Should exits follow the calendar's mood? And what the first live order taught about broker APIs

Months, no; expiry days, maybe. We let a walk-forward procedure pick the best trail-and-target combination for each context from prior years only, and applied it to the next year. The fixed rule (20% trail, no target) made +₹2.72L with profit-to-drawdown 4.8 on 992 trades. Picking by month collapsed to +₹84k (P/DD 0.8) — the picks flipped every year, which is what noise looks like. Picking by expiry versus non-expiry was the only context with a stable choice: a wide 40% trail with an 80% target on expiry days, chosen six years running, earned ₹533 per expiry trade out of sample versus ₹430 for the fixed rule.

Because the gain is modest, the sample is small, and the mechanism — zero-day in-the-money options are gamma noise, so a tight trail shakes you out — is plausible but not proven. The live rule stayed unchanged and the alternative went into a shadow ledger: the bot records what the expiry-day exit would have done on every expiry trade, to be judged after about twenty of them. October, incidentally, is the rule's weakest month historically (108 trades, +₹7.9k, 40% winners) — written down in advance so a weak October would not be read as a broken system.

It was rejected. The broker's API refused a plain market order on an option without a market-protection parameter — a rule the paper bot had never met, because no paper trade ever touches the order path. The bot handled the rejection correctly (fell back to paper and managed the position virtually), the fix was a one-line parameter with a limit-order fallback, and the lesson went on record: the order path is only tested the first time it meets the real broker, so build the fallbacks before that moment and expect the first live signal to be a test of plumbing, not of edge.