Where have you shown it could not fail in the model? You've so far only presented an argument based on hand calculations and did not include all the connections and distortions modelled by NIST in your hand calculations.
NIST showed in their analysis that the beam k3004 expanded only to the West. This assumption relies on the C38 connection remaining. The beam was removed along with the girder and the girder was removed because the flange was assumed to be incapable of bearing the load once the girder had shifted.
If you want a citation for that - read the report.