Late Lessons, Jensen Huang and AI
Markdown version

Hindsight check: LL2-18 (Part C divider and Ch 18 Late lessons from Chernobyl, early warnings from Fukushima)#

Source section: EEA, Late lessons from early warnings: science, precaution, innovation (EEA Report No 1/2013), pp. 429–457. Pages 429–431 are the Part C divider and contents list. Chapter 18 (pp. 432–457, including references) is by Paul Dorfman, Aleksandra Fucic and Stephen Thomas. Check window: publication (2013) to late September 2026. Checked: 25–26 September 2026.

Method note. - No general web search. The session’s web-search budget was used up, so I retrieved sources directly from primary repositories: - UNSCEAR (the 2012 Annex A, the 2020/2021 Fukushima annex, the Chernobyl thyroid white paper, and its overview pages); - Fukushima Prefecture’s English portal (health survey results, casualties, evacuees); - Japan’s Nuclear Regulation Authority (NRA) and METI (decommissioning presentations of 16 September 2025); - IAEA PDFs: the Director General’s 2015 Fukushima report and the 2023 report on the treated-water discharge. The IAEA’s web pages refused automated access. - the US Federal Register and its API; EU acts via legislation.gov.uk; the Eurostat API; GOV.UK’s content API; the EBRD; press releases from the operators the chapter names (EDF, TVO); - Europe PMC, Crossref and HAL for peer-reviewed abstracts. - Wikipedia as a finder, and other secondary sources. I used English, Japanese and German Wikipedia only to locate dates and primary references, and I label a point “secondary” wherever it rests on Wikipedia or on news reporting. Several such points concern 2025–2026 court rulings and national legislative votes that I could not confirm from court or parliamentary sites. - Access limits. - The OECD Nuclear Energy Agency website returned 403, so the date the revised liability conventions entered into force comes from a secondary source. - The French Cour des comptes website was unreachable, so its 2025 EPR report is not used. - The IAEA reactor database (PRIS) is JavaScript-only, so I could not rebuild the global series of construction starts. I use the World Nuclear Association’s table of 2026 starts instead. - Scope. I read only the section digest, the section’s source extract and web sources. Page numbers are report pages of the 2013 volume. - Annex 3. Chapter 18 is a new case in the 2013 volume; the 2001 volume had no nuclear-power chapter. So no Annex 3 update applies, and I did not open Annex 3. - Part C divider. Pages 429–431 contain only titles and a contents list, so there is no argument to check. The editors’ framing is that Part C covers technologies with “very little direct hindsight” (Introduction, p. 10, outside this section). For this chapter, thirteen years of hindsight now exist. - Author position. - The authors are known critics of nuclear power (see the digest). The chapter has no counter-panel. - After 2013, Thomas published an analysis of the Hinkley Point C decision [S87]. Fucic co-authored a review that reported raised chromosomal damage in children exposed after Chernobyl [S24]. That is a biomarker finding, not a disease outcome. - None of the evidence below comes from the authors except where noted.

Overview#

The chapter is several things at once: - a health assessment of two accidents; - a critique of probabilistic risk assessment (PRA); - an economic case against new nuclear build in Europe; - a snapshot of post-Fukushima policy.

Thirteen years on, these parts have fared very differently.

1. The institutional and economic diagnoses have held up or strengthened. - Official findings. The IAEA’s own 2015 report found that “a major factor” in the accident was the “widespread assumption in Japan that its nuclear power plants were so safe that an accident of this magnitude was simply unthinkable”. It added that this assumption “was not challenged by regulators or by the Government”. It also found that the plant’s probabilistic safety assessments had not considered internal flooding and had made “optimistic” assumptions about human performance [S33]. Japan’s new regulator used the same “safety myth” language in 2013 [S35]. - Liability. The revised Paris and Brussels liability conventions, which the chapter described as proposals, entered into force in 2022 with the same €700m and €1.5bn ceilings (secondary [S45]; €700m confirmed in [S44]). - Fukushima’s costs are now officially estimated at ¥21.5 trillion (government, 2016), and at ¥35–80 trillion by an independent Japanese institute (2019) [S46, S47]. On any of these figures the cost is about 100 times the €1.5bn European ceiling. - TEPCO’s liability was unlimited, yet the state still had to become its controlling shareholder [S33]. - New build. - Olkiluoto 3 began regular output in April 2023, not 2014 [S51]. - Flamanville 3 reached the grid in December 2024 and full power in December 2025, not 2016 [S53, S54]. - The Pyhäjoki project was cancelled [S57]. - Of the UK’s “up to eight” plants, two have gone ahead. Hinkley Point C is under a revenue contract and is now estimated at £31–35bn (2015 values). Sizewell C reached a final investment decision in 2025 with a 44.9% UK government stake [S58–S60]. - Chernobyl’s legacy. Costs and risks kept rising. The New Safe Confinement was completed in 2016–2019, but a drone strike in February 2025 breached it. The EBRD now puts repairs at €500m or more, and the structure must be “fully functional again by 2030” [S80, S81].

2. The health claims about Fukushima have weakened as forecasts of radiation-caused disease. They have strengthened as warnings about harms from other causes. - Radiation. UNSCEAR’s 2020/2021 assessment found that “no adverse health effects among Fukushima residents have been documented that are directly attributable to radiation exposure”. It judged that future radiation effects are “unlikely to be discernible” [S1]. - Thyroid cancers. The prefecture’s screening programme had found 365 malignant or suspected thyroid cancers by 2025. UNSCEAR and the programme’s own evaluation subcommittee attribute them mainly to highly sensitive screening, not radiation [S1, S6, S9, S10]. A minority of researchers dispute this [S11, S12]. - Other harms. Fukushima Prefecture counts 2,351 “disaster-related deaths”, from stress and failing health during evacuation, which is more than its direct disaster deaths [S7]. Some 22,600 people remained displaced in May 2026 [S8]. - What the chapter missed. Its recommended “multi-factorial” biomonitoring was largely put in place. But the chapter did not anticipate that protective action and screening would themselves become major sources of harm.

3. Chernobyl health numbers are still contested, in the way the chapter described. - The authors’ estimate. Their collective-dose extrapolation of 17,000–68,000 cancer deaths (p. 435) is arithmetically consistent with UNSCEAR’s later collective-dose figures. It is also the same kind of calculation as the mainstream IARC-led projection [S1, S16]. - UNSCEAR’s position. It still declines to make such projections, and it advises against “multiplying very low doses by large numbers of individuals” except for notional comparisons [S3]. - Low-dose evidence. Large worker cohorts since 2013 strengthen the low-dose basis for such projections, and suggest risk per gray may be higher than protection standards assume [S18, S20]. - Yablokov. The 93,000 figure for Belarus alone has been rejected in mainstream reviews on methodological grounds [S17]. - Observed thyroid cancers. About 20,000 cases occurred in 1991–2015 among people exposed as children, with a best-estimate radiation-attributable fraction of about 0.25 [S5].

4. The critique of PRA was largely accepted, with nuance, and is now being partly reversed in one major jurisdiction. - What changed. Regulators in the EU, Japan and the US added requirements for beyond-design-basis and severe accidents, and for extreme external hazards [S33–S35, S37]. - What did not. They broadened PRA rather than abandoning it. - Recurrence rate. Empirical studies since 2013 support the order of magnitude of the chapter’s point: observed frequencies of severe accidents are far above pre-Fukushima PRA figures. The “once every 20 years” figure sits at the pessimistic end [S29–S31]. No accident on the Chernobyl or Fukushima scale has happened in the 15 years since. - Reversal in the US. A 2019 US rule dropped a requirement tied to reevaluated flood and seismic hazards. A dissenting commissioner said this “ignores primary lessons from the Fukushima Daiichi accident” [S37]. A 2025 executive order and 2026 proposed rules direct the US regulator towards “credible, realistic risks” and propose removing the “as low as reasonably achievable” (ALARA) requirement [S38–S40].

5. The policy snapshot was accurate for 2011–12 but mostly did not last. - Germany completed its phase-out in April 2023 [S65]. - Elsewhere the direction reversed: - Belgium repealed its phase-out law in 2025 [S66]. - Switzerland’s parliament voted in June 2026 to lift the 2017 ban on new build, subject to a possible referendum [S67]. - Italy’s Senate gave final approval on 23 September 2026 to an enabling law for a return to nuclear power [S68]. - The Netherlands is planning new reactors [S70]. - Globally, 38 countries have endorsed a pledge to triple nuclear capacity by 2050 [S71, S72], and the World Nuclear Association’s database already lists ten reactor construction starts in 2026 [S62]. - EU share. The chapter’s forecast that the EU nuclear share would decline “after 2020” held. The share fell from 27.6% in 2013 to 21.6% in 2022, then recovered to 23.0% in 2024 [S64]. The predicted emphasis on extending the lives of existing reactors also held [S73].

6. One of the chapter’s numbers was too high. - The chapter’s figure. It put the atmospheric release of caesium-137 at 3.5 × 10^16 Bq, about 40% of Chernobyl’s (pp. 440–441). This relied on an early independent estimate [S25]. - The later consensus is 6–20 PBq, with most estimates near the lower end, or about 20% of Chernobyl [S1]. - The official estimate. The Japanese regulator’s figure of 15 PBq falls inside that range. So the chapter’s theme that independent science corrects official figures did not hold for this particular number. - What did hold. The ocean release was the largest from any single accident [S1, S26]. The unit errors the digest flags (TBq printed for PBq) are confirmed by the published source [S25].

Fair reading. - What the hindsight supports. The chapter’s strongest lessons are institutional and economic: - regulatory capture and the “safety myth”; - probabilistic assessments that miss common-cause cascades; - liability caps that are orders of magnitude too small; - first-of-a-kind cost overruns that require the state to carry the risk; - legacies lasting many decades. - Where it is weaker. Its health framing leaned on contested and advocacy sources. It anticipated radiation disease that has not been detected, and it missed that the harms of evacuation and screening would dominate at Fukushima. - What it could not know. Its policy snapshot captured a moment and could not have seen the energy-security politics of 2022–2026. - A point that looks prescient. The authors objected that security threats (aircraft strikes, terrorism) were excluded from the EU stress tests (p. 444). The war-related threats to Zaporizhzhia and to the Chernobyl confinement since 2022 bear this out [S80–S82, S86].

Claim-by-claim#

Claim 1: Fukushima health consequences “may start to arise and be documented over the next 5–40 years”; biomonitoring must integrate cancer and non-cancer outcomes, age, gender, geographic dispersion, psychosocial status and cancer-specific latency (pp. 432, 442, 448)#

Original claim. - The chapter’s conclusion (p. 448) calls for monitoring that would: - integrate cancer and non-cancer disease data; - take a “complex approach”, considering age, gender, geographic dispersion “and the psychological, educational and social status of victims”; - evaluate latency cancer by cancer. - It says that because evacuees were dispersing across Japan, follow-up “will need to account for geographic dispersion” (p. 442). - It suggests a “complex radiochemical environment” may shape the eventual biological effect (p. 448). - The framing throughout is that radiation-caused disease is likely to emerge.

Subsequent developments - UNSCEAR’s assessments. - The 2013 report (published 2014) and three white papers (2015–2017) were followed by the 2020/2021 report (launched 9 March 2021), which “broadly confirms” the 2013 findings [S2]. - It concluded: “No adverse health effects among Fukushima residents have been documented that are directly attributable to radiation exposure… future radiation-associated health effects are unlikely to be discernible” [S1, conclusion (q)]. - Its statistical power analyses of children “did not indicate that excess thyroid cancer would likely be discernible up to either age 30 or 40”. Excess leukaemia “is still unlikely to be discernible”, although there “would be some possibility” of detection if true doses were at the upper bound [S1, paras 245, 247]. - It found no credible evidence of excess congenital anomalies, stillbirths, preterm births or low birthweight [S1, para 248]. - Collective thyroid dose was about 3% of Chernobyl’s European figure, and collective effective dose about 10–15% [S1, appendix B and para on collective dose]. - Thyroid screening results. - Fukushima Prefecture’s Health Management Survey (FHMS) screened people aged about 18 or under in 2011. The results pages were updated 23 July 2026 [S6]. They report cases of “malignancy or suspected malignancy” by round:

| Examination round | Cases |
|---|---|
| Baseline, 2011–13 | 116 |
| Round 2 | 71 |
| Round 3 | 31 |
| Round 4 | 39 |
| Round 5 | 50 |
| Round 6 (to 30 Sep 2025) | 21 |
| Age-25 milestone examinations | 27 |
| Age-30 milestone examinations | 10 |
| **Total (my sum)** | **365** |

Verdict: partly held up. - Weakened as a forecast of radiation-caused disease. After 15 of the 40 years, none has been documented, and the mainstream assessment expects none to be discernible. - Strengthened as a call to monitor many outcomes, including psychosocial ones, across a dispersed population. The chapter did list psychosocial status (p. 448). - Missed. The chapter did not foresee that evacuation and screening, both protective measures, would produce the largest measurable harms.

Implications for weight. - The general lesson carries weight: after a disaster, monitor broadly and for a long time, and do not assume the headline hazard is the only pathway to harm. - The chapter’s implied expectation of radiation disease should carry little weight. - A lesson the chapter did not draw is well supported by later evidence: protective measures and surveillance carry their own harms, and those harms need to be weighed in the same analysis.

Claim 2: Chernobyl cancer mortality is deeply uncertain: about 4,000 (Chernobyl Forum) against 93,000 fatal cancers in Belarus alone (Yablokov); the authors extrapolate 17,000–68,000 over 50 years (600,000 person-Sv × 0.057/Sv, ± a factor of 2); leukaemia latency “can exceed more than 40 years” (pp. 434–436)#

Original claim. - The chapter cites the Chernobyl Forum’s “about 4 000” for the three most affected countries. - It cites Yablokov et al. (2006) for 270,000 cancers and 93,000 deaths in Belarus. - It notes that UNSCEAR (2008) declined to project numbers “because of unacceptable uncertainties”. - It multiplies a worldwide collective dose of 600,000 person-Sv by the ICRP coefficient of 0.057 per Sv to get about 34,000 deaths. Doubling or halving this for linear no-threshold (LNT) model uncertainty gives 17,000–68,000 (p. 435). - It reports rising leukaemia among clean-up workers and in children aged 0–5 at the time of the accident, and says leukaemia latency can exceed 40 years (p. 436).

Subsequent developments - UNSCEAR’s position on projections. - UNSCEAR’s 2012 report (published 2015) states that it “does not recommend multiplying very low doses by large numbers of individuals to estimate numbers of radiation-induced health effects” at or below background levels. - It adds that such projections “could be useful” for comparisons if applied consistently, with uncertainties fully taken into account, and not presented as “other than notional” [S3, paras 30–31]. - So the chapter’s calculation is the kind of calculation UNSCEAR discourages. It is not the kind UNSCEAR forbids. - The inputs are consistent with later UNSCEAR figures. UNSCEAR’s 2020/2021 comparison table gives collective effective doses for 1986–2005 of about 200,000 man-Sv in Belarus, Russia and Ukraine and about 160,000 man-Sv in the rest of Europe [S1, appendix B]. Extended to 50 years, 600,000 person-Sv is plausible. So the arithmetic is defensible; the dispute is about whether the calculation should be made. - The mainstream projection. Cardis et al. (2006, IARC-led) projected 16,000 thyroid cancer cases (95% uncertainty interval 3,400–72,000) and 25,000 other cancer cases (11,000–59,000) in Europe by 2065. They noted that this burden would be undetectable in national cancer statistics [S16]. These are case counts, not deaths, so they are not directly comparable. The chapter’s central 34,000 deaths is at the high end of this mainstream projection, but the same order of magnitude. - Low-dose epidemiology since 2013. - INWORKS 2023 (309,932 nuclear workers in France, the UK and the US) found solid-cancer mortality rising 52% per Gy (90% CI 27–77%). The estimate was steeper still below 100 mGy, and “larger than estimates currently informing radiation protection” [S18]. - A 2020 National Cancer Institute (NCI) meta-analysis of 26 low-dose studies found that the evidence of excess risk was not explained by bias, particularly for leukaemia [S20]. - Together these strengthen, rather than weaken, the LNT basis for collective-dose projections. - Yablokov. Balonov (2012) examined the approach of Yablokov and colleagues, as set out in their 2009 New York Academy of Sciences book. He concluded that its reliance on ecological rather than analytical epidemiology “resulted in the overestimation of the number of accident victims by more than 800 000 deaths” [S17]. I found no mainstream body that endorses Yablokov’s figures. - Observed thyroid cancer. - UNSCEAR’s 2018 white paper counts about 20,000 thyroid cancers registered in 1991–2015 among those under 18 in 1986, in Belarus, Ukraine and the four most contaminated Russian regions. - It puts the attributable fraction for non-evacuated residents at “of the order of 0.25”, with an uncertainty range “at least from 0.07 to 0.5” [S5]. That implies roughly 5,000 radiation-attributable cases, with a range of about 1,400–10,000 (my arithmetic). - UNSCEAR’s 2020/2021 table says a “substantial fraction of the 19 000 thyroid cancers observed (up to 2016)” is attributable to radiation [S1, appendix B]. - Leukaemia. - UNSCEAR reports a dose-related increase in leukaemia and cataracts among higher-exposed emergency and recovery workers. In the general population, leukaemia “does not appear to be elevated” [S1, appendix B; S4]. - An ecological study in Ukraine found childhood leukaemia more frequent after the accident in contaminated regions (frequency ratio 1.59, 95% CI 1.36–1.86) [S22]. It is ecological, the design that Balonov criticises. - On latency, the atomic-bomb survivor study found radiation-associated leukaemia risks, “especially for acute myeloid leukemia, had persisted throughout the follow-up period out to 55 years” [S21]. This supports the chapter’s statement. - Heritable effects. The chapter also discussed genetic effects in offspring (per the digest). Whole-genome sequencing of 130 children born to exposed parents found no increase in germline mutations [S23].

Verdict: contested (the state of the science the chapter described persists). - Held up: the authors’ own range, as a notional projection. Its inputs match later UNSCEAR dose estimates and later low-dose epidemiology supports its method, but mainstream bodies still decline to make such projections. - Weakened: the Yablokov figure, rejected by mainstream reviewers. - Held up: the leukaemia-latency statement. - Weakened: the heritable-effects concern.

Implications for weight. - The lesson that “methodological choices shape the answer” in low-dose controversies (digest insight 10) is strongly supported. The same collective dose yields “about 4,000”, “about 34,000” or “no projection” depending on population boundaries, risk model and whether projection is thought legitimate at all. - Readers should not treat the chapter’s placing of Yablokov alongside mainstream estimates as even-handed. Any lens built on this chapter should separate the authors’ defensible extrapolation from the advocacy figures it cites.

Claim 3: Fukushima released about 3.5 × 10^16 Bq of Cs-137 to air, “roughly twice the official government figure” (NISA’s 15,000 TBq) and about 40% of Chernobyl’s release; the ocean release was the largest ever from a single accident (pp. 440–441)#

Original claim. - The chapter uses Stohl et al.’s inverse-modelling study, quoted from the 2011 discussion paper, to say releases “started earlier, lasted longer, and were therefore higher than earlier official estimates”. - It quotes deposition figures of “6.4 TBq” on Japan and “0.7 TBq” on other land (p. 440). - It also cites plutonium deposition (p. 440). It frames the gap as a case of unreliable operator and official data being corrected from outside Japan.

Subsequent developments - The published source. Stohl et al. (2012, final version) give a total caesium-137 emission of 36.6 PBq (range 20.1–53.1), “about 43% of the estimated Chernobyl emission”. They state that “6.4 PBq… were deposited over Japanese land areas” and “0.7 PBq” elsewhere [S25]. This confirms that the chapter’s “TBq” is a transcription error for PBq. - The later consensus. - UNSCEAR (2013, reaffirmed 2020/2021) put total atmospheric caesium-137 at “generally in the range of 6 to 20 PBq”. It notes that estimates made since 2013 “generally confirm the ranges… with most of the estimates at the lower end”. - Its preferred source term, from the Japan Atomic Energy Agency (JAEA, 2020), is 10 PBq caesium-137 and 120 PBq iodine-131. - Only one post-2013 estimate, based on an accident-progression code rather than environmental measurements, came close to Stohl’s: 29 PBq [S1, paras 23–24, 35–36, table 1]. - On averages of published estimates, Fukushima’s caesium-137 release was “about… 20%” of Chernobyl’s (85 PBq) [S1, para 24; appendix B]. - The Nuclear and Industrial Safety Agency’s (NISA) figure of 15 PBq, the official Japanese regulator’s estimate, lies inside UNSCEAR’s range. The independent estimate the chapter relied on lies above it. - Plutonium and strontium. UNSCEAR judged releases of “less volatile radionuclides (e.g., 90Sr and 239Pu)” to be “negligible” [S1, conclusion (a)]. - The ocean. - UNSCEAR estimates direct releases of about 3–6 PBq caesium-137 in the first three months. A further 5–8 PBq was deposited on the ocean from the atmosphere, and groundwater and rivers continued to carry smaller amounts [S1, conclusion (c)]. - For Chernobyl, the corresponding deposition figures are about 2.8 PBq on the Black Sea and about 3.0 PBq on the Baltic [S1, appendix B]. - Buesseler et al. (2012) called Fukushima’s ocean releases “unprecedented”. They found risks to marine life and seafood consumers “below those generally considered harmful” [S26, S27]. - The chapter’s “largest ever from a single accident” holds. - Later marine management. Discharge of treated water that still contains tritium began on 24 August 2023. An IAEA review in 2023 found the plan “consistent with relevant international safety standards” with “negligible radiological impact”. It stressed that its report is “neither a recommendation nor an endorsement” [S28]. By September 2025, 14 batches had been discharged [S83].

Verdict: weakened for the atmospheric magnitude and the Chernobyl ratio; held up for the ocean claim; the unit errors are confirmed.

Implications for weight. - The chapter’s broader point was that early official figures under a captured regulator deserved scepticism, and that outside monitoring matters. That point remains sound; see Claim 9 on TEPCO’s disclosure record. - But this case shows that independent estimates can also err high. The lesson that survives is about many independent measurements, made in the open, reaching agreement over time. It is not that outside estimates are right and official ones wrong. - Magnitudes from the chapter should be replaced with UNSCEAR’s ranges in any later use.

Claim 4: Observed major accidents imply recurrence “once every 20 years”, against pre-Fukushima estimates of about 1:100,000; PRA “has proven very limited” and “an urgent re-appraisal… seems overdue” (pp. 432, 447–449)#

Original claim. - The chapter argues that probabilistic assessment “can only be based on sets of pre-conditioning assumptions”. It says PRA missed the cascade from earthquake to tsunami to failures of reactors and spent-fuel pools (p. 447). - It cites Goldemberg (a 2011 blog) for a major accident every 20 years. The digest notes that the chapter’s probability arithmetic is garbled. - It endorses Bunn and Heinonen’s call for planning “for events beyond design bases”, together with international reviews and binding standards (p. 448).

Subsequent developments - Empirical frequency studies after 2013. - Ha-Duong and Journé (2014) rebuilt the accident record and found “the worldwide historical frequency of a nuclear major accident, defined as an INES level 7 event, is 14%” in a random five-year window. That is roughly one every 30–35 years (my conversion) [S29]. - Escobar Rangel and Lévêque (2014) found that a time-varying model shows “a greater increase in the risk of a core meltdown accident” after Fukushima than classic Poisson models [S30]. - Wheatley, Sovacool and Sornette (2016) estimated “a 50% chance that… a Fukushima event (or larger) occurs every 60–150 years”. They found costs heavy-tailed, with Chernobyl and Fukushima costing “nearly five times the sum of the 173 other events”, and the International Nuclear Event Scale (INES) “inconsistent” with cost and release data [S31]. - These studies support the chapter’s order-of-magnitude point: observed frequencies far exceed pre-Fukushima PRA-based figures. The “20 years” sits at the pessimistic end of later estimates. - No INES level 7 accident has occurred in the 15 years since March 2011. - The IAEA’s diagnosis. The IAEA Director General’s report (2015) found that the plant’s safety analyses “did not fully address the possibility of a complex sequence of events”. It found that “the probabilistic safety assessments did not address the possibility of internal flooding” and that human-performance assumptions “were optimistic”. It also noted that “the possibility of several reactors at the same facility suffering a crisis at the same time was not considered”. But its remedy was “a comprehensive probabilistic safety assessment” used together with deterministic analysis, not a move away from PRA [S33]. - EU reforms. Council Directive 2014/87/Euratom (8 July 2014) added: - a binding safety objective of avoiding “early radioactive releases” and “large radioactive releases”; - defence-in-depth covering “extreme external natural and unintended man-made hazards” and “severe conditions”; - Europe-wide topical peer reviews every six years; - regulator independence and transparency requirements [S34]. - Japan. The NRA’s requirements, in force from 8 July 2013, re-evaluated external hazards (earthquakes, tsunamis, volcanoes, tornadoes, forest fires). They required countermeasures against station blackout, severe accidents and “malicious airplane crash”, applied retroactively to existing plants, and were framed on “an underlying assumption that severe accidents could occur at any moment” [S35]. - United States. - The Nuclear Regulatory Commission’s (NRC) Near-Term Task Force concluded that defence-in-depth, “supported and modified as necessary by state-of-the-art probabilistic risk assessment techniques”, should remain the organising principle. - The 2019 rule on mitigating beyond-design-basis events made the post-Fukushima mitigation orders permanent. But the Commission majority removed the requirement that mitigation strategies address the reevaluated flooding and seismic hazards, along with requirements for drills and staffing. - Commissioner Baran’s dissent, printed in the rule, said this “ignores primary lessons from the Fukushima Daiichi accident”. He noted “incontrovertible evidence that the current design bases for some plants do not address a flood hazard identified by the licensees’ own analyses” [S37]. - Executive Order 14300 (23 May 2025) calls the NRC’s approach an attempt to “insulate Americans from the most remote risks”. It orders “thresholds to ensure that reactor safety assessments focus on credible, realistic risks”, 18-month licensing deadlines, and reconsideration of LNT and ALARA [S38]. - An NRC proposed rule of 15 July 2026 would “remove references to the ALARA principle” in favour of “determinate” thresholds [S39]. A further proposed rule of 24 September 2026 covers reactor licensing and oversight [S40]. - Security and war. - The chapter criticised the exclusion of aircraft strikes and terrorism from the EU stress tests (p. 444). - Since 2022, the IAEA has said the occupation of Zaporizhzhia violates all “seven pillars” of nuclear safety. By May 2023 the plant had lost off-site power seven times and fallen back on diesel generators (secondary [S86]). - A drone strike breached the Chernobyl confinement in 2025 [S80, S82]. - These are hazards outside the scenario lists of conventional PRA.

Verdict: partly held up. - Accepted: the core critique, that PRA scenario lists and independence assumptions miss common-cause cascades. The IAEA, EU and Japanese reforms institutionalised planning beyond the design basis. - Not borne out: the call to move away from PRA. PRA was broadened, not abandoned. - Pessimistic: the specific recurrence figure is at the high end of later estimates. - Reversal: since 2025 the US has been moving in the opposite direction, towards narrowing what counts as a credible risk.

Implications for weight. - The mechanism (insight 1: scenario-bounded probability misses cascades) is well supported by official post-accident findings. - A lesson the chapter did not draw is also supported: reforms made after a disaster can be diluted in later rulemaking, or reversed when political priorities shift from safety to speed. - The numbers in the chapter’s recurrence arithmetic should not be reused.

Claim 5: European liability regimes (operator cap of €169m; €1.5bn total under the revised Paris/Brussels protocols) are “very unlikely” to prove adequate; full insurance could add up to €2.36/kWh; required liability €6.09 trillion; caps cause under-deterrence and distort competition (pp. 446, 448)#

Original claim. - The chapter contrasts European caps with Fukushima liabilities of €76–152bn and clean-up of €190bn (p. 446). - It cites the Versicherungsforen Leipzig actuarial study. The digest notes the study was commissioned by a renewable-energy industry body, which the chapter does not disclose. - It cites Faure and Fiore (2009) on caps as a subsidy that weakens deterrence.

Subsequent developments - The revised conventions are now in force. - The 2004 Protocols to the Paris and Brussels Conventions entered into force on 1 January 2022 for most western European parties (secondary [S45]). - The UK order implementing them confirms the “Paris Convention limit of €700 million” for operators [S44]. - So the “proposals” the chapter described are now law. At €1.5bn in total, they remain about two orders of magnitude below realised costs. - Fukushima’s realised costs. - The government’s December 2016 estimate was ¥21.5 trillion. By 2022, according to Board of Audit sources, ¥12.1 trillion had been spent, including ¥7.1 trillion in compensation [S46, a news report]. - The Japan Center for Economic Research (JCER, 2019) estimated ¥35–80 trillion over 40 years, and about ¥35 trillion by 2050 in a scenario that leaves fuel debris in place [S47]. - Even the official figure is on the order of €150bn, about 100 times the European ceiling. - Who paid. The IAEA reports that TEPCO’s liability “was unlimited in amount” and that TEPCO was not granted the natural-disaster exemption. Payment nonetheless required “financial support to TEPCO by the Nuclear Damage Compensation and Decommissioning Facilitation Corporation (NDF), and NDF becoming the controlling shareholder of TEPCO” [S33]. Even an uncapped regime ended in public rescue. - Insurance-cost estimates. Laureto and Pearce (2016) spread reported Fukushima costs of $20–525bn over the plant’s lifetime output, giving $0.22–5.78/kWh. That is “far higher than the current insurance costs by Japanese law of $0.01/kWh” [S48]. This is a plant-specific calculation after the event, not a premium for a whole fleet. Like the Leipzig study, it is a high-end framing. It confirms the order of magnitude of the €2.36/kWh figure, not its precision. - Heavy tail. Wheatley et al. found that, even if no event larger than Chernobyl or Fukushima is assumed, the expected annual accident cost worldwide “bracket[s] the cost of a new plant” [S31]. - Distortion of competition. European courts did not treat support for nuclear power as unlawful. The Commission approved Hinkley Point C’s state aid in 2014, and the Court of Justice rejected Austria’s challenge in September 2020 (secondary, via [S59] and World Nuclear News). I found no EU instrument that raises liability beyond the 2004 Protocols. I could not check EUR-Lex directly, so this is a statement about what I found. - Elsewhere. The US Price-Anderson regime, which pools operator contributions up to a limit, was renewed in 2024 through 2045 (secondary [S50]).

Verdict: strengthened on inadequacy. The magnitude comparison is now based on realised, officially estimated costs. The under-deterrence and competition-distortion arguments remain plausible but untested. Courts and legislators have not acted on them, and the Japanese case shows that uncapped liability also ends in socialised costs when the operator cannot pay.

Implications for weight. - The lesson that caps and state backstops move tail risk onto the public (insight 7) deserves high weight. It now rests on realised, official figures rather than one commissioned study. - The precise €2.36/kWh and €6.09 trillion figures should not be reused without disclosing who commissioned them. - A refinement is warranted. When one failure can exceed an operator’s value, the public ends up bearing the cost whether or not liability is capped. The deterrent effect of liability rules is then weak either way.

Claim 6: The EU nuclear share of electricity will decline relatively after 2020, and emphasis will shift to extension, upgrading and retrofitting of existing reactors (p. 445)#

Original claim. The chapter bases this on limited new build since 2000, an ageing fleet and phase-outs (p. 445, citing Lévêque 2011).

Subsequent developments - The share fell. Eurostat gross-generation data for the EU-27 give a nuclear share that I computed from the dataset [S64]:

Year Nuclear share of EU-27 generation
2013 27.6%
2019 26.3%
2020 24.5%
2022 21.6%
2023 22.4%
2024 23.0%

Mochovce 4 reached first criticality in August 2026 (secondary [S61]), and construction of Paks II-1 in Hungary began in February 2026 [S62].

Verdict: held up. Both parts of the forecast were borne out. The decline had begun before 2020 and partly reflected fleet performance problems as well as closures.

Implications for weight. This modest structural forecast was sound. It follows from a general pattern: when new build is slow and costly, incumbents extend existing assets. That pattern carries moderate weight as a lens.

Claim 7: New build will require implicit or explicit public subsidy. Olkiluoto 3 was expected online late 2014 at €5.7bn+ (originally 2009, €3bn); Flamanville hoped operational by 2016 at €6bn (originally 2012, €3.3bn); Pyhäjoki construction expected in 2015; the UK approved “up to eight” new plants (pp. 443–444)#

Original claim. Box 18.9 (p. 444) presents the European Pressurised Reactor (EPR) projects as first-of-a-kind overruns. The surrounding text lists the new-build plans of 2011–12 (p. 443).

Subsequent developments - Olkiluoto 3. - Regular production began on 16 April 2023. That was 14 years later than the original date and 8.5 years later than the chapter’s forecast [S51]. - TVO and Areva settled their dispute in 2018 [S52]. TVO’s investment is reported at about €5.5bn, and total cost including the supplier’s losses at about €11bn (secondary [S52]). - Flamanville 3. - It was connected to the grid on 21 December 2024 and reached 100% power on 14 December 2025 [S53, S54]. - EDF’s cost at completion was €12.7bn in January 2022 (in 2015 euros, per EDF’s convention), up from €12.4bn [S55]. That is almost four times the original €3.3bn. - In 2020 the French Cour des comptes put the cost at up to €19.1bn including financing costs (secondary, [S56]). - Pyhäjoki (Hanhikivi 1). The project was never built. In 2022 the Finnish government said it would not issue a construction licence, and Fennovoima ended its contract with Rosatom in May 2022 (secondary [S57]). - United Kingdom. - Hinkley Point C is being built under a revenue contract that guarantees the operator a fixed price. The Commission approved it as state aid, and the Court of Justice upheld that in 2020 (secondary). - EDF’s January 2024 update put its cost at £31–34bn in 2015 values, with Unit 1 in 2029–2031 [S58]. EDF’s February 2026 results reportedly moved Unit 1 to 2030 at £35bn (2015 prices) (secondary [S59]). - Sizewell C reached a final investment decision on 22 July 2025. The government took a 44.9% stake as “the single biggest equity shareholder”, with a French export-credit guarantee [S60]. - I found no other UK large-reactor project that has reached a final investment decision. The digest records other sites as later abandoned; I did not verify each. - State aid elsewhere. Hungary’s Paks II, the other EU new build, began construction in February 2026 [S62].

Verdict: strengthened. Every schedule and cost in Box 18.9 was exceeded, several times over. Pyhäjoki was cancelled. The UK’s “up to eight” has become two, both with explicit state support: a guaranteed price for one, and public equity plus debt guarantees for the other.

Implications for weight. The lesson that first-of-a-kind megaprojects overrun and push risk onto the public (insight 14) is among the best supported in the chapter. It should carry high weight as a technology-neutral pattern. The size of the overruns shown here, three to four times on cost and 9–14 years on schedule, gives a useful calibration.

Claim 8: Policy trajectories as of 2011–12: German phase-out by 2022; Swiss non-replacement by 2034; Italian ban by binding referendum; Belgian phase-out; Borssele open to 2033; the IAEA’s 2035 growth forecast halved; construction starts fell from 15 to 2 (pp. 443–444)#

Original claim. The chapter reports these as the post-Fukushima policy landscape. It suggests Germany’s decision “may prove significant for European energy policy as a whole” (p. 444).

Subsequent developments - Germany. The phase-out was completed with a short delay. The last three reactors, Emsland, Isar 2 and Neckarwestheim 2, shut on 15 April 2023, 3.5 months after the original date, following an emergency extension during the 2022 energy crisis (secondary [S65]). - Switzerland. - Voters approved the Energy Strategy 2050 on 21 May 2017, which banned new nuclear plants. - In its message of 13 August 2025, the Federal Council recommended rejecting a popular initiative that would allow all low-carbon generation, but proposed, as an indirect counter-proposal, removing the 2017 new-build ban from the Nuclear Energy Act. - Both chambers of parliament adopted it on 19 June 2026. A referendum has been launched against it; if 50,000 valid signatures are collected by 8 October 2026, a vote will follow in 2027 (secondary [S67]). - Italy. - On 23 September 2026 the Senate gave final approval to an enabling law on “sustainable nuclear energy”. The Chamber had approved it in June. - The government now has 12 months to issue decrees covering plants, safety, waste and siting. The minister’s target for first reactors is 2034–35, “not a deadline” [S68]. - Commentators note that the next legislature will decide whether anything is built [S69]. - Belgium. It extended Doel 4 and Tihange 3 by ten years in 2022–23, and parliament repealed the phase-out law in May 2025 (secondary [S66]). - Netherlands. The government designated the Borssele site for two new reactors (2022) and is studying operation of the existing plant beyond 2033. In 2025 the minister called a 2035 start date “unrealistic” (secondary [S70]). - Global forecasts. - In December 2023, more than 20 countries launched a declaration at COP28 to triple nuclear capacity by 2050 [S71]. By 2026, 38 countries had endorsed it, including Belgium, Italy, the Netherlands, Japan and Sweden [S72]. - I could not retrieve the IAEA’s current projection series or the global series of construction starts (see Method note). - The World Nuclear Association’s reactor database lists ten construction starts worldwide in January–September 2026 alone, six of them in China [S62]. - Japan (context the chapter discusses on p. 442). The NRA lists 11 units “in operation” on 16 September 2026, including TEPCO’s Kashiwazaki-Kariwa Unit 6, and 23 units in or under decommissioning [S36].

Verdict: weakened as a durable trajectory. The snapshot was accurate for 2011–12. Germany’s phase-out held. Most others have been reversed or are being reversed, 12–15 years on. The reversals were driven by energy-security politics after 2022 and by climate commitments.

Implications for weight. - The chapter’s insight 15 (focusing events shift policy quickly, and the response depends on political culture) holds. - It needs a second half: shifts made in response to a focusing event are only as durable as the political coalition behind them. A later crisis of a different kind can reverse them within a decade. - Only the policy that was physically carried out became irreversible: Germany’s reactors are closed. - The durability of any post-disaster policy should be treated as an open question, not inferred from the size of the initial reaction.

Claim 9: Fukushima was “profoundly manmade” and resulted from “regulatory capture” (NAIIC); regulators had industry ties that “create a conflict of interest”; TEPCO retracted data at least four times (pp. 440, 442–443)#

Original claim. - The chapter quotes the Diet commission (Box 18.7, p. 443) and a Nature editorial on conflicts of interest (p. 442). - It notes that TEPCO retracted radionuclide and reactor data at least four times (p. 440). - It gives the 1,000-year tsunami evidence from the 869 Jōgan tsunami as the ignored early warning (Box 18.5, p. 438).

Subsequent developments - Official findings that support it. - The IAEA Director General’s report (2015) found that the “safety” assumption was “accepted by nuclear power plant operators and was not challenged by regulators or by the Government”. It found that responsibilities “were divided among a number of bodies”, and that the plant’s vulnerability to external hazards “had not been reassessed” [S33]. - Japan’s new regulator described the pre-accident “safety myth” as having “critically impeded efforts for nuclear safety” [S35]. - Disclosure failures that came out later. - TEPCO admitted on 22 July 2013 that contaminated groundwater had been leaking to the sea (secondary [S78]). - A 2016 investigation found staff had been told not to use the term “core meltdown” for two months after the accident (secondary [S78], citing Nuclear Engineering International, 24 June 2016). - The criminal trial “revealed many facts that were previously unknown, concealed, or denied” [S75]. Among them was TEPCO’s 2008 internal study, which used the government’s 2002 long-term assessment to estimate a possible tsunami of 15.7 m (secondary [S78]). - Institutional reform. - The Nuclear Regulation Authority was established on 19 September 2012 and imposed retroactive requirements from July 2013 [S35]. - The NRA’s original framework limited reactors to 40 years, with a single 20-year extension [S35]. - Koppenborg’s studies find that the NRA “stood its ground against attempts to undermine its position of power”. With lawsuits by citizens, this meant the pro-nuclear network “lost policy implementation power”: only six reactors were operating in 2020 [S42, S43]. - Signs of backsliding. - Legislation enacted in 2023 allows operation beyond 60 years by excluding shutdown periods, “subject to the economy minister’s approval”. This moves an important part of lifetime decisions from the regulator’s statute towards the promoting ministry (secondary [S74]). - TEPCO restarted Kashiwazaki-Kariwa Unit 6 in 2026. It was briefly suspended after the restart and is listed in operation in September 2026 [S36, S79]. - Courts take a narrower view of responsibility. - On 17 June 2022 the Supreme Court held the state not liable to evacuees (secondary [S78]). - Three former TEPCO executives were acquitted of negligence in 2019. The acquittal was upheld in 2023 and became final on 11 March 2025, after the Supreme Court found “no foreseeability” of the accident (secondary [S76]). - A 2022 district court order that four executives pay TEPCO ¥13 trillion was overturned by the Tokyo High Court in June 2025, again on foreseeability (secondary [S77]). - The courts have therefore not accepted, as a finding of individual or state legal responsibility, the Diet commission’s view that the accident was preventable and “manmade”. - Comparable regulatory changes elsewhere. The EU’s 2014 directive requires regulators to be “functionally separate from any other body or organisation concerned with the promotion or utilisation of nuclear energy” [S34]. By contrast, the US executive order of 2025 directs the NRC to reorganise, with reductions in force, to speed up licensing, and to cut its Advisory Committee on Reactor Safeguards “to the minimum necessary” [S38].

Verdict: held up as an institutional diagnosis, endorsed by the IAEA and by Japan’s own reform. It is contested in law: Japanese courts have repeatedly found that the tsunami was not legally foreseeable.

Implications for weight. - Insight 5 (regulators tied to the industry weaken prevention) deserves high weight. - The hindsight adds two lessons: - independence won after a disaster erodes when promotional ministries regain authority over key decisions such as operating lifetimes; - legal accountability runs on a narrower standard of foreseeability than inquiries use. A disaster can be judged “manmade” and yet leave no one legally responsible. - Both matter for any lens on how governance is captured and then rebuilt.

Claim 10: Fukushima decommissioning “will take up to 40 years”; Chernobyl remediation is expected to continue to 2065, with “less than half” of the needed funding raised (pp. 433, 442)#

Original claim. - TEPCO’s revised timetable put decommissioning at up to 40 years (p. 442). - For Chernobyl, the chapter says remediation “is likely to continue until 2065”, with less than half the resources raised and the completion date slipped by a decade (p. 433).

Subsequent developments - Fukushima’s schedule. - The official roadmap still targets completion in “2041–2051” [S83, S84]. - Retrieval of melted fuel debris formally began in September 2024. Two trial retrievals, in November 2024 and April 2025, recovered “totally 0.9g of samples” [S83]. - Total debris is estimated at about 880 tonnes (secondary [S85]). - JCER’s 2019 study already modelled a “confinement-managing” alternative, which would postpone full debris removal [S47]. - With under 1 g recovered 14 years in, the 2051 end date looks doubtful. The official roadmap has not changed. - Fukushima’s treated water. Discharge began in August 2023 after an IAEA review and continues [S28, S83]. - Chernobyl’s funding and structure. - The funding gap the chapter described was eventually closed for the confinement. The Shelter Implementation Plan (the programme to make the reactor site safe) reached €2.1bn. The arch was slid into place in 2016 and handed over in 2019. - The EBRD says it has “mobilised more than €2.5 billion in international contributions” plus more than €700m of its own resources [S80]. - Then the 14 February 2025 drone strike breached “the inner and outer cladding, damaging around 200 m² of panels”. It disabled the main crane and ventilation needed to dismantle the old shelter [S80]. - In December 2025 the IAEA reported that the structure “had lost its primary safety functions, including the confinement capability” (secondary [S82]). - The EBRD estimates repairs at “at least €500 million”. It says the structure “must be fully functional again by 2030” to avoid irreversible corrosion. Donors approved €30m for the first phases in April 2026 [S81]. - Earlier, Russian forces occupied the exclusion zone in February–March 2022 (secondary [S82]). - I found no updated official end date for remediation of the site.

Verdict: strengthened. Both timelines remain very long, and costs have risen. At Chernobyl a hazard nobody anticipated, war, has reopened a problem that funding had closed.

Implications for weight. Insight 8 (costs of catastrophic failure run for decades and exceed provisions) carries high weight. The hindsight adds that decades-long legacies are exposed to hazards outside their original risk frame, and that a funding gap closed once can reopen.

Summary of verdicts#

# Claim (pages) Verdict
1 Fukushima health consequences over 5–40 years; integrated, multi-outcome biomonitoring (pp. 432, 442, 448) Partly held up (no radiation-attributable effects documented; UNSCEAR expects none discernible; thyroid excess attributed to screening, with minority dissent; 2,351 disaster-related deaths and continuing displacement; monitoring put in place, with screening harms now recognised)
2 Chernobyl mortality: 4,000 vs 93,000 (Belarus) vs authors’ 17,000–68,000; leukaemia latency >40 years (pp. 434–436) Contested (authors’ extrapolation consistent with UNSCEAR doses and later low-dose evidence, but mainstream bodies decline to project; Yablokov rejected; about 20,000 thyroid cancers with attributable fraction about 0.25; leukaemia latency supported; no germline effect)
3 Cs-137 to air 3.5 × 10^16 Bq, twice official, 40% of Chernobyl; largest ocean release (pp. 440–441) Weakened (consensus 6–20 PBq, about 20% of Chernobyl; official 15 PBq within range); ocean claim held; unit errors confirmed
4 Recurrence every 20 years vs 1:100,000; PRA “very limited”; urgent reappraisal (pp. 432, 447–449) Partly held up (critique accepted and beyond-design-basis rules added; PRA broadened, not abandoned; frequency at pessimistic end; US reversal since 2025)
5 Liability caps inadequate; €2.36/kWh; under-deterrence and distortion (pp. 446, 448) Strengthened on inadequacy (protocols in force at €1.5bn; Fukushima at least ¥21.5 trillion; uncapped TEPCO still needed a state rescue); deterrence and distortion arguments untested
6 EU nuclear share declines after 2020; shift to life extension (p. 445) Held up (27.6% in 2013 to 21.6% in 2022, 23.0% in 2024; widespread life extensions)
7 New build needs subsidy; OL3 2014, Flamanville 2016, Pyhäjoki 2015, UK up to eight (pp. 443–444) Strengthened (OL3 2023, Flamanville 2024–25 at about four times budget; Pyhäjoki cancelled; UK two projects, both state-backed)
8 2011–12 phase-outs and slowdown (pp. 443–444) Weakened as a trajectory (Germany completed 2023; Belgium, Italy, Switzerland and the Netherlands reversed or reversing; tripling pledge; construction starts recovered)
9 “Profoundly manmade”; regulatory capture; TEPCO data retractions (pp. 440, 442–443) Held up institutionally (IAEA and Japanese reform; further disclosure failures); contested in law (acquittals final 2025; state not liable 2022)
10 Fukushima decommissioning up to 40 years; Chernobyl to 2065, underfunded (pp. 433, 442) Strengthened (0.9 g of about 880 t of debris retrieved by 2025; Chernobyl confinement breached in 2025, repairs at least €500m)

Technology-neutral lessons this check supports (for later use as a lens)#

Each lesson is tied to the section’s pages and to the later evidence above.

  1. Safety cases built on a scenario list miss the cascades that cause catastrophes. After failure, the usual institutional response is to widen the assessment rather than abandon it (pp. 432, 447–449). The IAEA, EU and Japan added beyond-design-basis, multi-unit and external-hazard requirements while keeping probabilistic methods [S33–S35]. (Claim 4)
  2. Hard-won precautionary reforms can be diluted in later rulemaking, or reversed when political priorities shift from safety to speed. Examples are the US rule in 2019, the US executive order in 2025, the proposed rules in 2026, and Japan’s operating-life legislation in 2023 [S37–S40, S74]. (Claims 4, 9)
  3. Protective actions and surveillance carry their own harms, and these can exceed the harm of the hazard they target. At Fukushima, the deaths and harms that can be measured came from evacuation, and screening led to overdiagnosis. None has been documented from radiation (pp. 442, 448; [S1, S6, S7, S14, S15]). (Claim 1)
  4. When effects fall below what epidemiology can detect, “the number of victims” becomes a choice about method and framing. Institutions legitimately differ on whether to compute it at all. Advocacy and mainstream figures should be kept apart (pp. 434–435; [S3, S16, S17]). (Claim 2)
  5. Independent measurement matters, but independent estimates can err in either direction. What proved reliable was many estimates, made in the open, agreeing over time. The early outside figure the chapter used was later revised down towards the official one (pp. 440–441; [S1, S25]). (Claim 3)
  6. Liability caps and state backstops move tail risk onto the public. When a single failure can exceed the operator’s value, costs are socialised even without a cap. Caps have persisted after a demonstration that losses can exceed them by about 100 times (pp. 446, 448; [S33, S44–S47]). (Claim 5)
  7. First-of-a-kind megaprojects overrun by multiples. Here, cost rose three to four times and schedules slipped 9–14 years. Projects that follow depend on explicit state finance and risk-sharing (pp. 443–444; [S51–S60]). (Claim 7)
  8. Policy shifts made in response to a focusing event are only as durable as the coalition behind them. A different later crisis can reverse them within a decade; only changes already carried out physically become irreversible (pp. 443–445; [S65–S72]). (Claim 8)
  9. Regulatory independence is built after a disaster but erodes as authority drifts back to promoting ministries. Legal accountability runs on a narrower test of foreseeability than inquiries use (pp. 442–443; [S33, S35, S42, S74–S77]). (Claim 9)
  10. Legacies lasting decades are exposed to hazards outside their original risk frame, and funding gaps closed once can reopen. The chapter’s objection to excluding security threats from stress tests (p. 444) was borne out by wartime damage to a completed confinement (pp. 433, 442, 444; [S80–S82, S86]). (Claims 4, 10)

Sources#

All retrieved 25–26 September 2026 unless noted. “Abstract only” means I read the abstract via Europe PMC, Crossref or HAL, not the full text. “Secondary” means the point rests on Wikipedia or news reporting that I could not confirm from a primary source. For those, I give both the Wikipedia page and the primary or news reference it cites.

UNSCEAR and international radiation-health assessments - [S1] UNSCEAR 2020/2021 Report, Volume II, Scientific Annex B: Levels and effects of radiation exposure due to the accident at the Fukushima Daiichi Nuclear Power Station: implications of information published since the UNSCEAR 2013 Report. United Nations, New York, 2022 (© December 2021; launched 9 March 2021). https://www.unscear.org/unscear/uploads/documents/unscear-reports/UNSCEAR_2020_21_Report_Vol.II.pdf. Paragraphs used: 23–24 and table 1, 35–36, 244–250; chapter IX conclusions (a), (c), (q), (r); appendix B comparison tables. - [S2] UNSCEAR, “Fukushima Daiichi Nuclear Power Station Accident” overview page (2013 Report, White Papers 2015–2017, 2020/2021 Report). https://www.unscear.org/unscear/en/areas-of-work/fukushima.html - [S3] UNSCEAR 2012 Report, Scientific Annex A: Attributing health effects to ionizing radiation exposure and inferring risks. United Nations, New York, 2015, paras 30–31. https://www.unscear.org/unscear/uploads/documents/publications/UNSCEAR_2012_Annex-A.pdf - [S4] UNSCEAR, “Chernobyl accident” overview page. https://www.unscear.org/unscear/en/areas-of-work/chernobyl.html - [S5] UNSCEAR Secretariat, White Paper: Evaluation of data on thyroid cancer in regions affected by the Chernobyl accident (published 2018). https://www.unscear.org/unscear/uploads/documents/publications/Chernobyl_WP_2017.pdf

Fukushima Prefecture official data - [S6] Fukushima Prefecture, “An Overview of Fukushima Health Management Survey” (page updated 23 July 2026; thyroid round results as of 31 March 2018 to 30 September 2025; Evaluation Subcommittee summary, July 2025). https://www.pref.fukushima.lg.jp/site/portal-english/en-4-3-1.html - [S7] Fukushima Prefecture, “Damage caused by the earthquake and tsunami” (casualties as of 1 May 2026). https://www.pref.fukushima.lg.jp/site/portal-english/en-1-1-1.html - [S8] Fukushima Prefecture, “Transition of the number of evacuees” (as of 1 May 2026). https://www.pref.fukushima.lg.jp/site/portal-english/en-1-4-1.html

Peer-reviewed studies: Fukushima health - [S9] Takahashi H et al. (2024). Detection of thyroid cancer among children and adolescents in Fukushima, Japan: a population-based cohort study of the Fukushima Health Management Survey. EClinicalMedicine. https://doi.org/10.1016/j.eclinm.2024.102722 (abstract). - [S10] Shimura H et al. (2023). Confounding factors and biases involved in regional differences in the detection rate of thyroid cancer in the second-round Thyroid Ultrasound Examination. Journal of Radiation Research. https://doi.org/10.1093/jrr/rrad044 (abstract). - [S11] Tsuda T, Tokinobu A, Yamamoto E, Suzuki E (2016). Thyroid cancer detection by ultrasound among residents ages 18 years and younger in Fukushima, Japan: 2011 to 2014. Epidemiology 27(3). https://doi.org/10.1097/EDE.0000000000000385 (abstract). - [S12] Kato T, Yamada K, Hongyo T (2023). Area dose-response and radiation origin of childhood thyroid cancer in Fukushima based on thyroid dose in UNSCEAR 2020/2021. Cancers 15(18):4583. https://doi.org/10.3390/cancers15184583 (abstract). - [S13] IARC Expert Group on Thyroid Health Monitoring after Nuclear Accidents (2018). Thyroid Health Monitoring after Nuclear Accidents (IARC Technical Publication). https://publications.iarc.who.int/571 (publication summary only). Companion paper: Togawa K et al. (2018), Lancet Oncology, https://doi.org/10.1016/S1470-2045(18)30680-6 (bibliographic record only). - [S14] Uchi Y et al. (2024). Features of causes of indirect certified disaster-related death in areas affected by the Fukushima Daiichi nuclear power plant accident. BMJ Open. https://doi.org/10.1136/bmjopen-2024-084009 (abstract). - [S15] Zhao T, Tsubokura M (2024). Risks associated with evacuation and disaster-related death after a radiation disaster: summary of research results from Hamadori region, Fukushima. Radiation Protection Dosimetry. https://doi.org/10.1093/rpd/ncae078 (abstract).

Chernobyl and low-dose epidemiology - [S16] Cardis E et al. (2006). Estimates of the cancer burden in Europe from radioactive fallout from the Chernobyl accident. International Journal of Cancer 119(6). https://doi.org/10.1002/ijc.22037 (abstract). - [S17] Balonov MI (2012). On protecting the inexperienced reader from Chernobyl myths. Journal of Radiological Protection 32(2):181. https://doi.org/10.1088/0952-4746/32/2/181 (abstract). - [S18] Richardson DB et al. (2023). Cancer mortality after low dose exposure to ionising radiation in workers in France, the United Kingdom, and the United States (INWORKS): cohort study. BMJ 382:e074520. https://doi.org/10.1136/bmj-2022-074520 (abstract). - [S19] Richardson DB et al. (2025). Site-specific cancer mortality after low-level exposure to ionizing radiation: INWORKS update. American Journal of Epidemiology. https://doi.org/10.1093/aje/kwae256 (abstract; background only). - [S20] Hauptmann M et al. (2020). Epidemiological studies of low-dose ionizing radiation and cancer: summary bias assessment and meta-analysis. JNCI Monographs 2020(56). https://doi.org/10.1093/jncimonographs/lgaa010 (abstract). - [S21] Hsu WL et al. (2013). The incidence of leukemia, lymphoma and multiple myeloma among atomic bomb survivors: 1950–2001. Radiation Research 179(3). https://doi.org/10.1667/RR2892.1 (abstract). - [S22] Liubarets TF et al. (2019). Childhood leukemia in Ukraine after the Chornobyl accident. Radiation and Environmental Biophysics 58. https://doi.org/10.1007/s00411-019-00810-4 (abstract). - [S23] Yeager M et al. (2021). Lack of transgenerational effects of ionizing radiation exposure from the Chernobyl accident. Science 372(6543). https://doi.org/10.1126/science.abg2365 (abstract). - [S24] Fucic A et al. (2016). Follow-up studies on genome damage in children after Chernobyl nuclear power plant accident. Archives of Toxicology 90. https://doi.org/10.1007/s00204-016-1766-z (abstract; co-authored by a chapter author).

Source term and ocean - [S25] Stohl A et al. (2012). Xenon-133 and caesium-137 releases into the atmosphere from the Fukushima Dai-ichi nuclear power plant: determination of the source term, atmospheric dispersion, and deposition. Atmospheric Chemistry and Physics 12:2313–2343. https://acp.copernicus.org/articles/12/2313/2012/ (abstract). - [S26] Buesseler KO et al. (2012). Fukushima-derived radionuclides in the ocean and biota off Japan. PNAS 109(16). https://doi.org/10.1073/pnas.1120794109 (abstract). - [S27] Buesseler K et al. (2017). Fukushima Daiichi-derived radionuclides in the ocean: transport, fate, and impacts. Annual Review of Marine Science 9. https://doi.org/10.1146/annurev-marine-010816-060733 (abstract). - [S28] IAEA (4 July 2023). IAEA Comprehensive Report on the Safety Review of the ALPS-Treated Water at the Fukushima Daiichi Nuclear Power Station. https://www.iaea.org/sites/default/files/iaea_comprehensive_alps_report.pdf

Accident frequency, PRA and the IAEA’s Fukushima findings - [S29] Ha-Duong M, Journé V (2014). Calculating nuclear accident probabilities from empirical frequencies. Environment Systems and Decisions 34. https://doi.org/10.1007/s10669-014-9499-0 (abstract via HAL: https://hal.science/hal-01018478v2). - [S30] Escobar Rangel L, Lévêque F (2014). How Fukushima Dai-ichi core meltdown changed the probability of nuclear accidents? Safety Science 64. https://doi.org/10.1016/j.ssci.2013.11.017 (abstract via HAL: https://hal.science/hal-01110974v1). - [S31] Wheatley S, Sovacool B, Sornette D (2016, online 22 March 2016; Risk Analysis 37(1), 2017). Of disasters and dragon kings: a statistical analysis of nuclear power incidents and accidents. https://doi.org/10.1111/risa.12587 (abstract). See also Wheatley, Sovacool, Sornette (2016), Reassessing the safety of nuclear power, Energy Research & Social Science 15, https://doi.org/10.1016/j.erss.2015.12.026 (bibliographic record only). - [S33] IAEA (2015). The Fukushima Daiichi Accident: Report by the Director General (STI/PUB/1710), Vienna. https://www-pub.iaea.org/MTCD/Publications/PDF/Pub1710-ReportByTheDG-Web.pdf. Parts used: Foreword; Summary on beyond-design-basis accidents; section 2 on probabilistic safety assessment; section 5.4.1 on compensation.

Regulation (EU, Japan, US) - [S34] Council Directive 2014/87/Euratom of 8 July 2014 amending Directive 2009/71/Euratom (nuclear safety). ELI: http://data.europa.eu/eli/dir/2014/87/oj. Text read at https://www.legislation.gov.uk/eudr/2014/87/contents/adopted (Articles 5, 8a, 8b, 8e; recitals 5, 12). - [S35] Nuclear Regulation Authority, Japan. “Enforcement of the New Regulatory Requirements for Commercial Nuclear Power Reactors” (8 July 2013; English document August 2013). https://www.nra.go.jp/data/000067212.pdf (index: https://www.nra.go.jp/english/regulatory/index.html). - [S36] Nuclear Regulation Authority, “Present States of Operation” (last update 16 September 2026). https://www.nra.go.jp/english/nuclearfacilities/operation.html - [S37] US NRC, “Mitigation of Beyond-Design-Basis Events”, final rule, 84 FR 39684, 9 August 2019 (including Commissioner Baran’s dissenting views). https://www.federalregister.gov/documents/2019/08/09/2019-16600/mitigation-of-beyond-design-basis-events (full text: https://www.federalregister.gov/documents/full_text/text/2019/08/09/2019-16600.txt). - [S38] Executive Order 14300, “Ordering the Reform of the Nuclear Regulatory Commission”, signed 23 May 2025, published 29 May 2025. https://www.federalregister.gov/documents/2025/05/29/2025-09798/ordering-the-reform-of-the-nuclear-regulatory-commission - [S39] US NRC, “Reforming and Modernizing the NRC’s Radiation Protection Framework”, proposed rule, 91 FR 43456, 15 July 2026 (comments closed 31 August 2026). https://www.federalregister.gov/documents/2026/07/15/2026-14208/reforming-and-modernizing-the-nrcs-radiation-protection-framework - [S40] US NRC, “Regulatory Enhancements for Reactor Licensing, Decommissioning, and Operational Oversight”, proposed rule, 24 September 2026 (abstract read). https://www.federalregister.gov/documents/2026/09/24/2026-19568/regulatory-enhancements-for-reactor-licensing-decommissioning-and-operational-oversight - [S41] Council Directive 2013/59/Euratom of 5 December 2013 (Basic Safety Standards), recital 14 and Article 9(3)(a). ELI: http://data.europa.eu/eli/dir/2013/59/oj. Text read at https://www.legislation.gov.uk/eudr/2013/59/contents/adopted - [S42] Koppenborg F (2020, online). Nuclear restart politics: how the “nuclear village” lost policy implementation power. Social Science Japan Journal. https://doi.org/10.1093/ssjj/jyaa046 (abstract). - [S43] Koppenborg F (2023). “The Fissured ‘Nuclear Village’”, in Japan’s Nuclear Disaster and the Politics of Safety Governance, Cornell University Press. https://doi.org/10.7591/cornell/9781501770043.003.0006 (chapter abstract). - [S87] Thomas S (2016). The Hinkley Point decision: an analysis of the policy process. Energy Policy 96. https://doi.org/10.1016/j.enpol.2016.06.021 (bibliographic record only; chapter author).

Liability and costs - [S44] The Nuclear Installations (Liability for Damage) Order 2016 (UK SI 2016/562), Explanatory Note. https://www.legislation.gov.uk/uksi/2016/562/note/made - [S45] Secondary: Wikipedia, “Paris Convention on Third Party Liability in the Field of Nuclear Energy” (ratification table showing entry into force of the 2004 Protocols on 1 January 2022; cites the NEA status page http://www.oecd-nea.org/law/paris-convention-ratification.html, which refused automated access). https://en.wikipedia.org/wiki/Paris_Convention_on_Third_Party_Liability_in_the_Field_of_Nuclear_Energy - [S46] Asahi Shimbun, “12.1 trillion yen spent so far on Fukushima nuclear disaster”, 7 November 2022 (news; cites Board of Audit sources and the government’s 2016 estimate of ¥21.5 trillion). https://www.asahi.com/ajw/articles/14762193 - [S47] Japan Center for Economic Research, “Accident Cleanup Costs Rising to 35–80 Trillion Yen in 40 Years”, 3 July 2019. https://www.jcer.or.jp/english/accident-cleanup-costs-rising-to-35-80-trillion-yen-in-40-years - [S48] Laureto JJ, Pearce JM (2016). Nuclear insurance subsidies cost from post-Fukushima accounting based on media sources. Sustainability 8(12):1301. https://doi.org/10.3390/su8121301 (abstract). - [S50] Secondary: Wikipedia, “Price–Anderson Nuclear Industries Indemnity Act” (renewal through 2045 in the ADVANCE Act, 2024). https://en.wikipedia.org/wiki/Price%E2%80%93Anderson_Nuclear_Industries_Indemnity_Act

New build - [S51] TVO press release, “Regular electricity production has started at Olkiluoto 3 EPR”, 16 April 2023. https://www.tvo.fi/en/index/news/pressreleasesstockexchangereleases/2023/regularelectricityproductionhasstartedatolkiluoto3epr.html - [S52] TVO, settlement agreement release, 11 March 2018. https://www.tvo.fi/news/1966. Cost figures (TVO investment about €5.5bn; total about €11bn) are secondary, from Wikipedia, “Olkiluoto Nuclear Power Plant”: https://en.wikipedia.org/wiki/Olkiluoto_Nuclear_Power_Plant - [S53] EDF press release, “Update on the Flamanville EPR: the reactor produces its first electrons on the national electricity grid”, 21 December 2024. https://www.edf.fr/en/the-edf-group/dedicated-sections/journalists/all-press-releases/update-on-the-flamanville-epr-the-reactor-produces-its-first-electrons-on-the-national-electricity-grid - [S54] EDF press release, “Update on the Flamanville EPR: the reactor has reached 100% of nuclear thermal power”, 14 December 2025. https://www.edf.fr/en/the-edf-group/dedicated-sections/journalists/all-press-releases/update-on-the-flamanville-epr-the-reactor-has-reached-100-of-nuclear-thermal-power - [S55] EDF, “Point d’actualité sur l’EPR de Flamanville”, 12 January 2022 (cost at completion €12.4bn to €12.7bn). https://www.edf.fr/groupe-edf/espaces-dedies/journalistes/tous-les-communiques-de-presse/point-d-actualite-sur-l-epr-de-flamanville-0 - [S56] Secondary: Le Monde, “Nucléaire : la Cour des comptes éreinte l’EPR”, 9 July 2020 (Cour des comptes estimate up to €19.1bn; via Wikipedia “Flamanville Nuclear Power Plant”; not opened). https://www.lemonde.fr/economie/article/2020/07/09/nucleaire-la-cour-des-comptes-ereinte-l-epr_6045707_3234.html - [S57] Secondary: Wikipedia, “Hanhikivi Nuclear Power Plant” (cancellation, 2022; cites Yle, 2 May 2022, https://yle.fi/news/3-12425648). https://en.wikipedia.org/wiki/Hanhikivi_Nuclear_Power_Plant - [S58] EDF press release, “Hinkley Point C Update”, 23 January 2024. https://www.edf.fr/en/the-edf-group/dedicated-sections/journalists/all-press-releases/hinkley-point-c-update-1 - [S59] Secondary: Wikipedia, “Hinkley Point C nuclear power station” (EDF 2025 results: Unit 1 in 2030, £35bn in 2015 prices, citing New Civil Engineer, 20 February 2026; Court of Justice decision of September 2020 on Austria’s challenge, citing World Nuclear News, 22 September 2020). https://en.wikipedia.org/wiki/Hinkley_Point_C_nuclear_power_station - [S60] GOV.UK, “Sizewell C gets green light with final investment decision”, 22 July 2025. https://www.gov.uk/government/news/sizewell-c-gets-green-light-with-final-investment-decision - [S61] Secondary: Wikipedia, “Mochovce Nuclear Power Plant” (Unit 3 commercial operation 17 October 2023; Unit 4 first criticality, citing World Nuclear News, 10 August 2026). https://en.wikipedia.org/wiki/Mochovce_Nuclear_Power_Plant - [S62] World Nuclear Association, “Nuclear Power in the World Today” (reactor database table of 2026 construction starts). https://world-nuclear.org/information-library/current-and-future-generation/nuclear-power-in-the-world-today

Energy statistics - [S63] Eurostat, “Nuclear energy statistics”, Statistics Explained (data from January 2025). https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Nuclear_energy_statistics - [S64] Eurostat dataset nrg_bal_peh (gross electricity production, EU27_2020, siec N900H and TOTAL, GWh), queried via API: https://ec.europa.eu/eurostat/api/dissemination/statistics/1.0/data/nrg_bal_peh?format=JSON&lang=en&geo=EU27_2020&nrg_bal=GEP&siec=N900H&siec=TOTAL&unit=GWH (shares computed by me).

Policy trajectories - [S65] Secondary: Wikipedia, “Nuclear power in Germany” (last three reactors shut 15 April 2023; cites Politico Europe, 18 April 2023). https://en.wikipedia.org/wiki/Nuclear_power_in_Germany - [S66] Secondary: Wikipedia, “Nuclear power in Belgium” (ten-year extension of Doel 4 and Tihange 3; phase-out law repealed, citing DW, 15 May 2025, https://www.dw.com/en/belgian-parliament-scraps-nuclear-phaseout-plan/a-72560001). https://en.wikipedia.org/wiki/Nuclear_power_in_Belgium - [S67] Secondary: German Wikipedia, “Kernenergie in der Schweiz” (2017 vote; Federal Council message 13 August 2025; parliament 19 June 2026; referendum signature deadline 8 October 2026; Leibstadt long-term operation). https://de.wikipedia.org/wiki/Kernenergie_in_der_Schweiz - [S68] RAI News, “Nucleare, il Senato dà il via libera definitivo alla normativa. Meloni: ‘Promessa mantenuta’”, 23 September 2026 (news). https://www.rainews.it/articoli/2026/09/nucleare-il-senato-da-il-via-libera-definitivo-alla-normativa-sulle-centrali-atomiche-meloni-promessa-mantenuta-725e96c3-e021-4247-a6b5-c611fa3849e6.html - [S69] Il Post, “Che intenzioni ha il governo sul nucleare”, 5 June 2026 (news). https://www.ilpost.it/2026/06/05/governo-parlamento-legge-delega-nucleare/ - [S70] Secondary: Wikipedia, “Nuclear power in the Netherlands” (Borssele site for new reactors; study of operation beyond 2033; 2025 delay). https://en.wikipedia.org/wiki/Nuclear_power_in_the_Netherlands - [S71] US Department of Energy, “At COP28, Countries Launch Declaration to Triple Nuclear Energy Capacity by 2050…”, 1 December 2023. https://www.energy.gov/articles/cop28-countries-launch-declaration-triple-nuclear-energy-capacity-2050-recognizing-key - [S72] World Nuclear Association, “Declaration to Triple Nuclear Energy” (38 endorsing nations). https://world-nuclear.org/net-zero-nuclear/tripling-pledges/declaration-to-triple-nuclear-energy - [S73] Secondary: Wikipedia, “Nuclear power in France” (ASN approval February 2021 for 900 MWe units beyond 40 years; 1300 MWe units, citing World Nuclear News, 4 July 2025). https://en.wikipedia.org/wiki/Nuclear_power_in_France - [S74] Secondary: Wikipedia, “Nuclear power in Japan” (2023 law excluding shutdown periods from the 60-year limit, subject to the economy minister’s approval). https://en.wikipedia.org/wiki/Nuclear_power_in_Japan

Accountability, courts and disclosure - [S75] Johnson DT, Fukurai H, Hirayama M (2020). Reflections on the TEPCO trial: prosecution and acquittal after Japan’s nuclear meltdown. The Asia-Pacific Journal: Japan Focus 18(2). https://doi.org/10.1017/S1557466020029307 (abstract). - [S76] Secondary: Japanese Wikipedia, “武黒一郎” (Takekuro Ichirō): acquittals in 2019 and 2023; Supreme Court Second Petty Bench rejected the appeal in March 2025 (“no foreseeability”); acquittal final on 11 March 2025. https://ja.wikipedia.org/wiki/%E6%AD%A6%E9%BB%92%E4%B8%80%E9%83%8E - [S77] Secondary: Japanese Wikipedia, “無過失責任” (strict liability): Tokyo High Court, June 2025, overturned the ¥13 trillion shareholder judgment on foreseeability. https://ja.wikipedia.org/wiki/%E7%84%A1%E9%81%8E%E5%A4%B1%E8%B2%AC%E4%BB%BB - [S78] Secondary: Wikipedia, “Fukushima nuclear accident” (Supreme Court ruling 17 June 2022; 2016 finding on the “core meltdown” wording, citing Nuclear Engineering International, 24 June 2016; the 2002 assessment and 2008 TEPCO study of a 15.7 m tsunami, citing Mainichi, 20 October 2018) and “Discharge of radioactive water of the Fukushima Daiichi Nuclear Power Plant” (TEPCO admission of groundwater leaks, 22 July 2013). https://en.wikipedia.org/wiki/Fukushima_nuclear_accident ; https://en.wikipedia.org/wiki/Discharge_of_radioactive_water_of_the_Fukushima_Daiichi_Nuclear_Power_Plant - [S79] Secondary: Wikipedia, “Kashiwazaki-Kariwa Nuclear Power Plant” (Unit 6 restart and suspension, citing BBC, 22 January 2026). https://en.wikipedia.org/wiki/Kashiwazaki-Kariwa_Nuclear_Power_Plant

Chernobyl and Fukushima legacy management - [S80] EBRD, “Making Chornobyl safe” (Shelter Implementation Plan €2.1bn; arch slid into place 2016, handed over 2019; more than €2.5bn mobilised; damage from the 2025 strike; “fully functional again by 2030”). https://www.ebrd.com/home/what-we-do/focus-areas/nuclear-safety/making-chornobyl-safe.html - [S81] EBRD, “EBRD donors back plan to repair Chornobyl’s protective shield”, 1 April 2026. https://www.ebrd.com/home/news-and-events/news/2026/ebrd-donors-back-plan-to-repair-chornobyl-s-protective-shield.html - [S82] Secondary: Wikipedia, “Chernobyl Nuclear Power Plant drone strike” (IAEA statement of December 2025 on loss of confinement function, citing RTÉ, 6 December 2025, https://www.rte.ie/news/2025/1206/1547589-chornobyl/, and Politico Europe, 6 December 2025) and “Chernobyl New Safe Confinement”. https://en.wikipedia.org/wiki/Chernobyl_Nuclear_Power_Plant_drone_strike - [S83] METI, “Overview of progress on Decommissioning and Fukushima reconstruction” (16 September 2025), https://www.meti.go.jp/english/earthquake/nuclear/decommissioning/pdf/1_20250916.pdf ; TEPCO, “Current Status of the Decommissioning at FDNPS” (16 September 2025), https://www.meti.go.jp/english/earthquake/nuclear/decommissioning/pdf/2.r20250916.pdf - [S84] METI, “Mid-and-Long-Term Roadmap towards the Decommissioning of TEPCO’s Fukushima Daiichi Nuclear Power Station Units 1–4” (5th revision, 27 December 2019, and index). https://www.meti.go.jp/english/earthquake/nuclear/decommissioning/index.html - [S85] Secondary: Wikipedia, “Fukushima Daiichi Nuclear Power Plant” (estimate of about 880 t of fuel debris, citing ABC News, 5 April 2023). https://en.wikipedia.org/wiki/Fukushima_Daiichi_Nuclear_Power_Plant - [S86] Secondary: Wikipedia, “Zaporizhzhia Nuclear Power Plant crisis” (IAEA September 2022 report on the “seven pillars”; seven losses of off-site power by May 2023, citing World Nuclear News, 31 May 2023). https://en.wikipedia.org/wiki/Zaporizhzhia_Nuclear_Power_Plant_crisis