In Fe-Pd alloys, the competing geometric (fcc versus bcc) and magnetic tendencies result in rich phase stability and ordering physics. Here, we study these alloys via a first principles mixed-basis cluster expansion (CE) approach. Highly accurate fcc and bcc CEs are iteratively and self-consistently constructed using a genetic algorithm, based on the first principles results for ∼100 ordered structures. The structural and magnetic "filters" are introduced to determine whether a fully relaxed structure is of fcc/bcc and high-/low-spin types. All structures satisfying the Lifshitz condition for stability in extended phase diagram regions are included as inputs to our CEs. We find that in a wide composition range (with more than 1/3 atomic content of Fe), an fcc-constrained alloy has a single stable ordered compound, L1 0 FePd. However, L1 0 is higher in energy than the phase-separated mixture of bcc Fe and fcc-FePd 2 (β2 structure) at low temperatures. In the Pd-rich composition range, we find several fcc β2-like ground states: FePd 2 (β2), Fe 3 Pd 9 , Fe 2 Pd 7 , FePd 5 , Fe 2 Pd 13 , and FePd 8 , yet we do not find FePd 3 with the the experimentally observed L1 2 structure. Fcc Monte Carlo simulations show a transformation from any of the attempted β2-like ground states directly into a disordered alloy. We suggest that the phonon and/or spin excitation contributions to the free energy are responsible for the observed stability of L1 2 at higher temperatures, and likely lead to a β2 ↔ L1 2 transition. Finally, we present here a complete characterization of all the fcc and bcc Lifshitz structures, i.e., the structures with ordering vectors exclusively at high-symmetry k points.