17k worst drawdown. Not a solid finding, so nothing changes; a 0.25% stop is noted as a ready patch if the paper record ever shows losers larger than the backtest's.