Problem with RAS-CI calculation

I started toying around with atomic RAS-CI. Is there something wrong with the following input

molecule {
0 5
Fe
}

set basis pcseg-0
set reference rohf
set scf_type pk
set mcscf_type conv

# Ar core
set restricted_docc [3, 0, 0, 0, 0, 2, 2, 2]
# spd active space
set ras2 [3, 1, 1, 1, 0, 1, 1, 1]
cas_e, cas_wfn = energy('rasscf', return_wfn=True)

# Dynamical correlation from Ar core
set restricted_docc [0, 0, 0, 0, 0, 0, 0, 0]
set ras1 [3, 0, 0, 0, 0, 2, 2, 2]
ras_e, ras_wfn = energy('detci', ref_wfn=cas_wfn, return_wfn=True)

The output from the second part looks very weird:
==> Starting CI iterations <==

    H0 Block Eigenvalue = -1259.69953111

    Simultaneous Expansion Method (Block Davidson Method)
    Using 1 initial trial vectors

     Iter   Root       Total Energy       Delta E      C RMS

   @CI  0:     0   -104.599203359018   -1.0460E+02   1.7097E+00
   @CI  1:     0   -114.034313448065   -9.4351E+00   2.8252E+01
   @CI  2:     0   -126.656421888822   -1.2622E+01   1.2979E+01
   @CI  3:     0   -129.395267421081   -2.7388E+00   9.7215E+00
   @CI  4:     0   -131.864369339767   -2.4691E+00   1.1087E+01
   @CI  5:     0   -144.156387935524   -1.2292E+01   2.5547E+01
   @CI  6:     0   -152.844222654133   -8.6878E+00   1.0207E+01
   @CI  7:     0   -154.032962698529   -1.1887E+00   2.7989E+00
   @CI  8:     0   -154.146131021761   -1.1317E-01   2.2249E+00
   @CI  9:     0   -154.270646128935   -1.2452E-01   2.5911E+00
   @CI 10:     0   -158.914747893258   -4.6441E+00   2.6410E+01
   @CI 11:     0   -170.712879724983   -1.1798E+01   1.7935E+01
   @CI 12:     0   -175.721291161824   -5.0084E+00   1.0286E+01
   @CI 13:     0   -176.954792530500   -1.2335E+00   5.8794E+00
   @CI 14:     0   -177.398912783427   -4.4412E-01   2.7091E+00
   @CI 15:     0   -177.468521952843   -6.9609E-02   1.1821E+00
   @CI 16:     0   -177.486922248949   -1.8400E-02   8.1383E-01
   @CI 17:     0   -177.510798336187   -2.3876E-02   1.4520E+00
   @CI 18:     0   -177.679485706644   -1.6869E-01   3.7443E+00
   @CI 19:     0   -178.051070273708   -3.7158E-01   3.6720E+00  
   @CI 20:     0   -178.247547328802   -1.9648E-01   2.2715E+00  
   @CI 21:     0   -178.343356506871   -9.5809E-02   1.6707E+00  

The H0 block eigenvalue looks alright, but it’s as if detci is screwing up its guess vector, since the energy is 1100 Hartree too high…