"Does This T-Shirt Make My Estimate Look Big?" In Agile estimating and planning, self-similar scales such as Planning Poker, Fibonacci Numbers, and T-Shirt Sizing are commonly used to assess the size of software development efforts. We start with the simple question of "What if we are off by one [size in either direction]?" Since such estimates typically rely on Expert Judgment, this supposition is not unreasonable. Exploring various discrete and continuous probability distributions, including the Uniform, Triangular, and Lognormal, this paper derives closed-form solutions for expected growth from the point estimate (i.e., Risk) and the accompanying Uncertainty in the form of a coefficient of variation (CV), and generalizes to any choice of confidence level. It considers the trade-offs in the ratio of these scales, with smaller ratios (like Planning Poker) offering finer gradations but an increased chance of being "wrong," while larger ratios (like T-Shirt Sizing) sacrifice granularity for accuracy. Not only do Fibonacci numbers approach the Golden Ratio (approx. 1.618), but they also have the intuitive advantage that any given size is the sum of the two preceding sizes. Drawing on the author's earlier work ("Understatement of Risk and Uncertainty by Subject Matter Experts," SCEA, 2011), this paper proposes improvements to the traditional elicitation process by replacing abstract numbers on the self-similar scale with analogous data points. Finally, it tests the accuracy of forecasts made using both scales for knowable and future quantities.