Safety: Defense in Depth, Radiation, and the Three Accidents Everyone Asks About

Nuclear safety engineering starts from a blunt premise: machines fail, people err, and the design must absorb both. The organizing philosophy is defense in depth — no single barrier is trusted, so barriers are stacked. The fuel itself is a ceramic that traps fission products in its crystal lattice; around it, the sealed zirconium cladding; around that, the steel reactor vessel; around that, the containment building — steel-lined concrete a meter or more thick, designed for the pressure of a complete pipe rupture. In parallel run redundant and independent safety systems: multiple emergency cooling injection paths, multiple emergency diesel generators, and a scram system resting on the physics gifts from the chain-reaction page — delayed neutrons and negative feedback. The newest designs extend the philosophy to passive safety: the AP1000s at Vogtle cool their containment by gravity-fed water and natural air circulation, removing decay heat for days with no pumps and no power at all.

The regulator enforcing all this — the Nuclear Regulatory Commission — is covered on its own page, but one operational fact belongs here: every US plant is continuously inspected by NRC staff stationed at the plant, and the daily power report those inspectors' agency publishes is literally the data source for this map's live layer.

Radiation, calibrated honestly. Radiation is energy emitted by unstable atoms, it is everywhere, and dose is measured in millisieverts (mSv). The average American receives about 3 mSv per year from natural background (cosmic rays, radon, the potassium in your own body) — roughly double that including medical imaging; a chest CT is ~7 mSv; airline crews pick up a few extra mSv annually from altitude. Against that scale: living next door to a nuclear plant adds about 0.01 mSv per year — a hundredth of a single coast-to-coast flight's worth. US plant workers are limited to 50 mSv/year and typically receive a small fraction of it. The lowest dose with clearly demonstrated health effects sits around 100 mSv of acute exposure; regulation conservatively assumes risk scales linearly all the way to zero, which is prudent policy but worth understanding when small doses make headlines. Per terawatt-hour generated — counting accidents — nuclear's death rate is on par with wind and solar and hundreds of times lower than coal, whose air pollution kills continuously and invisibly.

Three Mile Island, 1979. A stuck-open relief valve drained coolant from TMI Unit 2's brand-new PWR; confusing instrumentation led operators to throttle emergency cooling; roughly half the core melted. And then: containment worked. Radioactive release was minimal — average dose to the surrounding population was about 0.01 mSv, a day of background — and epidemiology found no attributable health effects. TMI killed no one, but it transformed the industry: operator training, control-room human factors, and the industry's own watchdog (INPO) date from its aftermath. The accident's deepest irony is current events: TMI Unit 1, the undamaged twin that ran until 2019, is being restarted under a 20-year Microsoft power agreement — as the Crane Clean Energy Center, visible in this map's deals layer.

Chernobyl, 1986. The worst nuclear accident in history was a design that could not be licensed in the West meeting a test run that violated its own procedures. The Soviet RBMK used graphite moderation with water cooling, giving it the property the chain-reaction page flagged as disqualifying: under low-power, low-flow conditions, boiling increased reactivity — positive feedback. During a botched test, Unit 4's power excursion ruptured the core; with no containment building, the burning graphite lofted fission products across Europe. Two workers died that night, 28 more from acute radiation within months; the most rigorous UN assessments attribute several thousand eventual thyroid cancers (overwhelmingly treatable) to contaminated milk, with long-term population-wide effects too small to distinguish statistically. Chernobyl is genuinely terrible — and genuinely inapplicable to vessel-and-containment light-water reactors with negative feedback, which is every reactor on this map.

Fukushima Daiichi, 2011. A magnitude-9 earthquake; every operating unit scrammed correctly. The 14-meter tsunami that followed overtopped the seawall and drowned the basement-mounted diesel generators — a station blackout, the loss of all AC power. Without cooling, decay heat (that 6–7%-and-falling afterglow) melted three cores over the following days; hydrogen generated by overheated cladding exploded in the reactor buildings, producing the footage the world remembers. Radiation killed no one in the immediate aftermath; Japan attributes one later worker death to radiation-linked cancer. The chaotic evacuation of 150,000+ people, by contrast, caused well over a thousand deaths among the elderly and ill — a hard lesson that disaster response carries its own risks. Fukushima's engineering lessons — flood-hardened backup power, portable emergency equipment staged nationwide (the US "FLEX" program), filtered venting — were retrofitted across the world fleet.

Three accidents in seven decades of commercial power, one of which killed no one and another of which was a design excluded from the Western fleet: that is the actual record against which the technology's risks should be weighed — alongside, fairly, the genuinely hard problem it does have, which is the next page.