Wet-gap crossing has become one of the most dangerous engineering tasks on the contemporary battlefield, because a pontoon bridge is a large, fixed, thermally and geometrically conspicuous structure that must remain in place long enough for a force to cross, all under persistent unmanned reconnaissance. This article asks a narrow question: what determines whether a crossing survives, and where should engineering effort be spent to reduce the exposure that gets bridging units killed? The study reframes survivability as a race between two clocks. The first is the total exposure time of the crossing site — the sum of assembly, trafficking, and recovery; the second is the adversary's reconnaissance-strike loop, the interval from first detection to delivered effect. Its original contribution is the Crossing Exposure Index (CEI), defined as the ratio of total exposure time to the reconnaissance-strike loop time, together with a concealment-adjusted survivability function in which probability of completing the crossing before a strike arrives decays exponentially with the exposure that overhangs the enemy loop, scaled by a concealmentreduced detection rate. The model yields a sharp design rule absent from the bridging and operations-research literatures alike: when exposure is driven below the loop time, the crossing completes inside the adversary's decision cycle and timely engagement becomes near-impossible, whereas above that threshold survivability falls off rapidly and concealment, not further speed, governs the residual. Illustrative computation across assembly times of 30 to 150 minutes, loop times of 10 to 40 minutes, and concealment effectiveness of zero to eighty percent produces modelled crossing-survival probabilities ranging from below 0.10 for a slow, unconcealed crossing against a fast loop to above 0.90 for a rapid, well-concealed crossing. The analysis concludes that exposure reduction is a systems problem: modular and parallel assembly to compress the timeline, signature management to lower the detection rate, and sequencing to keep the detectable window beneath the enemy loop, each insufficient alone. All numerical values are model-derived and require empirical and exercise validation before use.