Caveats — what these numbers do and do not claim.

- the reference grid on this run is the k-spacing rule's, not a converged dense reference; absolute voltages carry the metallic-reference offset
- PBE+U (Fe=4 eV, as applied) trend-level accuracy: ranks candidates within one series; not comparable to Materials Project or to experiment
- ferromagnetic approximation (true ground states are AFM): ~0.01-0.05 V
- full delithiation average, not the practical voltage window of single-phase chemistries
- every cell — the cathode pair and the Li-metal reference alike — was sampled under one k-spacing rule (Δk = 0.3 1/Å) and therefore at its OWN grid, recorded per leg in physics.txt; these numbers compare only with runs at the same Δk
- the delithiated phase is NOT a frozen host: Li is stripped in the MACE-relaxed cell and the host is relaxed, and relaxed again under QE when verify_calculation is `relax` (this run: relax). The frozen-host bias is MEASURED per candidate as host_relaxation_v in table.csv and is not corrected
