Randomized Experiment Fails Despite Fair Coin: Missing Join Key

A team built a cryptographically fair randomized experiment but discovered weeks later it was unanalyzable because no join key linked coin flips to tasks. The coin was verified unbiased, but the dispatcher logged draws without recording which task they applied to, making intention-to-treat analysis impossible. The post highlights that randomization alone is insufficient without proper bookkeeping of assignment-to-outcome links.