The preceding figures and table focus on model parameter constraints, but as a description of the observational situation it is most useful to characterize the precision, redshift range, and systematic uncertainties of the basic expansion and growth measurements. At present, supernova surveys constrain distance ratios at the 1-2% level in redshift bins of width Δz = 0.1 over the range 0 < z < 0.6, with larger but still interesting error bars out to z ≈ 1.2. These measurements are currently limited by systematics tied to photometric calibration, extinction, and reddening, and possible evolution of the SN population. BAO surveys have measured the absolute distance scale (calibrated to the sound horizon rs) to 4.5% at z = 0.11, 2% at z = 0.35 and z = 0.57, 6% at z = 0.73, and 3% at z = 2.5. Multiple studies have used clusters of galaxies or weak lensing cosmic shear or galaxy-galaxy lensing to measure a parameter combination σ8 Ωmα with α ≈ 0.3-0.5. The estimated errors of these studies, including both statistical contributions and identified systematic uncertainties, are about 5%. RSD measurements constrain the combination f(z) σ8(z), with recent determinations spanning the redshift range 0 < z < 0.9 with typical estimated errors of about 10%. These errors are dominated by statistics, but shrinking them further will require improvements in modeling non-linear effects on small scales. Direct distance-ladder estimates of H0 now span a small range (using overlapping data but distinct treatments of key steps), with individual studies quoting uncertainties of 3-5%, with similar statistical and systematic contributions. Planck data and higher resolution ground-based experiments now measure CMB anisotropy with exquisite precision.
A flat ΛCDM model with standard radiation and neutrino content can fit the CMB data and the BAO and SN distance measurements to within their estimated uncertainties. However the Planck+WP+BAO parameters for this model are in approximately 2σ tension with some of the direct H0 measurements and most of the cluster and weak lensing analyses, disagreeing by about 10% in each case. Similar tensions are present when using WMAP data in place of Planck+WP data, but they are less evident because the WMAP errors are larger and the best-fit Ωm value is lower. Moving from ΛCDM to wCDM can relieve the tension with H0, but only by going to w < -1 (which would be more physically startling than w > -1), and this change on its own does not produce better agreement with the structure growth data. It is not clear whether current tensions should be taken as a sign of new physics or as a sign that at least some of the experiments are underestimating their systematic uncertainties. Factor-of-two reductions in error bars, if convincing, could lead to exciting physical implications, or to a resolution of the existing mild discrepancies. Moving forward, the community will have to balance the requirement of strong evidence for interesting claims (such as w ≠ -1 or deviations from GR) against the danger of confirmation bias, i.e., discounting observations or error estimates when they do not overlap simple theoretical expectations.
There are many ongoing projects that should lead to improvement in observational constraints in the near-term and over the next two decades . Final analyses of Planck temperature and polarization maps will significantly tighten the CMB constraints, including an important reduction of the uncertainty in the matter fluctuation amplitude that will sharpen tests based on structure growth. Final data from the SDSS-III BOSS survey, finishing in 2014, will reduce BAO errors by a factor of two at z = 0.3, 0.6, and 2.5. Its SDSS-IV successor eBOSS will yield the first BAO measurements in the redshift range 1 < z < 2 and improved precision at lower and higher redshifts. The HETDEX project will measure BAO with Lyman-α emission line galaxies at z = 2-3. The same galaxy surveys carried out for BAO also provide data for RSD measurements of structure growth and AP measurements of cosmic geometry, and with improved theoretical modeling there is potential for large precision gains over current constraints from these methods. The Dark Energy Survey (DES), which started operations in August 2013 and will run through 2018, will provide a sample of several thousand Type Ia SNe, enabling smaller statistical errors and division of the sample into subsets for cross-checking evolutionary effects and other systematics. DES imaging will be similar in depth but 50 times larger in area than CFHTLens, providing a much more powerful weak lensing data set and weak lensing mass calibration of enormous samples of galaxy clusters (tens of thousands). Weak lensing surveys from the newly commissioned Hyper Suprime-Cam on the Subaru telescope will be smaller in area but deeper, with a comparable number of lensed galaxies. Reducing weak lensing systematics below the small statistical errors of these samples will be a major challenge, but one with a large payoff in precision measurements of structure growth. Uncertainties in direct determinations of H0 should be reduced by further observations with HST and, in the longer run, by Cepheid parallaxes from the GAIA mission, by the ability of the James Webb Space Telescope to discover Cepheids in more distant SN Ia calibrator galaxies, and by independent estimates from larger samples of maser galaxies and gravitational lensing time delays.
A still more ambitious period begins late in this decade and continues through the 2020s, with experiments that include the Dark Energy Spectroscopic Instrument (DESI), the Subaru Prime Focus Spectrograph (PFS), the Large Synoptic Survey Telescope (LSST), and the space missions Euclid and WFIRST (Wide Field Infrared Survey Telescope). DESI and PFS both aim for major improvements in the precision of BAO, RSD, and other measurements of galaxy clustering in the redshift range 0.8 < z < 2, where large comoving volume allows much smaller cosmic variance errors than low redshift surveys like BOSS. LSST will be the ultimate ground-based optical weak lensing experiment, measuring several billion galaxy shapes over 20,000 deg2 of the southern hemisphere sky, and it will detect and monitor many thousands of SNe per year. Euclid and WFIRST also have weak lensing as a primary science goal, taking advantage of the high angular resolution and extremely stable image quality achievable from space. Both missions plan large spectroscopic galaxy surveys, which will provide better sampling at high redshifts than DESI or PFS because of the lower infrared sky background above the atmosphere. WFIRST is also designed to carry out what should be the ultimate supernova cosmology experiment, with deep, high resolution, near-IR observations and the stable calibration achievable with a space platform.
Performance forecasts necessarily become more uncertain the further ahead we look, but collectively these experiments are likely to achieve 1-2 order of magnitude improvements over the precision of current expansion and growth measurements, while simultaneously extending their redshift range, improving control of systematics, and enabling much tighter cross-checks of results from entirely independent methods. The critical clue to the origin of cosmic acceleration could also come from a surprising direction, such as laboratory or solar system tests that challenge GR, time variation of fundamental "constants," or anomalous behavior of gravity in some astronomical environments. Experimental advances along these multiple axes could confirm today's relatively simple, but frustratingly incomplete, "standard model" of cosmology, or they could force yet another radical revision in our understanding of energy, or gravity, or the spacetime structure of the Universe.