However, their presence indicates that the designers were concerned about inadequate stiffness in the girder's fully seated, symmetrically loaded position.
Indeed, such a stiffener is almost always meant to prevent buckling of the web, not the deformation of the flange. Since the bearing reaction on a girder is concentrated at its endpoints, those regions of the web are more susceptible to out-of-plane buckling. Further, for many load profiles, the shear within the web builds up at the girder bearings. Thus the same web stiffeners at the bearing ends prevent this internal shear from causing diagonal buckling in the web near the bearings.
AE911truth offers nothing but unsubstantiated speculation that the plates were also strong enough to prevent failure if the full load was shifted to only one side of the flange...
Or even intended to do so. The load-bearing element of a I-section beam or girder is the web, not the flange. The flange exists principally to keep the web in plane along its length. Most deformations of the web substantially reduce the load-bearing capacity of the member.
The AE911T scenario proposes that the bearing stiffener is meant to allow one side of the flange to accept the girder's reaction load after a lateral displacement by preventing the flange from deforming as the footprint of the web exits the bearing seat. They contend that by omitting the stiffener from some models, NIST's analysis allowed the girder-column connection to fail in a condition that, according to them, it would have otherwise survived for longer. They further contend that this omission was intended, in order to provide a straw-man collapse initiation theory.
But this fails for two principal reasons. First, by the time the girder displaces laterally such that the bearing reaction is borne entirely by one side of the girder flange -- and thus invoke the supposed reserve load-bearing intent of the bearing stiffener -- the girder's center of gravity will already have substantially passed the edge of the bearing seat and thus the combination of gravity load and reaction load will tend to rotate the girder. Hence this is not a credible design intent for the stiffener, and thus AE911T's belief in unaccounted reserve strength in the model is unfounded.
Second (and I'm working from memory here), the bearing stiffener does not attach to both upper and lower flanges. Lack of attachments to the flanges is a hallmark of web-only stiffeners. It is true that attachment of the web stiffener to the lower flange allows the portion of the bearing reactions borne by the flange
across its total width to be transmitted to the web by the stiffeners. But it is
not true that such an arrangement works with an eccentric load such as would occur at only one side of the flange. It works only when both stiffeners have an equal (but practically opposing) axial load.
In the AE911T scenario, an eccentric load on one side only of a flange applies a load through the stiffener to the web in an out-of-plane fashion, borne only by the yield strength in the thickness of the web, increasing (not decreasing) the tendency to buckle the web --
especially when the stiffener does not rise the full height of the web. Instead in the expected case, with a roughly equal portion of the total bearing load on both flange sides, the loads thus transferred from flange to the bearing stiffeners act out-of-plane but in practical symmetrical opposition through the effective section of the stiffeners
on both sides, and are borne easily by compression through the thickness of the web at the site of the stiffener attachment. The web itself does not (and cannot) bear those transmitted loads asymmetrically.
The accusation that NIST's failure to include the web stiffener in their computer model means the girder-column connection in the model was less robust than a real life walk-off hypothesis is simply not supported by design practice or careful analysis. It is based on its proponents' speculation of design intent that is not supported by design practice or analysis.
As I've said for many months now, the accusation is meant simply to fool laymen into thinking the accusers have some point that the relevant industries and sciences are expected to address. It appeals to the lay public because it can be expressed simply as discrepancies between one visual representation and another, the congruence of which can be judged without expertise. The implications, however, of that incongruence rely upon expert interpretation to determine how relevant they are to the fidelity of a computer model and how important they are in the further understanding of a forensic structural analysis.
AE911T is all to eager to provide a semblance of that expert interpretation, but one that only seems valid upon cursory reading and does not stand up to evaluation by the industry and relevant sciences. I believe the proponents of that interpretation understand the likely fate their theories would have in the hands of competent structural engineers, which is why they stick exclusively to grass-roots efforts and do not present their findings to the industry and relevant sciences through the customary channels and methods.