When can the Planck satellite measure spectral index running?

We use model selection forecasting to assess the ability of the Planck satellite to make a positive detection of spectral index running. We simulate Planck data for a range of assumed cosmological parameter values, and carry out a three-way Bayesian model comparison of a Harrison-Zel'dovich model, a power-law model, and a model including running. We find that Planck will be able to strongly support running only if its true value satisfies |dn/d ln k| > 0.02.