Implied grams, a contaminated baseline, and the cost of proving an edge

On 20 Aug 2026 GOLDBEES closed at ₹128.88, COMEX gold at $4,571.40 and USDINR at 95.4960, with duty at 15%.

(a) Compute the parity in rupees per gram and the implied grams of duty-paid gold per GOLDBEES unit, $u = C/\Pi$. (b) The 250-day median of $u$ at that date was 0.0081941 and the median since 13 May 2026 was 0.0080920. Compute the premium or discount $\rho = u/\text{baseline} - 1$ against each. Which baseline does a rule with a −2.5% threshold need, and why? (c) The calendar-year medians of $u$ were 0.008642 in 2016 and 0.008261 in 2025. Compute the implied average annual rate of decline over the nine years. Name two separate causes that this number mixes, and say how you would separate them with public data. (d) Before the trade, the rule's 13 completed trades had mean $E = 0.0406$R and standard deviation $s = 1.4552$R. Using M03 (3.2a) and (3.2b) with $z_{0.8} = 0.8416$, how many trades are needed for $t = 2$ in expectation and for 80% power? At 1.3 signals a year, what does this imply? (e) The fill was at ₹130.14 and the stop at ₹126.02. With ₹10,00,000 of capital and 1% risk, compute the position size from (3.23), its notional value and the loss at the stop. How many units would a trader who sized from the signal-day close have bought, and what R-multiple would the same stop have cost them?