What cognitive system(s) provides the conceptual resources that inform number-word learning? Le Corre and Carey (2007) made a strong set of arguments that the answer cannot be the analogue magnitude system. Here we re-examine the most powerful of these arguments, based on scalar variability (the standard deviation of answers growing linearly with the mean), which is a signature of the analogue magnitude system. Using adult data, we show that a certain additional nuance is needed to understand this signature: it applies to a hidden continuous part of the system, and not always to the discrete integers that people produce. We then show that this new description, combined with the knower-level framework, is consistent with a corpus of child data: young children do show scalar variability in their responses when asked how many items are on a card. This points towards a role for the analogue magnitude system in number-word learning.