The Repository Root Cause

The I-35W Bridge Did Not Fail From Neglect — It Failed From a Calculation Nobody Made

August 23, 2026 · Runtime 10:15 · Watch on YouTube

Within days, this became the story of American infrastructure: an old bridge, decades of deferred maintenance, a country that had stopped taking care of its roads. The National Transportation Safety Board considered that story directly and ruled it out — corrosion, cracking, temperature, pier movement, all excluded as causes. The real cause was signed off in 1965: three families of steel gusset plates, sized by a calculation method that worked everywhere else on the bridge, and simply never extended to cover the one thing that was different about them.

Chapters

Transcript

6:05 p.m., Interstate 35W 0:00

August first, two thousand and seven, five minutes past six in the evening, rush hour on Interstate thirty-five West in Minneapolis. The center span of an eight-lane steel truss bridge separates from the rest of the structure and drops into the Mississippi River. One thousand feet of deck truss come down. Four hundred and fifty-six feet of the main span fall a hundred and eight feet into water only fifteen feet deep. A hundred and eleven vehicles are on the section that collapses. Thirteen people die. A hundred and forty-five are injured. A security camera at a lock and dam just downstream catches part of it on video.

What everyone assumed, ruled out 0:42

Within days, this became the story of American infrastructure: an aging interstate system, decades of deferred maintenance, a bridge that was simply old. Investigators considered that story directly, and ruled it out, item by item. Corrosion damage: ruled out — and there was real corrosion on this bridge, measured at up to seventeen percent section loss on plates at a different set of joints, but the report is explicit that it played no role in the collapse. A cracked floor truss: ruled out. Preexisting cracking, temperature effects, movement in the piers: all ruled out. Whatever brought this bridge down, it was not decay.

What a gusset plate does 1:25

A deck truss carries its load through triangles of steel — diagonal members, vertical members, horizontal chords — meeting at flat steel connector plates called gusset plates, riveted to every member at the joint. At most of these joints, the plate does one simple job: pass tension straight through, from one chord section to the next. At three families of joints, named U4, U10, and L11, the geometry is different. Diagonal and vertical members converge there too, and their forces do not pass straight through the plate. They shear across it.

A method that worked everywhere else 2:03

The design firm, Sverdrup and Parcel, sized nearly every gusset plate on this bridge with one method: calculate the tension carried between chord sections, and size the plate to hold it. That method is correct, and investigators confirmed the firm knew how to go further. They had performed the full shear calculation on this same bridge's floor truss connections, and on the main truss of another bridge entirely, built around the same years. At the U4, U10, and L11 nodes, where shear from the diagonals was the larger force, that additional calculation was never performed. Not calculated wrong. Not calculated and checked and missed. Never calculated at all.

About one-half of what was required 2:49

Investigators rebuilt what those plates could actually carry. At the U4, U10, and L11 nodes, the ratio of demand to capacity for shear exceeded two — the load asked more than twice what the plate was rated to hold. Across every kind of stress the report checked, these gusset plates supplied, in the investigators' own words, about one-half of the resistance the design loads required. Twenty-four plates, across these three node types, every one of them a half inch of steel. Every other gusset plate on this bridge held. These did not, because they were never sized for the job they were doing.

Two things it wasn't 3:30

Investigators tested two other explanations for that thin steel, and ruled out both. Half of this bridge's members were originally meant to use a stronger, more expensive steel, later dropped for cost — but not at these three node families, whose specification never changed from the earliest drawings through construction. And the plate's unsupported edge, thirty inches of it on half an inch of steel, broke the stiffening ratio the era's code allowed by a quarter. That is a second, independent red flag, sitting in the drawings the entire time, that no design check ever caught. Even that flag would not have saved the bridge on its own: investigators found that adding the required stiffener would have done almost nothing for the actual overload. Two separate mistakes, in the same three joints, and neither one by itself was the whole story.

The check that checked the wrong thing 4:24

This is the root cause the report settles on. Sverdrup and Parcel's own quality-control procedures listed exactly which calculations a gusset plate required. What those procedures never did was verify that every required calculation had actually been produced before a plate went onto the drawings. The firm's design process involved several rounds of checking, back-checking, and rechecking — a real process, not a rubber stamp — but every round of it applied to work that had actually been submitted. A checker can catch a wrong number on a page. Nothing in this process was built to notice a page that was never written.

Forty years of adding weight 5:04

For four decades, the little margin that existed only shrank. In nineteen seventy-seven, a two-inch concrete overlay went onto the deck to fight corrosion in the rebar beneath it — a repair, meant to extend the bridge's life. It added more than three million pounds, thirteen and a half percent more dead load than the original design carried. In nineteen ninety-eight, new median barriers and railings added still more. By the time investigators ran their final numbers, this bridge was carrying nineteen and a half percent more dead weight than the day it opened. All of that weight rested on gusset plates under-strength from the very first drawing.

Not one inspector's blind spot — the country's 5:46

Here is the part that is not about this one bridge. The investigation found a practice shared by federal and state transportation officials nationwide: excluding gusset plates entirely from bridge load-rating calculations. It also found officials giving little attention during inspection to exactly this kind of distortion. The investigation names this directly as a contributing cause of the accident, alongside the design error itself — not a footnote about one state's inspectors, a finding about how the whole country rated bridges. Every inspector who ever walked this bridge was working inside a system that had never once asked them to check the one component that was actually failing.

The photograph nobody flagged 6:28

There are photographs. Seven of the eight gusset plates at these joints show visible bowing along one edge in pictures from nineteen ninety-nine and two thousand three — bulging as much as an inch out of plane. A Minnesota inspector reportedly noticed it, assumed it dated to the bridge's original construction, and moved on. No note in any inspection report. No follow-up analysis. Investigators later checked that assumption against the loads generated during erection, and found it almost certainly wrong: those loads were lower than the ordinary loads on the bridge every day since. Nobody was hiding this plate. Nobody was ever asked to explain it.

578,735 pounds, staged since 2:30 7:12

On the day of the collapse, four of the eight lanes were closed for repaving. Since two-thirty that afternoon, contractors had staged materials for that night's concrete pour directly above the weak point. A hundred and eighty-four thousand pounds of gravel, nearly two hundred thousand pounds of sand, and close to two hundred thousand pounds of parked equipment and personnel. Five hundred and seventy-eight thousand pounds, sitting on the inner lanes just north of pier six, directly over gusset plates that had never had the margin to carry it.

6:05 p.m., 1 August 2007 7:45

At five minutes past six, the center span gave way exactly where it had always been weakest. Steel that had carried this freeway for forty years came down in under a minute. The wreckage did not fall straight down: the southern section twisted off its piers and came to rest eighty-one feet east of where it had stood a moment earlier, carrying the traffic on it into the river below.

The Board's conclusion 8:09

The investigators' conclusion, close to their own words: the probable cause was the inadequate load capacity, from a design error, of the gusset plates at the U10 nodes. They failed under decades of added weight combined with the traffic and construction load on the bridge that day. Contributing causes: the design firm's own quality control, and a nationwide practice of excluding gusset plates from load-rating calculations altogether.

The word that entered the checklist 8:37

In January two thousand and eight, months before this final report was even published, the Federal Highway Administration issued its first directive in the history of American bridge engineering. It required gusset plates to be included in the load rating of nonredundant steel truss bridges. Forty-three years after the drawings were signed, that one word — gusset — finally entered the checklist.

Seven episodes, seven ways of not knowing 9:02

Seven episodes now. A theory correct only within the range it had been tested in. A checklist complete for every tower built before a new shape existed. A criterion asking about frequency when the problem was stability. A safety switch never once exercised. An alarm whose silence looked exactly like good news. A connection check that existed only on paper. And now, a part of the structure that no check anywhere in the country had ever been written to cover — until this one went into the river.

Next: the bridge with no backup plan 9:37

The idea that a bridge should never depend on one part with nothing to back it up is not new. Six weeks after this exact bridge opened to traffic, on December fifteenth, nineteen sixty-seven, the Silver Bridge came down at Point Pleasant, West Virginia. Forty-six people died, from a crack a tenth of an inch deep in a single steel eyebar with no partner to share the load when it broke. That collapse is the reason bridge inspection exists in this country at all. And forty years later, a part the resulting rules still didn't check took another one down.

Description and sources

On 1 August 2007, at 6:05 p.m. during rush hour, the center span of the I-35W bridge over the Mississippi River in Minneapolis collapsed. A thousand feet of deck truss came down; 456 feet of the main span fell 108 feet into water only 15 feet deep. 111 vehicles were on the collapsing section. 13 people died, 145 were injured.

The public explanation was immediate: an aging interstate system, deferred maintenance, a bridge that was simply old. The NTSB ruled that out point by point — corrosion, a cracked floor truss, preexisting cracking, temperature, and pier movement were all excluded. The real fault traced to the 1965 design: at the U4, U10, and L11 nodes, the design firm Sverdrup & Parcel sized the gusset plates by a method that only accounted for tension between chord members. At these three node types, shear from the diagonal and vertical members was the larger force, and the additional shear calculation that required — the same one the firm used correctly on this bridge's own floor truss, and on another bridge's main truss entirely — was never performed. The 24 resulting plates provided about one-half of the resistance the design loads demanded, with shear demand-to-capacity ratios over 2.

For four decades the margin only shrank. A 1977 concrete overlay added 13.4% more dead load; a 1998 barrier upgrade brought the cumulative increase to 19.5%. Photographs from 1999 and 2003 show visible bowing in most of the affected plates — a defect a Minnesota inspector reportedly saw and dismissed as leftover from construction, an assumption the NTSB later showed was almost certainly wrong. Investigators found a nationwide practice among federal and state officials of excluding gusset plates from load ratings entirely, and of giving them little attention during inspection.

On the day of the collapse, four of eight lanes were closed for repaving, and since mid-afternoon nearly 579,000 pounds of gravel, sand, and equipment sat staged directly above the weakest joints. In January 2008 the FHWA issued its first-ever directive requiring gusset plates in the load rating of nonredundant steel truss bridges nationwide.

PRINT-READY, FROM THIS CHANNEL

The Failure Atlas, Vol. 01 — Tacoma Narrows · Citicorp Center · Millennium Bridge · Apollo 13 · the 2003 blackout · Hyatt Regency

https://therepository.gumroad.com/l/failure-atlas

PRIMARY SOURCES

NTSB, Highway Accident Report: Collapse of I-35W Highway Bridge, Minneapolis, Minnesota, August 1, 2007, NTSB/HAR-08/03, 14 November 2008 — read in full for this episode. https://www.ntsb.gov/investigations/AccidentReports/Reports/HAR0803.pdf
FHWA, Technical Advisory T 5140.29, Load-Carrying Capacity Considerations of Gusset Plates in Non-Load-Path-Redundant Steel Truss Bridges, 15 January 2008 — the first nationwide requirement to include gusset plates in bridge load ratings.
Wikipedia, "I-35W Mississippi River bridge" — cross-checked for the collapse sequence and wreckage geometry. https://en.wikipedia.org/wiki/I-35W_Mississippi_River_bridge

The 13 technical plates in this video are illustrations generated for the channel by a diffusion image model, styled to match its cyanotype identity. They are diagrams of the system, not photographs of the hardware, and no person is depicted in any of them.

Root Cause investigates why engineered systems fail, using the official investigation reports and the primary technical literature. Sources for this episode are linked above.

The technical drawings in this video are cyanotype-style illustrations produced for the channel. They are diagrams, not photographs of the real hardware or the actual bridge. The charts and dimensioned comparisons are drawn from the figures and text in the sources listed above.

#engineering #failureanalysis #rootcause #structuralengineering #i35w