Here are some plots made with gwinc with Kevin's help of the effective loss for squeezing from different mode mismatches.  

The top left plot shows a reasonable guess of what the mismatch sizes were in O4: 2.5% sqz to OMC mismatch,  2.5% SRC to arm mismatch (G2600967), and 1% arm to OMC mismatch: G2600435 (indicated it could be as bad as 1%, or better).  How these three combine to make an effective loss for squeezing is frequency dependent and depends on the phase of the mismatch, which we do not know.  This results in up to 10% effective loss at a few kHz. 

The top right shows what happens if the sqz to OMC mode matching is improved to 0.3%, which we hope it will be after our recent psams changes (91551).  In this case the effective loss at high frequency varies from 5% to 1.6%.  

If the IFO to OMC mode mismatch is also improved to 0.3%, in which case the higher range of this loss is 2%.  When the two mismatches have similar size, it appears that they can cancel each other out to give very low loss.  In reality, these mode mismatches are likely limited by astigmatism and beam quality, and don't have a single mismatch phase like what is used in this model, and can't produce this neat cancelation effect, and the lowest loss options here won't be realistic.

The last plot shows that if we also reduced the arm to SRC mismatch to 0.3%, that has a small impact on the effective loss.