No single cosmological observable arrives to us cleanly separated from intervening astrophysical processes, thus a unified understanding is needed to test our models of cosmology. This thesis uses mock observations, simulated skies matched to the statistical and instrumental properties of real and proposed surveys, to disentangle these components and give a direct, quantitative point of comparison between theoretical models and observational data. We build on the Peak Patch-WebSky pipeline, forward modelling dark matter (DM) halo catalogues into cosmic microwave background (CMB) foreground maps and tomographic line intensity mapping (LIM) observables. These mocks are used to search for observable traces of primordial non-Gaussianity (PNG) left in the large-scale structure (LSS) of the universe. Of particular interest is primordial intermittent non-Gaussianity (PING), a form of PNG that arises generically in multi-field inflation.

We introduce the WebSky2.0 mocks, achieving an order of magnitude finer mass resolution and over 40% greater volume to previous WebSky CMB foreground mocks and extending WebSky to various LIM observables. We demonstrate sensitivity to PING in the thermal Sunyaev–Zel’dovich (tSZ) power spectrum, and the number density of halos passing high-mass and ellipticity selection cuts. Finally, we forecast the sensitivity of upcoming [CII] and CO LIM survey using the voxel intensity distribution (VID) as a summary statistic. Forecasts for a Stage-2 CCAT survey of the cosmological [CII] signal can clearly distinguish ΛCDM, LIM astrophysics and PING model effects, placing competitive constraints on the latter two. Implementing a new CO mass-luminosity model fit to FIRE simulations, a joint Fisher forecast combining the CO VID with the three-dimensional CO power spectrum shows a marked improvement in constraining power from the joint analysis.

Together, these results establish mock observations, built from a single simulation pipeline and pointed at three different stages of cosmic history, as a disciplined way to confront fundamental early-universe physics with the noisy, foreground-laden data real surveys deliver.