Currency futures under every rule we own, what FX research actually supports, an AI forecaster re-priced on real contracts, a straddle that only works on expiry day, and a crore-scale verified P&L read month by month

No, on any pair. FYERS serves rupee pairs as a continuous front-month one-minute series back to early 2018 — expired contracts cannot be fetched by name, which cost a morning — and on that data all sixteen combinations of four rules and four pairs lose money, in both halves of the record and in almost every year. Before costs each rule earns between minus ten and plus eight rupees per lot per trade, which is zero; costs are about fifty rupees. The shape is right — winners are longer than losers in fifteen of sixteen cells, by 1.2 to 1.7 times — but an average winner is seventy to a hundred and fifty rupees per lot, six to twelve paise of movement, and that is not enough to pay for a breakout. On NIFTY the same rule's winner is nearly three thousand rupees and costs are four percent of it. A breakout needs a market that keeps travelling after the break; the rupee pairs move a few paise a day.

Thirty-nine runs, each rule copied from its bot, with contract rolls detected and removed and NIFTY-sized thresholds also scaled to each pair's range. All thirty-nine negative; none positive in more than two of nine years. Again the trend rules keep their NIFTY shape — winners 1.6 to 1.9 times losers — and again the size is missing: the pivot system's average winner on the pound pair is under five hundred rupees a lot against fifteen thousand on NIFTY. Every rule's edge before costs sits within seventy rupees of zero. Two books could not be tested because there is no currency option data.

The published record is thinner than the folklore. Retail contest winners describe trend-following, "smart money concepts" or Elliott waves and, mostly, oversized positions for a few weeks; the repeat winners of the professional championship run undisclosed diversified rule-based futures portfolios. What the research supports is slow: carry (earning the interest gap), three-to-twelve-month momentum, and Turtle-style breakouts; the popular day trades, such as the London-open breakout, lose in published ten-year tests. We pre-registered ten of these and ran them on the three rupee pairs. The slow trend family is the only one with the right shape — winners two to five times losers — and it shows most on the euro pair, where four of those rules finish positive. But the best of them earns about twelve hundred rupees a year per lot, loses money in its first half, and nothing works on two of three pairs. Everything counter-trend or intraday — fading big days, fading gaps, month-end dollar demand, session drift — has no edge at all. Currency research is closed on our side unless a currency-specific idea arrives.

No. Its saved daily forecasts, run through its own rules and priced for the first time on tradable futures at the morning fill it actually uses, lose on both pairs over eight and a half years: winners about the size of losers, gross near zero, costs the rest. Even filled at the forecast day's own close, which it cannot do live, it is roughly breakeven. The real-money profit was genuine but was mostly one short on the pound during a week the pound fell two and a half percent — a handful of trades that cannot be told from luck. It stays a paper record.

It does not. The paper record is one strong month: a loss over its first fifty-four days, then a large gain in the twenty-five days since the start of October, an average that is still within what a zero-edge strategy produces by chance. The backtest reproduces the bot's own days with a correlation of 0.88 and shows the same losing summer and good October. Over seven years the structure earns on expiry days — three to six hundred rupees per expiry, every year positive — and loses on every other day since 2023. Its winning day is 0.78 times its losing day: small wins, bigger losses, the wrong shape. Run daily it would need roughly four lakh per lot to survive its own drawdown for an expectation near zero; run on expiry days only, which is what the separate expiry bot already does, about two lakh per lot. Capital goes to the expiry-day version, after its paper gate, or nowhere.

Each is "most volatile" in a different sense. Out-of-the-money moves most in percentage: it is cheap, so a small rupee change is a large swing, and it can double or go to zero in hours — a lottery ticket. In-the-money moves most in rupees and least in percentage: most of its price is real value, so it follows the index and decays slowly. At-the-money is where the behaviour changes fastest — a move in your favour makes it act in-the-money, a move against makes it act out-of-the-money — and it carries the most time value, so it loses the most rupees per day when nothing happens. That is why the weekly pivot book, which holds for days, buys in-the-money and the intraday breakout book buys close to the money; pure out-of-the-money buying has never passed a test here because the decay on the quiet days outweighs the occasional big win.

The mechanical version of that idea has been tested and lost: twelve years of "read the day, deploy the right tool" rules, predict-then-switch, detect-then-hand-off, and a foundation model's 09:45 read as a filter — the always-on fleet beat every one, out of sample. A judgement call has no information the data lacks — gap, volatility, trend and levels are already the gates the bots carry, the parts of regime reading that survived testing — and it cannot be backtested, so it can never earn the bars the bots had to pass. What is worth doing is a visible regime card and a drift monitor that pauses a bot whose live record leaves its backtest distribution. If the question must be settled rather than argued, the fair test is a pre-registered paper book: a one-line morning call, logged before the market shows its hand, compared with the always-on fleet after sixty sessions.

Read month by month from the verified data rather than the headline, it is not a system; it is one discretionary bet on a gold bull run, in very large size. Three gold futures contracts made about twenty-one crore; everything else — silver, copper, stock futures, index options bought in lakhs of units — lost about fifteen crore. From December to May the account made fifteen crore; from June onwards it gave back seven. Only a hundred and nine of two hundred and fifty-four trading days show any realised result, so positions are held for weeks and booked on a few days; the best and worst days, plus seven crore and minus six, were both gold, two days apart. The shape is the trend-trader's — many losers, a few enormous winners — but it depended on catching one regime, and it is now in a drawdown the size of half the headline. The testable idea inside it is the one we already know: hold a trending metal while the trend holds and size the winners, which is what our own daily-trend tests on silver showed, and with the same caveat — the profit lives in one regime.