At 6:05 p.m. on August 1, 2007, the center span of the I-35W bridge over the Mississippi River in Minneapolis gave way. Construction crews had spent the afternoon staging gravel, sand, and equipment on the two closed southbound lanes, nearly 580,000 pounds in all, ahead of a concrete pour scheduled for that evening. A security camera at a lock and dam downstream recorded it: about 1,000 feet of deck truss collapsed at once, carrying 111 vehicles, and some 456 feet of the main span fell 108 feet into the river. Thirteen people died; 145 were hurt.
The National Transportation Safety Board traced the collapse to steel connector plates, called gusset plates, at a joint its engineers designated U10. When the bridge's designer, Sverdrup & Parcel and Associates, sized those plates in the 1960s, it calculated only the tension passing through the nearby chord splice and never added the shear forces the diagonal and vertical truss members also placed on the same connection, a step the 1961 design code required. The same firm had run that fuller calculation correctly elsewhere, on its floor trusses. At U10, and at two further joints built just as thin, the board concluded the calculation had never been performed at all. The result was a half-inch steel plate that, by the board's analysis of the design loads, provided only about half the resistance its design called for.
That error should have surfaced the first time anyone recalculated the bridge's safe capacity. Minnesota rated the bridge three times in the decades that followed: in 1979, after a deck overlay added more than three million pounds of concrete; in 1995, using a newly adopted load-factor method; and in 1997, ahead of a median-barrier replacement that added over a million pounds more. None of the three rated the gusset plates. Minnesota's rating software had no field for a connection's own strength, and when the board checked bridge-rating programs across fourteen states, it found none that did. The shared assumption, that a gusset plate is always stronger than the beams it joins, was one no earlier bridge collapse had ever forced anyone to question. The 1995 rating's own weakest point, on paper, was a beam in the bridge's south approach spans, nowhere near the truss. The two joints that had been built, since the 1960s, with only about half the capacity their own design required never appeared on a rating sheet.
That afternoon's construction staging added one more load directly over those joints, and the plates gave way under weight they had never had the strength to carry. The board later wrote that had the national guidance for bridge ratings included gusset plates at all, Minnesota would have had multiple chances to catch the error. It never had one, because the question the rating existed to answer did not cover the part of the bridge that failed.
A load rating is supposed to be a promise that someone checked. Minnesota's own ratings had small errors too (a mismeasured barrier dimension, a missing item in a dead-load tally), but none of that is why the bridge fell. The rules the engineers correctly followed never asked about a gusset plate's strength at all. The bridge did not fail its rating. It failed a question the rating had never been built to ask.