In the late 1990's, observations of 93 Type Ia supernovae were analysed in
the framework of the FLRW cosmology assuming these to be `standard(isable)
candles'. It was thus inferred that the Hubble expansion rate is accelerating
as if driven by a positive Cosmological Constant $\Lambda$. This is still the
only direct evidence for the `dark energy' that is the dominant component of
the standard $\Lambda$CDM cosmological model. Other data such as BAO, CMB
anisotropies, stellar ages, the rate of structure growth, etc are all
`concordant' with this model but do not provide independent evidence for
accelerated expansion. Analysis of a larger sample of 740 SNe Ia shows that
these are not quite standard candles, and highlights the "corrections" applied
to analyse the data in the FLRW framework. The latter holds in the reference
frame in which the CMB is isotropic, whereas observations are made in our
heliocentric frame in which the CMB has a large dipole anisotropy. This is
assumed to be of kinematic origin i.e. due to our non-Hubble motion driven by
local inhomogeneity in the matter distribution. The $\Lambda$CDM model predicts
how this peculiar velocity should fall off as the averaging scale is raised and
the universe becomes sensibly homogeneous. However observations of the local
`bulk flow' are inconsistent with this expectation and convergence to the CMB
frame is not seen. Moreover the kinematic interpretation implies a
corresponding dipole in the sky distribution of high redshift quasars, which is
rejected by observations at 4.9$\sigma$. The acceleration of the Hubble
expansion rate is also anisotropic at 3.9$\sigma$ and aligned with the bulk
flow. Thus dark energy may be an artefact of analysing data assuming that we
are idealised observers in an FLRW universe, when in fact the real universe is
inhomogeneous and anisotropic out to distances large enough to impact on
cosmological analyses.