Theoretical models often differ significantly from measured data in their predictions of the magnitude of nuclear reactions that produce radionuclides for medical, research, and national security applications. In this paper, we compare a priori predictions from several state-of-the-art reaction modeling packages (CoH, EMPIRE, TALYS, and ALICE) to cross sections measured using the stacked-target activation method. The experiment was performed using the Lawrence Berkeley National Laboratory 88-Inch Cyclotron with beams of 25 and 55 MeV protons on a stack of iron, copper, and titanium foils. Thirty-four excitation functions were measured from 4–55 MeV, including the first measurement of the independent cross sections for $$^{\mathrm{nat}}\hbox {Fe}$$
nat
Fe
(p,x)$$^{49,51}\hbox {Cr}$$
49
,
51
Cr
, $$^{51,{\mathrm{52m}},{\mathrm{52g}},56}\hbox {Mn}$$
51
,
52
m
,
52
g
,
56
Mn
, and $$^{{\mathrm{58m,58g}}}\hbox {Co}$$
58
m
,
58
g
Co
. All of the models, using default input parameters to assess their predictive capabilities, failed to reproduce the isomer-to-ground state ratio for reaction channels at compound and pre-compound energies, suggesting issues in modeling the deposition or distribution of angular momentum in these residual nuclei.