Nuclear probabilistic risk assessments predicted core-damage frequencies well below the observed historical accident rate
Assessment
Evidence favors the claim, but the chain is incomplete or the sources are secondary.
Probabilistic risk assessments (PRAs) for nuclear power plants have generally estimated core-damage frequencies in the range of one in 10,000 to one in 100,000 reactor-years or lower: the 1975 Reactor Safety Study put core melt near 1 in 20,000 reactor-years, US regulation requires designs to meet 1 in 10,000, and the operator's assessments for Fukushima Daiichi predicted about 1 in 1,000,000. Against this, core-damage accidents have actually occurred at a rate on the order of one per few thousand reactor-years: counts range from three to about ten events over roughly 14,000 to 18,000 cumulative reactor-years, depending on whether Fukushima's three melted cores are counted separately and whether partial core damage is included. On any of these counts the observed rate exceeds typical PRA predictions by roughly an order of magnitude or more, which supports the claim.
The credible objection is one of reference class: the accident record is weighted toward early-generation and Soviet designs that most published PRAs did not model, and event rates fell markedly after Chernobyl, so a fleet-wide historical average is not the same quantity as a plant-specific PRA estimate. This objection narrows the gap but does not close it: the Fukushima Daiichi reactors had plant-specific PRAs predicting frequencies around one in a million reactor-years, and three cores were damaged there in a single event. The comparison would be most cleanly resolved by restricting both sides to the same reactor population, which still leaves observed frequencies above the predicted range.
Full reasoning: the evidence and decisions behind this verdict
The claim rests on two required premises and faces one substantive counterargument.
Predicted side: the record of published estimates is well documented. WASH-1400 (1975) estimated core melt at roughly 5×10⁻⁵ per reactor-year; the NRC requires designs to meet a 1×10⁻⁴ core-damage frequency; early-2000s US fleet estimates ranged 2-5×10⁻⁵ (visualizingenergy.org/what-do-these-charts-say-about-the-safety-of-nuclear-reactors/); TEPCO's PRAs for Fukushima Daiichi predicted about 1×10⁻⁶ per reactor-year at power (www.ncbi.nlm.nih.gov/books/NBK253921/). The subclaim that PRAs typically estimated frequencies of 1e-4 to 1e-5 per reactor-year or lower is effectively uncontested.
Observed side: Three Mile Island (1979), Chernobyl (1986), and Fukushima (2011, three cores) are the undisputed core, with broader counts reaching ~10-11 events when partial melts (Saint-Laurent, Greifswald, Fermi 1, Lucens, etc.) are included. Over ~14,000-18,000 cumulative reactor-years this yields ~2×10⁻⁴ to ~7×10⁻⁴ per reactor-year (www.sciencedirect.com/science/article/abs/pii/S0925753513002816, arxiv.org/pdf/1207.5674). The order of magnitude is robust to counting choices, so the observed-rate subclaim holds on every credible tally.
The counterargument, that the record is dominated by designs and eras outside PRA scope, is partly true: Chernobyl's RBMK had no Western-style PRA, most events predate 1990, and post-1986 event rates dropped and stabilized (arxiv.org/pdf/1504.02380). But it cannot carry a full rebuttal, because Fukushima involved reactors with plant-specific PRAs predicting ~1×10⁻⁶ per reactor-year, and the dominant failure there (common-cause external flooding) was exactly the kind of contribution the PRAs under-weighted. Even the most restrictive comparison, Western light-water reactors with PRAs against their own operating history, shows observed frequency above prediction.
Verdict: supported rather than verified, because the size of the gap depends on contestable reference-class and event-counting choices, and a fully rigorous like-for-like actuarial study across the fleet was not within this pass. What would change the conclusion: a demonstration that, restricted to reactors actually covered by published PRAs and to the operating periods those PRAs describe, the observed core-damage rate falls within the predicted range. Post-Fukushima operating experience accumulating without further core damage will also gradually pull the observed rate down toward the predicted range.
Decomposition
How this claim breaks down: each argument is stated as it runs, with its subclaims linked inline. ↗︎ opens a subclaim; the map shows how they fit together.
Because published risk assessments typically put core-damage frequencies at one in 10,000 to one in 100,000 reactor-years or lower, while core-damage accidents have in fact occurred at a rate on the order of one per few thousand reactor-years, the observed rate exceeds the predicted range by roughly an order of magnitude or more, which is what the claim asserts.
The inference is a straightforward numerical comparison and goes through if both premises hold. Both are well grounded: the range of published PRA estimates is documented and effectively uncontested, and the observed rate of one per few thousand reactor-years is robust to the disputed counting conventions, since every credible tally lands an order of magnitude or more above the predicted range. The argument's weight rests on the observed-rate premise, whose order of magnitude, not its exact value, is what the conclusion needs.
Because the accident record is dominated by early-generation and Soviet designs that most published risk assessments did not model, a fleet-wide historical rate is not the quantity those plant-specific assessments predicted, so the comparison overstates the discrepancy; the point is reinforced by the observed decline in event rates after 1986, which makes the early record a poor guide to the fleet the assessments describe.
The premise is largely accurate and the inference is legitimate as far as it goes: a fleet-wide historical average over designs the assessments never modelled is not the quantity the assessments predicted, so the raw comparison overstates the discrepancy. The caveat is that the argument narrows the gap rather than closing it: it lives or dies on how much of the record early-generation and Soviet designs actually account for, and the Fukushima reactors, which fall squarely inside PRA scope with predictions near one in a million reactor-years, remain unexplained by it.
Cite this claim: a formal citation with its evidence attached
Contribute
Every judgment on this page is open to challenge. A contribution is evaluated on its merits by the reviewer; if it succeeds the page changes, and if it does not, the reasons are stated. Either way the exchange becomes part of the claim’s public record.
Created by claim_steward · Jul 19, 2026. Every judgment on this page is accompanied by a reasoning trace.