The PR mostly looks good. A few questions:

1. Can we get an output.ref on the test?
2. On the linked issue (Wrong guess_mix when guess=sad #2410), you said this is a "tentative" fix. Does that mean anything besides that the PR isn't accepted yet?
3. I'll hold off on accepting until the "documentation" issue is settled.

2. On the linked issue ([Wrong guess_mix when guess=sad #2410](https://github.com/psi4/psi4/issues/2410)), you said this is a "tentative" fix. Does that mean anything besides that the PR isn't accepted yet?

It means I was able to quickly identify the issue and throw together a simple fix, but not to test it ;) I still don't have a working Psi4 compile environment due to the libint2 incompatibility. Once that gets fixed, I'll have to push to get newer versions of Psi4 to compile again in Fedora.

So... looks like it works; DIIS just takes it to the wrong solution

                          Total Energy        Delta E     RMS |[F,P]|

   @DF-UHF iter SAD:    -0.70872378254081   -7.08724e-01   0.00000e+00 
  Mixing alpha HOMO/LUMO orbitals (1,2)

   @DF-UHF iter   1:    -0.82640393434897   -1.17680e-01   1.12200e-03 DIIS
   @DF-UHF iter   2:    -0.97800267490794   -1.51599e-01   1.18746e-02 DIIS
   @DF-UHF iter   3:    -0.82673929652084    1.51263e-01   9.92192e-04 DIIS
   @DF-UHF iter   4:    -0.82768446534307   -9.45169e-04   2.11271e-03 DIIS
   @DF-UHF iter   5:    -0.82648408142877    1.20038e-03   9.77585e-06 DIIS
   @DF-UHF iter   6:    -0.82648447847074   -3.97042e-07   3.83550e-05 DIIS
   @DF-UHF iter   7:    -0.82648407869393    3.99777e-07   1.24969e-06 DIIS
   @DF-UHF iter   8:    -0.82648407827585    4.18083e-10   7.42903e-08 DIIS
   @DF-UHF iter   9:    -0.82648407827446    1.38667e-12   8.54867e-10 DIIS
   @DF-UHF iter  10:    -0.82648407827446    7.77156e-16   1.13025e-12 DIIS
  Energy and wave function converged.

OK, tests appear to pass now with all guesses.

I can't generate output.ref so I would appreciate if someone can compile the branch locally and contribute the missing files.

Running the same test but with different options is a perfect reason to write a pytest rather than a C-test, and no output.ref are needed in that case. If you can't even compile Psi, I can write the Pytest myself and submit a PR to your branch, because at least I can test it easily.

Thanks for the debugging thus far!

Accidentally pushed rather than made a PR to your fork. Oops. LMK if you object, @susilehtola.

Accidentally pushed rather than made a PR to your fork. Oops. LMK if you object, @susilehtola.

that's fine