Steel truss bridges are designed with specific load paths in mind. Each member — chord, diagonal, vertical — carries forces calculated by the engineer to flow through the structure from deck to foundation. The design specifies which members are in tension, which are in compression, and by how much. Remove a critical member, and the load path breaks. According to the design model, the bridge should collapse.
Many don't. José M. Adam and colleagues at the Polytechnic University of Valencia and the University of Vigo spent three years investigating why, using a scaled physical model of a Pratt-type railway truss bridge and over two hundred virtual failure scenarios (Nature, 2025). They found that when a primary member fails, the bridge activates six secondary resistance mechanisms that were never part of the design: global twisting of the entire structure, lateral distortion, hinge-like articulation at joints, local deformation of members adjacent to the failure, membrane action in the deck, and catenary action in remaining members.
These mechanisms are fundamentally different from the primary load paths. Twisting redistributes forces through three-dimensional coupling that a planar truss analysis ignores. Hinge formation allows controlled rotation at joints that were designed to be rigid. Lateral sway uses the bridge's width — a dimension the two-dimensional design model treats as irrelevant — to absorb loads that the designed path can no longer carry.
The critical detail: these mechanisms were always available. The geometry that permits global twisting existed from the day the bridge was built. The joints that can form hinges could always have formed hinges. The lateral stiffness was always present. None of these were designed into the bridge. They are consequences of the structure's three-dimensional geometry that a two-dimensional design analysis excludes from consideration. The bridge has capabilities its engineers never calculated.
The gap between design intent and structural reality is the margin of survival. Engineers calculate primary paths using idealized models — pinned joints, planar loading, elastic behavior. The real structure operates in three dimensions, tolerates plastic deformation, and activates coupling between modes that the design model treats as independent. When the primary path fails, the structure doesn't consult the design. It finds whatever path the geometry permits.
The structural observation: a designed system's behavior under normal conditions is a strict subset of its behavior under extreme conditions. The design captures what the engineer intended. The geometry contains what the structure can do. The two diverge precisely when the design's assumptions break — when a member fails, when loads exceed specifications, when the idealized model no longer applies. At that point, the engineer's intent becomes irrelevant. What matters is the full set of structural behaviors the geometry supports, including the ones no one calculated.
This inverts the usual framing of engineering safety. Safety factors are conventionally understood as excess capacity along the designed path — a bridge rated for 10 tons can carry 15 because the members are oversized. The latent mechanisms are different. They are not excess capacity along the designed path. They are entirely different paths that activate only when the designed path breaks. The safety margin lives in dimensions the design doesn't model.