Before you start: (a) define one dimensionless target metric and its range; (b) write down the specific variables you plan to measure; (c) ask: with only these variables, is the claimed accuracy realistic—or are key, independent effects missing? If uncertain, add one clearly independent variable or rethink the setup; if not, proceed. Let’s build an RG thread of short templates: task → variables measured → promised accuracy → result → was the plan realistic? The aim is practical: fewer costly studies that were unlikely to succeed from the start, and faster launches for the ones that can.