A trend follower's short on silver, the course's last two modules, Minervini's VCP tested without bias, and what four lots should look like

From March to August it took seven short legs for a net of about ₹1.6 lakh per lot, with a worst drawdown of ₹66k and its last three legs all losers in a row. From September to early November, on the current contract's daily candles: two September losses of ₹61k together, then one short entered on 24 September and still open with about ₹2.5 lakh of unbooked gain. That is what a trend follower looks like inside a trend — several losing flips, then one long winner that pays for them all. It does not change the earlier reading: the system's entire record of profit on silver is this year's fall, and it lost money every year before it. It stays on the shelf, to be revisited only if silver's 2026 behaviour carries into 2027.

On real option prices with our costs, the 3-minute version made about ₹1.15 lakh over 169 trades and the 5-minute version about ₹1.25 lakh over 150, both front-loaded into the first half of the sample. The finding that matters: the 3-minute strategy fires on the same day as the advanced call already in the bot 94% of the time, and the 5-minute one 87% — all three are the same 09:18 gap-up continuation wearing different filters, with most entries on the first bar of the day. Adding them to the bot's two legs raises the rupees but raises the drawdown by almost as much, because it is more size on the same days, not more opportunities. The course's claim that combining its strategies cuts the drawdown does not hold on real data for the call variants. Nothing was added; the two modules go on record as parameter neighbours that confirm the installed call leg is not a lucky setting.

The rules were pre-registered before the run: his trend template, relative strength in the top 20% of the universe by six-month return, a base with at least two successively shallower contractions and a volume dry-up, a buy-stop at the pivot filled on the first minute that crosses it, a 7% stop on minute lows, breakeven after one unit of risk, a 20-day-average trail after two, his "train schedule" exit if the trade has not moved after ten days, ₹1 lakh per position, at most five open, delivery costs. On 274 stocks over eleven years the pattern produced 130 trades, 38% winners, a profit factor of 1.7 and about ₹2.6 lakh — of which ₹2.1 lakh was a single year, 2017; 2022 through 2026 were all negative. The decisive test was the placebo: random entry days with the same stop, breakeven and trail made more money per trade and were positive in nine of eleven years. So the profit in this structure is the exit discipline applied to stocks in a bull decade, not the pattern. His money management — small losses, two-to-one, scale only when trades work — is sound and already in use here. As an entry signal the contraction fires about a dozen times a year across 270 names and its edge on Indian data is one year. Nothing was built.

The code as written books exactly one lot at the target whatever the size, so four lots would run as one out, three trailing. Both shapes were backtested on the same seven years: two out, two trailing is exactly the two-lot system doubled — same signals, same fills, about ₹36.9 lakh with a worst drawdown of ₹6.3 lakh. One out, three trailing earns 4.6% more over the whole sample but with 14% more drawdown, a worse profit-to-drawdown ratio, and it loses to the half-and-half shape in two of the last three years. The reason is one the record already showed: the trailing lot is the weak lot — on its own it loses money at two lots and barely breaks even with three — while the partial lot is where the profit is. The decision: when size goes to four, book half, as backtested. It is a one-line change and it waits for the scheduled maintenance window; the plan of record stays at two lots through December.

In order, and the order is the point. Write the rule in three sentences with every number in it, and write down what result would make you reject it, before you test. Get real data — minute bars and actual option prices, saved locally; a formula-priced backtest hides the spread, the decay and the gaps. Have an LLM write the backtest from your written rule: fills on the next bar, real costs, a trade list you can check against a chart. Make it reliable before you believe it: both halves of the sample positive, a placebo with random entries and the same exits that does clearly worse, parameters that survive a 20% wiggle, costs doubled, lot size fixed. Build the bot as a copy of the backtest, sharing its indicator code, with a log of every decision and a fake-broker test harness. Paper-trade it on a small Linux server for weeks and compare its record to the backtest day by day; a difference is a bug or a lie, and you must find which. Then real money behind two locks, one lot, with a bot that never sells what it does not hold, retries every exit until confirmed, halts on anything unknown, and whose order path has been reviewed by someone trying to break it. Start as a buyer, where the loss is capped at the premium. The code is the easy part; the discipline in the first, fourth and last steps is the whole game.