One idea, thirteen versions, one deployed.

Three things have to be true at once. Remove any one and this session doesn't happen.

The original idea

NIFTY monthly puts, 2024-01 → 2026-07, minute-level replay, costs included.

A ₹50 put that never doubles decays to near zero, so each miss costs almost the whole premium. The trail caps winners near +₹7.7k while losers run to −₹6k. You need better than 44% of entries to double just to break even; the idea delivered 34%. Only 2026 — the bear year — was positive.

Inverting it

If buying decay loses, sell it. Version 2 bought one ₹100 put and sold two ₹50 puts — roughly zero cost to enter — holding to expiry unless the index traded below the short strike.

Profitable, but the median cycle earns ₹139 — the entire result is one good month minus one bad month. Remove the single best cycle and two and a half years nets −₹8,200. Not a strategy; a coin-flip with extra steps.

The mirror test

Before trusting any of it, the same structure was run in reverse — short the ATM, long two cheaper puts — and the two results were compared cycle by cycle.

Four variants, one winner

The value-neutral ratio — long one ATM put, short two half-price puts, stop when the package loses ₹50 per unit — became the base. Then four ideas were tested against it, each a real strategy, each built in about a minute.

Research model, 2024-01 → 2026-07. Identical entry logic; only the management rule differs.

Removing the stop was the most instructive failure. It nets roughly zero while risking seven times the drawdown — one month alone cost −₹139,165. And take-profit fails for the opposite reason: the winning months pay several hundred points, so exiting at +50 sells the only thing that works.

The winner was the trader's own idea: each time the index runs 500 points, add another spread at the new strikes. It stacks profit zones 500 points apart, so a shallow pullback that misses the original still lands in a later one. It added ₹48,258 with no additional drawdown.

Then: does it survive COVID?

Two and a half years is a small sample containing no true crash. The window was pushed back to January 2020 — 1,614 trading days, 139 cycles, including the fastest index collapse in modern Indian market history.

The drawdown barely moved — −₹24,262 over 6.5 years versus −₹21,019 over 2.5 — despite adding the worst crash available. March 2020 cost about ₹9,000 in total: the stop fired two minutes after entry on one cycle and three days in on another. The deepest drawdown in the whole history is still a 2025 episode, not COVID.

Equally important, the longer sample reduced the headline: 12.2% CAGR and Calmar 1.02, against 24% and 2.36 on the shorter window. That is what a longer sample is supposed to do to a flattering number, and it is the figure worth planning around.

Does it work on calls too?

The identical structure mirrored to the call side — long ATM call, short two cheaper calls — over the same 6.5 years.

Every year negative. The asymmetry is structural, not a fitting artifact: markets fall in sharp, bounded moves and rise in slow grinds. The put structure's profit zone gets fed by frequent pullbacks; the call version needs a sized rally that grind-up markets overshoot slowly and crash-bounces blow straight through. Same logic, mirrored instrument, opposite regime fit.

How many should we sell against it?

The structure sells two puts per long put. Is two right? Four ratios were tested, normalised so each is measured per unit of long exposure.

1 × 3 earns 32% more money and is a far worse strategy. It wins more often — the deeper short strike is harder to reach — but below it the position loses at double the slope. One month cost −₹100,189, which is 46% of the entire six-year profit in a single cycle.

The curve rises smoothly to 2 and then collapses: 6.67 → 8.19 → 8.99 → 2.63. That shape is a genuine optimum rather than a fitted parameter — and it lands on a clean integer because two is where the premium sold exactly funds the put bought, leaving net short exposure of one unit.

And where should the strikes sit?

The ratio question was “how many to sell.” The other half is “where to place them.” The whole structure was slid up and down — buying deeper in-the-money, then further out-of-the-money — keeping every other rule identical.

Research model, 2020-01 → 2026-07, per unit of long exposure.

Both directions lose, for opposite reasons. Buying in-the-money reduces time decay — a sound instinct — but the dearer long leg has to be funded by selling a richer short put, which drags it closer to the money. Ordinary noise then trips the stop: 38 stop-outs at the money becomes 63 at 300 points in.

Going out-of-the-money fails differently. The cheap long leg needs almost no funding, so the far short lands right next to the middle one — and the structure quietly collapses into a narrow spread with two adjacent short puts. That is why the drawdown more than doubles even though it gets stopped out less often. The giveaway is in the last column: months that finish above everything rise from 63 to 95, while profitable months fall from 30 to 21. It stops losing and stops winning at the same time.

Only at the money does the premium sold exactly fund the option bought while keeping the far short genuinely far away. Two independent parameters — how many to sell, and where to place them — both peak on the same clean setting. That is the difference between a structural optimum and a curve-fitted one.

A good hypothesis that failed

The losing cycles clearly clustered at high volatility, so the obvious improvement is: don't enter when VIX is high, and exit if VIX spikes. The observation was correct — cycles whose peak VIX exceeded 25 lost ₹23,526 between them, while every bucket below 25 was positive.

The fix did not work, at any threshold.

The more VIX exits taken, the worse the result — and the drawdown never improved at any threshold. Fourteen configurations were tested, entry gates and exit gates, from 16 to 40. Not one reduced the drawdown by a rupee.

The reason is a classic trap: high VIX is a symptom that arrives with the loss, not a warning that precedes it. By the time VIX prints 25 the market has already fallen and the position is already marked down — and from there it often recovers, because this structure profits as the index grinds into its profit zone. Exiting on the VIX print locks in the loss and forfeits the recovery.

The ladder — and the decision that mattered

A statistical tie on risk-adjusted return — 9.15 against 8.99 is noise. So the tiebreak became a question the backtest doesn't answer directly: which one is easier to actually hold?

The ladder's worst day is 35% smaller. Its bad days are 27% smaller. For a strategy you have to live with for months, that difference is worth more than a few percent of return — and it costs nothing risk-adjusted.

The out-of-sample check that reversed a recommendation

One more question: was COVID carrying the result? The four live candidates were re-run on the last two years only.

This test killed a recommendation made an hour earlier. The wider ₹100 stop had looked like a mild improvement over the full period — best in the table at 9.46. Out of sample it is the worst of the three viable configs. That gain lived entirely in 2020–2023 and did not survive. The stop stayed at ₹50.

The ladder, by contrast, ranked first in both windows. Ranking first in a COVID-inclusive and a COVID-excluded sample is the strongest robustness signal the session produced — so the ladder with a ₹50 stop is what got built and deployed to paper.

What the tooling caught that a human wouldn't

The session, in order

Why it takes all three

The verdict, on the platform’s own engine

The two finalists were built as real strategy templates and run through Tradetron’s backtest engine over an identical window — same costs, same data, same rules. This is the comparison the deployment decision was made on.

Tradetron backtest engine, 2020-04-15 → 2026-07-24, net of costs, on the deployed templates.

The ladder wins on every measure that matters: more money, half the drawdown, and twice the return per unit of risk. It also delivers the gentler day-to-day path shown earlier — which, for a position held for months, is what determines whether you actually stay in it.

Both reports are public. Every fill, every exit reason, every month is in them.