We show that the assumption of a flat universe induces critically large errors in reconstructing the dark energy equation of state at z 0.9 even if the true cosmic curvature is very small, O(1%) or less. The spuriously reconstructed w(z) shows a range of unusual behaviour, including crossing of the phantom divide and mimicking of standard tracking quintessence models. For 1% curvature and ΛCDM, the error in w grows rapidly above z ∼ 0.9 reaching (50%, 100%) by redshifts of (2.5, 2.9) respectively, due to the long cosmological lever arm. Interestingly, the w(z) reconstructed from distance data and Hubble rate measurements have opposite trends due to the asymmetric influence of the curved geodesics. These results show that including curvature as a free parameter is imperative in any future analyses attempting to pin down the dynamics of dark energy, especially at moderate or high redshifts.Introduction The quest to distinguish between a cosmological constant, dynamical dark energy and modified gravity has become the dominant obsession in cosmology. Formally elevated to the status of one of the most important problems in fundamental physics, [22] [23] uncovering the true nature of dark energy, as encapsulated in the ratio of its pressure to density, w(z) = p DE /ρ DE , has become the focus of multi-billion dollar proposed experiments using a wide variety of methods, many at redshifts above unity (see e.g. [1]).Unfortunately these experiments will only measure a meagre number of w(z) parameters to any precisionperhaps two or three [3] [24] -since the standard methods all involve integrals over w(z) and typically suffer from subtle systematic effects. As a result of this information limit, studies of dark energy have traditionally fallen into two groups. The first group (see e.g. [10]) have taken their parameters to include (Ω m , Ω DE , w) with w constant and often set to −1. The 2nd group are interested in dynamical dark energy and have typically assumed Ω k = 0 for simplicity while concentrating on constraining w(z) parameters (see e.g. [11] ) [25].In retrospect, the origins of the common practise of ignoring curvature in studies of dynamical dark energy are clear. Firstly, the curved geodesics add an unwelcome complexity to the analysis that has typically been ignored in favour of studies of different parametrisations of w(z). Secondly, standard analysis of the Cosmic Microwave Background (CMB) and Baryon Acoustic Oscillations (BAO) in the context of adiabatic ΛCDM also put stringent limits on the curvature parameter, e.g. Ω k = −0.003 ± 0.010 from WMAP + SDSS [12,13]. As a result it was taken for granted that the impact on the reconstructed w(z) would then be proportional to Ω k and hence small compared to experimental errors.Further support for the view that Ω k should not be included in studies of dark energy came from informa-