E12 — external validation vs CAMELS CV_0 (Arepo, LCDM)

Our PM evolved from CAMELS’ own 2LPT ICs (z=127→0, Ωm=0.3, 256³, 25 Mpc/h box). Node-rounding sanity: fraction of IC displacements beyond 0.45 voxels = 0.037.

• Tier 1 (truth quality): our PM vs Arepo, matched IDs: median offset 0.407 Mpc/h.

• Tier 2 (external stress test): ZA 1.140 vs damping model 1.127 Mpc/h median vs Arepo truth (0.1 Mpc cells — 10× beyond tested resolution range).

k (h/Mpc)	r our-PM	T our-PM	r ZA	r damp
0.35	0.999	0.918	0.982	0.983
0.66	0.984	0.897	0.924	0.925
1.26	0.937	0.859	0.702	0.711
2.40	0.837	0.864	0.144	0.202
4.57	0.574	0.795	0.017	0.012
8.69	0.295	0.709	-0.001	-0.002

Note on suite selection for the second-code check

SIMBA_DM CV_0 was downloaded first but found to be byte-identical to IllustrisTNG_DM CV_0 on the CAMELS server (range-request hash check): the DM-only CV “twins” of those two suites are one shared simulation. Astrid_DM CV_0 (MP-Gadget) is a genuinely distinct run (different bytes and size) and is used as the independent second code in E12-astrid.