Posted in My Daily Journal · 13 July 2026
₹23,500 of minute-level option prices for two falsifying years passed the option overlay; a free decade of per-contract data then showed it loses in eight years of eleven. The order of those two results is the lesson.
Why pay for two specific years of option data?
To falsify, not to confirm. A per-second sample showed the ATM weekly spread was about ₹0.15 median — 0.2% round trip — so minute data with a 0.1%-per-side penalty was tick-faithful enough. We bought 2021 and 2025 because they were the weakest trend years of the weekly-option era for the futures system; strong years cannot change a go/no-go decision, weak ones can.
Did the option overlay pass on those two years?
Marginally, after a bug. The first run showed +220% and +165% and was wrong: a ticker prefix matched three futures contracts at once, mixing price levels and creating phantom early fills — essentially a look-ahead. Fixed, the exact trigger-minute numbers were 2021 +₹73.5k (win 33%, drawdown ₹43.8k) and 2025 +₹39.9k (win 31%, drawdown ₹91.3k, profit-to-drawdown 0.44). A pass, with a poor ratio. The same replay vindicated the futures bot: real 2025 futures fills were about 30% better than the pessimistic backtest.
Then the decade test. What changed?
A free per-contract archive of 2014–2024 NIFTY options, audited against the paid data on the 2021 overlap (median per-trade difference ₹0, zero strike mismatches). The same signals, buying weekly ATM options, year by year: 2015 −174k, 2016 −140k, 2017 −112k, 2018 −146k, 2019 −81k, 2020 −144k, 2021 +91k, 2022 +44k, 2023 +12k, 2024 −15k, 2025 +33k. About −₹6.3L over the decade against +₹11.9L for futures on identical signals. The two paid years had accidentally sampled the overlay's best era. In 2020 the crash monsters were captured correctly; the loss was the post-crash bleed of re-buying 60–80 IV premium on every flip.
What was concluded?
Four things on record. The options wrapper is a regime bet, not a system — three positive years in eleven. Real-money go-live for the option-buying bot was switched off; its paper trial continued only to measure fills. No "when to wear the wrapper" filter would be hunted, because three positives cannot train a rule without curve-fitting. And the edge is the signal; capital belongs where the signal is cleanly expressed.
You also tested "withdraw at equity highs and re-enter later". Result?
Rejected, and instructive. On two years, lock-₹50k rules looked wonderful — 2025 net doubled, drawdown −33%. On the 11.5-year series every variant made less (₹1.2L to ₹4L less of ₹13.5L), drawdown barely moved, and the ratio was worse across the board. Trends cluster, so benching after windfalls skips positive-expectancy trades; an equity-curve moving-average filter was catastrophic everywhere. That closed a family: take-profits, volatility filters and equity-curve timing all fail. Profit protection lives in two places only — the ratcheting trail at trade level and a stand-down at catastrophe level.
A forum post claimed an 80%-win ITM scalping system made ₹3L in 2025. Real premiums?
−₹63,393, profit factor 0.89 against the claimed 1.30. The edge per trade was ₹127 against ₹120 of costs — toll-collector economics. The strategy as described was falsified; the author's real experience may be different, but the stated rules do not produce the stated result.