Context. The generation of energetic, non-thermal electrons during solar flares plays a critical role in energy transportation from the corona to the chromosphere, producing regions of observed intense X-ray emission. Turbulence in post-flare loops, particularly from Kelvin–Helmholtz instabilities (KHIs), has been suggested and investigated as a mechanism for trapping and accelerating electrons in such scenarios. Aims. Starting from past results, we aim to characterise the energisation process of electrons trapped in a turbulent post-flare looptop, quantifying the contributions of different acceleration mechanisms, and establishing a coherent numerical framework for describing particle energetics. Methods. We performed test-particle simulations with the guiding-centre approximation in addition to a 2.5D magnetohydrodynamic model of a time-evolving, post-flare coronal looptop. We implemented an improved formulation of the guiding-centre equations, which explicitly conserves energy, enabling a consistent analysis of electron acceleration in the turbulent plasma. Results. We find that, in the plasma turbulence inside the looptop, electrons develop supra-thermal energy distributions with tails compatible with hard X-ray emission. The dominant energisation channel arises from perpendicular gradient effects in the form of second-order, Fermi-like stochastic acceleration, while curvature effects are dominant for particles on long trajectories. Statistical correlations with the measured particle pitch angle confirm that the strongest acceleration occurs for electrons trapped in bouncing motions within turbulent magnetic structures. Conclusions. Our results provide an understanding of how KHI-induced turbulence in coronal looptops produces and sustains populations of trapped non-thermal electrons, supporting the interpretation of observed X-ray sources. We dissect and clarify the relative roles of different magnetic effects and the emergence of stochastic, Fermi-like energisation. We also demonstrate the advantages of the improved guiding-centre-approximation (GCA) formalism on a simple reproducible test, for the future benchmarking of GCA implementations.