Weak chemical tagging recovers a star’s birth population from its chemistry alone. Below [Fe/H] = −0.5 the thick disc, the in-situ halo and accreted debris overlap in the planes used to distinguish them, and phase-mixing limits what their dynamics

can add. This thesis asks what chemistry alone can resolve in the metal-poor Milky Way. Dimensionality reduction and Gaussian clustering applied to 17 abundance ratios for 9, 923 metal-poor GALAH DR3 giants return ten chemically distinct groups, nine of them known components, from the discs to the in-situ halo and Gaia-Enceladus-Sausage. The five in the thick disc share its metallicity and kinematics, and its α-enhancement to within their scatter, but separate in Zn, Ba and Sc, so its reported homogeneity does not extend to the iron-peak and neutron-capture ratios. Asteroseismic age information is then transferred from APOGEE DR17 to GALAH DR3 for the first time. Almost all groups peak between 8.6 and 9.0 Gyr except one, dominated by red clump stars, though a median age uncertainty of 2.4 Gyr leaves them unresolved from one another. Since neither chemistry nor age settles whether the groups are populations, the pipeline is benchmarked on mock stars from the accreted systems of 244 Milky Way Analogues in L-Galaxies, which follows all 17 ratios across more hosts than hydrodynamics can reach. Single systems are split across groups, because chemical enrichment traces a curve through abundance space and Gaussian components are straight. In the observations the structure we find survives a non-linear embedding and GALAH’s uncertainties, but one group proves to be a red clump selection rather than a population. Weak chemical tagging with a mixture model characterises accreted populations reliably, though the group count it returns is an upper limit.