Late Lessons, Jensen Huang and AI
Markdown version

Hindsight check: LL2-23 (Part D front matter and Ch 23 Understanding and accounting for the costs of inaction)#

Source section: EEA, Late lessons from early warnings: science, precaution, innovation (EEA Report No 1/2013). Part D title page, contents and list of panels (pp. 561–563; no introductory text), and Ch 23 by Mikael Skou Andersen and David Owain Clubb (pp. 564–580). Check window: publication (2013) to late September 2026. Checked: 26 September 2026.

Method note. - No general web search. The session’s web-search budget was exhausted, so I went directly to primary repositories: - EU law and Commission documents through the Publications Office (Cellar) and its SPARQL endpoint; - the US Federal Register API; - Europe PMC, Crossref and OpenAlex for journal articles (where a point rests on an abstract only, I say so); - the UN Treaty Collection; WHO; UNEP; NOAA CSL (host of the WMO/UNEP ozone assessments); the EEA; GOV.UK; the World Bank repository; IPBES; the Danish Ministry of Environment; AMAP. - Blocked sources. The Lancet full text (HTTP 403), the Minamata Convention secretariat site (403) and IISD’s Earth Negotiations Bulletin (403). As a result: - the Larsen and Sánchez-Triana lead-cost study is reported from its abstract; - the Minamata COP-6 dental-amalgam decision is reported from two 2026 journal articles, not from the decision text. - Annex 3. This chapter is new in 2013. Its SO2 and ozone-depleting-substance (ODS) boxes draw on 2001 Chapters 10 and 7. Under this task’s file restrictions I did not open Annex 3 of the 2013 report, so I cannot say what it adds on those cases. - Interests to keep in view. - Several of the chapter’s headline figures come from the lead author’s own work: the EXIOPOL nitrate estimates, and the EUR 1.50 per gram lead value (Andersen, 2010). - Temkin et al. (2019) [S16] are staff of an environmental advocacy organisation (the Environmental Working Group). - Larsen and Sánchez-Triana [S21] worked for or with the World Bank’s pollution programme. - The Danish expert group [S12] was convened by a national ministry that had already twice commissioned reviews of the nitrate limit. - Unit care (nitrate). Studies report either nitrate (NO3) or nitrate-nitrogen (NO3-N). 1 mg/L NO3-N ≈ 4.43 mg/L NO3. The EU and WHO limit of 50 mg/L NO3 ≈ 11.3 mg/L NO3-N. The Danish expert group had to reissue its report in January 2026 because it had mixed the two [S12].

Overview#

1. The chapter’s most general claim has been strongly vindicated: costs of inaction are routinely undercounted because whole categories of harm are left out. - Lead. The chapter’s own example, adult cardiovascular harm, is the clearest case. - Global lead-attributable deaths implied by the Global Burden of Disease (GBD) study rose as its exposure models changed: about 0.9 million (GBD 2019, implied by [S21]), 1.48 million (GBD 2021) [S24], and 3.5 million (GBD 2023, using cumulative bone-lead exposure) [S23]. WHO now repeats the 3.5 million figure [S25]. - A 2023 World Bank-linked estimate put the global cost of lead at USD 6.0 trillion (6.9% of world GDP) in 2019. About 77% of that is cardiovascular mortality, the category the chapter said had been omitted [S21]. - ODS. Later assessments added a climate co-benefit of the Montreal Protocol that the chapter could not count: protecting the land carbon sink from UV damage, worth a further 0.5–1.0 °C of avoided warming by 2100 [S42, S46]. - Caveat. Much of the upward revision reflects new modelling choices, not new harm. The chapter predicted this (“better modelling” shifts the ratios), but the estimates remain model-dependent and far apart.

2. Nitrate, the chapter’s most exposed advocacy case, has moved in its direction, but no binding standard has changed. - Evidence. Large cohort studies since 2013 report risks below 50 mg/L, most strongly for colorectal cancer and preterm birth [S5, S9–S11]. - Official reviews. - A Danish government-convened international expert group (December 2025, revised January 2026) concluded that nitrate in drinking water “poses a health risk” below 50 mg/L. It proposed a parametric value of 6 mg/L “as a precautionary approach” [S12]. - The Commission’s first full evaluation of the Nitrates Directive (15 July 2026) cites “increasing scientific evidence that nitrate levels in drinking water should be lower than 50 mg/l” [S13]. - What has not moved. - The EU’s 2020 recast of the Drinking Water Directive kept 50 mg/L and did not restore a guide value [S1]. - WHO’s current background document still finds the cancer evidence unconvincing [S2]. - Two of three meta-analyses found no association between drinking-water nitrate and colorectal cancer [S6, S7]. - The chapter was both overtaken and too cautious. - It said no European cost-benefit analysis of nitrate regulation existed (p. 570). The 2026 evaluation now provides one: benefits of EUR 10–22bn a year, three to seven times implementation costs [S13]. - If the Danish reading is right, the chapter’s “15–25 mg/L” thresholds and its 25 mg/L guide value were too lax, not too strict. - Its UK figure (EUR 2.6bn a year) rested mainly on bladder cancer, which is now the weakest of the candidate endpoints [S8, S12]. No one has replicated the figure.

3. The air-pollution claims (scale, and “the conservative bound is enough”) held up and were built into EU law. - The Commission’s December 2013 Clean Air Programme put health-related external costs at EUR 330–940bn in 2010 [S69], above the chapter’s CAFE figures. - The 2022 impact assessment for the ambient air quality recast reported benefit-cost ratios of 6:1 to 28:1 across options, even at the low end [S68]. The recast directive adopted in 2024 tightened the annual PM2.5 limit from 25 to 10 µg/m³ by 2030 [S67]. - EU deaths attributable to PM2.5 fell 45% between 2005 and 2022, but about 239,000 remained in 2022 [S70].

4. The mercury and ozone boxes largely held up; the specific projections have been superseded. - Mercury. - The Minamata Convention was adopted in October 2013, entered into force in 2017 and now has 154 parties [S32]. - The chapter’s point about harm displaced to the Arctic, and locked in there, has strengthened. Greenlandic exposure remains among the highest in the world and in places is rising [S38–S40]. - The AMAP projection of a 20% rise in global emissions is contested: inventories showed rises to 2015 [S30], but atmospheric observations imply Northern Hemisphere emissions fell between 2005 and 2020 [S31]. - Ozone. Recovery is on track. The 2022 WMO/UNEP assessment gives about 2040 for the near-global average, 2045 for the Arctic and 2066 for the Antarctic [S42]. The 2026 assessment is due at the end of 2026 [S43].

5. The valuation claims (Stern, discounting, value of a statistical life) were the most politically contingent, and hindsight shows how contingent. - Discount rates. - The US moved toward the chapter’s position in 2023, when OMB set a 2.0% default discount rate [S58]. - It reversed in 2025: Executive Order 14192 revoked that Circular and reinstated the 2003 version with its 3% and 7% rates [S59]. Executive Order 14154 withdrew the federal social cost of greenhouse gas estimates [S60]. - The UK (3.5% declining; 1.5% for health) and the EU (3%, with lower rates acceptable for health and environment) have held steady [S62, S63]. - Monetised mortality benefits. By February 2026 the US EPA was “refraining” from primary estimates of PM2.5 and ozone health benefits in a major rulemaking [S41]. - Credibility cuts both ways. A prominent high-damage climate estimate was retracted by Nature in December 2025 [S54].

6. What this means for weight. - Generalise freely. The mechanisms the chapter identifies, stated in neutral terms: - damage estimates count only what has already been quantified, so they are lower bounds; - valuation conventions (discount rate, VSL versus VOLY) move results by orders of magnitude and carry ethical choices; - reporting ranges and testing the conservative bound is a robust decision practice; - dispersal exports harm; - persistent stocks lock it in. - Give little weight to. The chapter’s own numerical extrapolations: - the UK nitrate EUR 2.6bn; - EU-27 ODS benefits of EUR 10–11bn; - 215 GtCO2-eq avoided; - lead costs of 4–6% of GDP. - Qualify heavily. The prediction that credible estimates will “justify precaution and strengthen diffuse interests” worked where institutions accepted monetised health benefits (the EU; Denmark on nitrate). It was reversed where the valuation conventions themselves became the object of political contest (US 2025–26). The durable lesson is not that numbers win. It is that who controls the accounting conventions decides what counts.

Claim-by-claim assessment#

Claim 1. Nitrate causes health effects well below 50 mg/L, with thresholds around 15–25 mg/L; the abandoned 25 mg/L guide value “deserves reconsideration” (pp. 570–571)#

What the chapter said. - The 1980 Directive had a 50 mg/L maximum and a 25 mg/L guide value. The guide value was dropped in 1998 for “absence of scientific proof”. - Only two cohort studies existed and their results were “ambiguous”. The Iowa study (Weyer et al., 2001) nonetheless “suggests that health effects can be detected well below the MAC-value and with lower thresholds of 15–25 mg NO3 per litre”. - The guide value “deserves reconsideration” (pp. 570–571). - The chapter does not mention that IARC classified ingested nitrate or nitrite, under conditions of endogenous nitrosation, as “probably carcinogenic” (Group 2A) in 2010. WHO’s 2016 background document confirms that classification [S2].

Subsequent developments. - The EU standard did not change. - Directive (EU) 2020/2184 (16 December 2020) keeps nitrate at 50 mg/L, with the combined condition nitrate/50 + nitrite/3 ≤ 1 [S1]. There is no guide value. - The recast was informed by a 2017 WHO Regional Office for Europe review of parameters [S1, recitals]. - In the same recast the lead value was halved to 5 µg/L by 2036, although WHO had recommended keeping 10 µg/L with concentrations “as low as reasonably practicable” [S1, recitals]. So the EU was willing to go beyond WHO advice for lead, but not for nitrate. - WHO kept 50 mg/L. - Its 2016 background document (WHO/FWC/WSH/16.52) bases the value on methaemoglobinaemia and thyroid effects in bottle-fed infants. - It states that “the weight of evidence does not clearly support an association between cancer and exposure to nitrate or nitrite per se” [S2]. - It estimates that endogenous NDMA formation at the guideline value would cause fewer than two extra cancers per 100,000 people over 70 years. - Epidemiology since 2013. - Iowa update. Jones et al. (2016) followed 34,708 Iowa women. Four or more years above 5 mg/L NO3-N (about 22 mg/L NO3) was associated with bladder cancer (HR 1.62; 95% CI 1.06–2.47) [S4]. This lands in the chapter’s 15–25 mg/L band. - The chapter misstates the original Weyer cohort: it had 21,977 women (16,541 on municipal supplies), not “10 000” [S20]. - Danish national cohort. Schullehner et al. (2018) followed 2.7 million people and found increased colorectal cancer risk “at drinking water levels above 3.87 mg/L”, well below both 50 and 25 mg/L [S5]. - Review. Ward et al. (2018) found the strongest evidence for colorectal cancer, thyroid disease and neural tube defects, with “many studies [observing] increased risk … below regulatory limits” [S3]. - Newer outcomes. - A 2023 meta-analysis found associations with preterm birth and neural tube defects [S9]. - A French prospective cohort (2025) reported breast cancer HR 1.51 for the top tertile of ingested nitrate [S10]. - A New Zealand national birth cohort (2026) linked low-level nitrate to preterm birth [S11]. - Against. - Picetti et al. (2022) found an association only with gastric cancer (OR 1.91 per 10 mg/L), and “no association … with colorectal cancer (10 studies)” [S6]. - Hosseini et al. (2021) found drinking-water nitrate “not associated with colorectal cancer risk” [S7]. - Arafa et al. (2022) found no pooled bladder-cancer association (OR 0.98), although the association rose to 1.36 after excluding high-risk-of-bias studies [S8]. - Official re-evaluation in Denmark. - DTU (Denmark’s National Food Institute) reviewed the evidence in 2019 and found it too uncertain. Its 2024 update concluded that “nitrate in drinking water constitutes a health risk … also … below the existing parametric value of 50 mg/L”. - The ministry then convened an international expert group (report 9 December 2025, revised 7 January 2026). It concluded that revision “is justified as a precautionary approach”. Using benchmark-dose modelling on the Schullehner cohort, it proposed a value of 6 mg/L [S12]. - It judged reproductive-outcome evidence “suggestive” but “not sufficiently consistent”. Bladder cancer and childhood CNS cancers were only “suggestive”. - I found no evidence that Denmark had adopted 6 mg/L by late September 2026. - EU evaluation (15 July 2026). - The Nitrates Directive evaluation cites the Danish recommendation. It states that the Directive’s objectives are “even more [relevant] when considering increasing scientific evidence that nitrate levels in drinking water should be lower than 50 mg/l to protect against carcinomic effects” [S13]. - Commission evaluations make no commitments to future action. - Exposure has not gone away. In 2016–2019, 14.1% of EU groundwater monitoring stations still exceeded 50 mg/L on annual average, and improvement had “slow[ed] since 2012” [S14].

Verdict: strengthened (directionally), but still contested at the level of WHO and pooled meta-analyses, and not yet reflected in any binding standard. - The concern about effects below 50 mg/L has gained a national expert endorsement and a Commission acknowledgement. - The chapter’s specific numbers look too lax rather than too strict: 15–25 mg/L thresholds, a 25 mg/L guide value, and bladder cancer as the anchor endpoint.

Implication for weight. Treat this as a well-documented instance of the chapter’s mechanism “proof lags evidence, and standards can be dropped just before the evidence arrives” (pp. 566, 570). The evidence grew for more than a decade while the binding standard stayed put. It is also a warning that the early warning’s specifics (the endpoint, the threshold) can be wrong even when its direction is right.

Claim 2. The costs of inaction on drinking-water nitrate are EUR 2.6bn a year for the UK alone; EUR 0.3/kg N mean, EUR 1.3/kg N in the UK and Belgium; colon cancer EUR 0.7/kg fertiliser N (p. 570)#

What the chapter said. - The UK figure is derived from Andersen et al. (2011, the EXIOPOL project), “based mainly on figures for bladder cancers derived from Weyer et al. (2001)”. - The colon-cancer figure is from van Grinsven et al. (2010). - The chapter adds that “no analysis has so far been carried out in Europe” comparing the benefits and costs of nitrate regulation (p. 570).

Subsequent developments. - No replication of the UK figure. I found no independent estimate of UK nitrate health costs to check EUR 2.6bn against. - The same method has been applied elsewhere. Estimates vary widely with endpoint choice and valuation. - Denmark: lowering the limit to 9.25 mg/L would avoid about 72 colorectal cancers a year, worth USD 179m a year, against mitigation costs of about USD 9m a year [S15]. - United States: 2,300–12,594 nitrate-attributable cancers a year, with USD 250m–1.5bn in medical costs plus USD 1.3–6.5bn in lost productivity [S16] (advocacy-organisation authors). - Minnesota: USD 745m a year. The authors add that “valuation methods that look only at avoided fatalities are likely to miss a substantial portion of the disease burden” [S17]. - New Zealand: USD 21–43m a year for colorectal cancer [S18]. - These fall broadly within the same order of magnitude per head as the chapter’s UK figure. That is my own rough comparison, not a published one. - Wider nitrogen costs. Van Grinsven et al. (2013) put total EU27 nitrogen social costs at EUR 75–485bn a year, with the benefits of agricultural nitrogen (EUR 20–80bn) below its pollution costs (EUR 35–230bn) [S19]. - The chapter’s “no European analysis” statement has been overtaken. The Commission’s 2026 evaluation estimates: - nitrogen pollution costs from agriculture of EUR 68–182bn a year; - Directive benefits of EUR 10–22bn a year; - implementation costs of EUR 2.8–3.1bn a year, a benefit-cost ratio of 3–7 [S13]. - It cites the Danish health-economic study [S13, fn 50]. - The endpoint has shifted away from bladder cancer. Bladder cancer is now the weakest of the candidate endpoints: a null pooled estimate [S8], and “suggestive” only in the Danish review [S12]. The case for costs of inaction now rests mainly on colorectal cancer and birth outcomes.

Verdict: unclear for the specific numbers (never replicated, and built on the endpoint that has fared worst). The chapter’s method, estimating a risk-based cost of inaction before proof, has been widely copied. In Denmark it helped trigger the precautionary re-evaluation [S12, Background].

Implication for weight. Do not carry the EUR 2.6bn or the per-kg figures forward as facts. The transferable point is that plausible, openly uncertain estimates can change the agenda, as the Danish sequence shows: cohort study, then economic analysis, then expert review, then a proposed limit. The specific figures are only as good as the endpoint they are anchored on.

Claim 3. Estimates of the costs of inaction “have often been grossly underestimated” because effects are left out: lead’s adult cardiovascular effects, the climate effects of ODS, and biodiversity (pp. 564, 568, 573, 576)#

Subsequent developments. - Lead: the omitted effect turned out to dominate. - Lanphear et al. (2018), using the NHANES-III cohort, attributed about 412,000 US deaths a year to low-level lead, including 256,000 from cardiovascular disease [S22]. - Larsen and Sánchez-Triana (2023) estimated 5.5 million lead-attributable cardiovascular deaths worldwide in 2019, “six times higher than the GBD 2019 estimate”, and a global cost of USD 6.0 trillion (6.9% of GDP). Of that, 77% was cardiovascular mortality [S21] (abstract). - GBD 2021 attributed about 1.48 million deaths to lead [S24]. GBD 2023, moving to cumulative bone-lead exposure, attributed 3.5 million deaths (5.8% of all deaths), making lead “the eighth leading risk for global mortality” [S23]. WHO’s June 2026 fact sheet adopts that figure [S25]. - The spread across these estimates (about 0.9m, 1.5m, 3.5m, 5.5m deaths) shows how far model structure drives the result. - ODS: climate costs are larger than the chapter counted. - WMO/UNEP (2022): Montreal Protocol controls avoid about 0.5–1 °C of warming by mid-century compared with an uncontrolled scenario [S42]. - “New evidence” suggests a further 0.5–1.0 K by 2100 from protecting the land carbon sink [S42, S46]. Young et al. (2021) put this at 325–690 billion tonnes more carbon held in plants and soils [S46]. - The 2016 Kigali Amendment, which targets HFC substitutes, is estimated to avoid 0.3–0.5 °C by 2100 [S42]. - Biodiversity and ecosystems. - The EEA’s own 2014 industrial-pollution costing excluded ecosystem-service impacts “such as harm to biodiversity” and said damage costs “are therefore likely to be under-estimated” [S71]. - IPBES (17 December 2024) estimated unaccounted-for costs of current economic activity, covering biodiversity, water, health and climate, at “at least $10–25 trillion per year” [S74]. - Undercounting in US regulatory analysis. The 2024 MATS amendments’ regulatory impact analysis could not monetise hazardous-air-pollutant benefits at all, as the 2026 repeal notes [S41]. That is an example of a zero entered for an unquantified effect. - Counter-evidence (upward revisions can be wrong). Kotz, Levermann and Wenz (2024) projected a committed 19% income loss from climate change by 2049. Nature retracted the paper on 3 December 2025 after others identified data anomalies [S54]. Upward revisions need the same scrutiny as downward ones.

Verdict: strengthened for lead and ODS, the two cases the chapter names. “Grossly” and “often” remain a generalisation from well-chosen cases. The chapter tests no example where a cost-of-inaction estimate proved too high.

Implication for weight. High weight for the mechanism: monetised estimates count what has been quantified, and each newly quantified pathway tends to raise them. Treat the size of any particular upward revision with caution. The lead estimates differ by a factor of about six depending on model choices, and a retracted climate study shows high-end numbers can fail.

Claim 4. EU air pollution costs EUR 276–427bn a year (3–5% of EU-25 GDP), 15–22% of GDP in Poland and the new Member States; SO2 damage EUR 5–9/kg against marginal abatement costs from below EUR 1/kg; even conservative VOLY-based bounds justify further abatement (pp. 571–573)#

Subsequent developments. - Scale confirmed and revised upward. - The Commission’s Clean Air Programme for Europe (COM(2013) 918, 18 December 2013) put 2010 health-related external costs at EUR 330–940bn. It found net benefits of its package “around €40 billion per year, according to the most conservative estimate” [S69]. - The EEA (2014) put damage from Europe’s registered industrial facilities alone at EUR2005 59–189bn in 2012 [S71]. - The conservative bound again justified action, and became law. - The 2022 impact assessment for the ambient air quality recast reported benefit-cost ratios of 10:1 to 28:1 (partial alignment with WHO guidelines), 7.5:1 to 21:1 (closer alignment) and 6:1 to 18:1 (full alignment), with net benefits of EUR 29–38bn [S68]. - Directive (EU) 2024/2881 records that “monetised health and environmental benefits significantly outweigh the expected implementation costs”. It sets a 2030 annual PM2.5 limit of 10 µg/m³ [S67]. - It also invokes the precautionary principle and gives individuals a right to compensation for health damage from breaches [S67]. That is a legal mechanism for the “diffuse interests” the chapter mentions. - Burden falling, but large. - EU-27 deaths attributable to PM2.5 above the WHO guideline fell 45% between 2005 and 2022, leaving 239,000 in 2022 (plus 70,000 from ozone and 48,000 from NO2). - Considering all concentrations, with no threshold, the PM2.5 figure would be 413,000 [S70]. - Poland and newer Member States. The World Bank (2022), using VSL and including household air pollution, estimated 2019 PM2.5 health costs at 9.8% of GDP for Poland, 16.3% for Bulgaria and 10.4% for Hungary, against 2.3% for France and 2.6% for the UK [S73]. Methods differ from CAFE, but the east–west gradient the chapter described persists. - SO2 abatement proved as cheap as claimed, until the low-hanging fruit ran out. - By 2022 all Member States but one met their 2020–29 SO2 reduction commitments, and 22 already met their 2030 commitments [S72]. - For fine particles the 2022 assessment found WHO-level concentrations “may not be possible at all for large parts of sampling points” by 2030, even with “virtually all technology options” [S68]. So marginal abatement costs do eventually rise.

Verdict: held up.

Implication for weight. This is the chapter’s strongest empirical case for its decision rule: report both VSL and VOLY bounds, then ask whether the conservative bound still justifies action. EU practice now follows it explicitly [S63, S68]. It supports the transferable lesson that reporting ranges, rather than forcing a single figure, can build agreement where valuation is contested.

Claim 5. Global mercury emissions are projected to rise 20% by 2020 relative to 2005 (AMAP 2011); marginal damage USD 1,500–6,000+ per kg; Greenland bears USD 59m a year with almost no local emissions (p. 569)#

Subsequent developments. - Treaty. The Minamata Convention was adopted at Kumamoto on 10 October 2013 and entered into force on 16 August 2017. It had 154 parties on 26 September 2026 [S32]. - In November 2025 the sixth Conference of the Parties (COP-6) reportedly decided to phase out the manufacture, import and export of dental amalgam by 2034. This rests on secondary sources [S33]. - The EU banned most dental-amalgam use from 1 January 2025 [S34]. - The emissions projection is contested. - The Global Mercury Assessment 2018 found 2015 anthropogenic emissions to air (2,220 t) “roughly 20% higher than … in 2010”. About 45% of that increase came from artisanal gold mining, and those estimates were judged “largely associated with improved information rather than a significant increase” [S30, pp. 3-20 to 3-21]. - Emissions fell in North America and the EU but rose 10–30% in most other regions [S30]. - Feinberg et al. (2024), using 51 monitoring stations, found Northern Hemisphere atmospheric mercury declining. They conclude that emissions “must have declined by at least 140 Mg between … 2005 and 2020” and that “existing emission inventories are incompatible with the observed … declines” [S31]. - Damage per kilogram: order of magnitude supported, with wide ranges. - Bellanger et al. (2013): EUR 8–9bn a year in IQ-related benefits of preventing methylmercury exposure in the EU [S37]. - Giang and Selin (2016): USD 339bn in cumulative lifetime US benefits from the Minamata Convention, with a range of USD 1.4bn to USD 575bn [S36]. - Zhang et al. (2021): USD 19 trillion (95% CI 4.7–54) in accumulated global health effects over 2010–2050 under current policy [S35]. - Arctic displacement and lock-in: strengthened. - AMAP’s 2021 health review describes Arctic blood mercury levels as “among the highest worldwide”. The Convention’s preamble cites Arctic and Indigenous vulnerability [S38]. - Methylmercury exposure from toothed-whale harvest exceeded tolerable intake in three of six Greenland municipalities between 1993 and 2020, and rose in one [S39]. - Isotope work shows ocean-borne legacy mercury driving present uptake. This explains why Arctic biota have not responded to falling deposition, which matters for any judgement of the Convention’s effectiveness [S40]. - US counting. The 2024 US MATS amendments were repealed in February 2026. EPA noted that hazardous-air-pollutant benefits “were not able to be monetized” [S41]. That supports the chapter’s point that unquantified effects enter decisions as zero.

Verdict: partly held up. - The damage and displacement claims held up, and on lock-in they strengthened. - The emissions projection is contested: inventories and observations disagree.

Implication for weight. Give high weight to the cross-border displacement and persistent-stock lock-in mechanisms: the people most exposed are not the emitters, and cuts at source take decades to reach them. Give low weight to any single emissions projection. Even the direction of the 2005–2020 trend depends on whether one trusts inventories or atmospheric observations.

Claim 6. The ozone layer should return to pre-1980 levels between 2050 and 2075; by 2050 about 47,000 skin cancers a year avoided in north-western Europe, with 14,000 extra a year still expected; EU-27 benefits EUR 10–11bn a year; the Montreal Protocol avoided about 215 GtCO2-eq, more than the Kyoto cuts (pp. 575–576)#

Subsequent developments. - Recovery timeline. The WMO/UNEP 2022 assessment (Executive Summary released January 2023) projects total column ozone returning to 1980 values: - “around 2066 in the Antarctic” (range 2049–2077); - “around 2045 in the Arctic”; - “around 2040 for the near-global average” [S42]. - The Antarctic date falls inside the chapter’s 2050–2075 window. Mid-latitude and global recovery is earlier. - The 2026 assessment is due “at the end of 2026” [S43]. - Compliance scare. - Unexpected CFC-11 emissions were detected in 2018. “Substantial emissions reductions followed” once the source region was identified [S42]. - Unexplained emissions of several other CFCs, carbon tetrachloride and HFC-23 persist, and monitoring gaps are “too large to determine whether all unexpected emissions have ceased” [S42]. - Treaty status. The Montreal Protocol has 198 parties. The Kigali Amendment (2016, in force 2019) has 174 [S48]. - Skin cancer. - Global modelling (van Dijk et al., 2013) found that by 2030 about two million skin cancers a year (14%) are prevented worldwide. At the peak, excess incidence in Western Europe under full compliance is 30–40 cases per million per year [S44]. - My rough comparison is that this is of the same order as the chapter’s 14,000 extra cases a year for north-western Europe. - US EPA modelling estimates 443 million skin cancers and 2.3 million skin-cancer deaths avoided for Americans born 1890–2100 [S45]. - Surface UV measurements confirm that UV Index values “remained essentially constant” rather than rising about 20% as they would have without the Protocol [S47]. - The EU-27 figure of EUR 10–11bn is the authors’ own scaling (from a US ratio in Sunstein, 2007). I found no independent check. - Climate. - The chapter’s “more than Kyoto” claim matches Velders et al. (2007) [S49]. - WMO/UNEP (2022) quantifies it as about 0.5–1 °C avoided by mid-century against a 3–3.5% annual growth scenario, close to the chapter’s 3% counterfactual [S42]. The carbon-sink and Kigali findings add more (Claim 3). - The specific 215 GtCO2-eq and EUR 2,150bn are the authors’ own illustrative calculations.

Verdict: held up. The recovery and health direction are confirmed, and the climate co-benefit is strengthened. The chapter’s own monetary and GtCO2 extrapolations remain unverified.

Implication for weight. The case supports two transferable lessons: - Lock-in. A 60-year latency means that harm already committed keeps arriving long after action (p. 576). - Co-benefits. Linked systems produce co-benefits that single-issue appraisals miss.

The 2018 CFC-11 episode adds a lesson the chapter did not draw: even a “solved” case needs continuing monitoring and verification capacity.

Claim 7. Stern: unabated climate change costs about 5% of GDP (up to 20%) against about 1% (at most 3.5%) for mitigation, and this “basic finding … was broadly accepted”; a social discount rate of about 1.4% is appropriate and corporate rates of up to 10% are not (pp. 565, 568, 574)#

Subsequent developments. - Benefits exceeding costs: supported, with medium confidence. - IPCC AR6 Working Group III (2022): “The global economic benefit of limiting warming to 2°C is reported to exceed the cost of mitigation in most of the assessed literature (medium confidence)”. - Mitigation pathways reduce 2050 global GDP by 1.3–2.7% compared with current policies, without counting avoided damages [S50]. This is close to Stern’s cost range. - Magnitudes: still contested. - Contemporary critics, notably Nordhaus (2007), held that Stern’s conclusions “depend decisively on the assumption of a near-zero time discount rate” [S51]. “Broadly accepted” was already an overstatement in 2013. - Stern himself argued in 2013 that standard models involve “gross underassessment of risk” [S52]. - Empirical studies since then span a wide range: Burke, Hsiang and Miguel (2015) projected about a 23% global income loss by 2100 [S53]. - One of the largest estimates, Kotz et al. (2024, a 19% committed loss by 2049), was retracted in December 2025 after data anomalies were identified [S54]. - Integrated-model estimates of the social cost of carbon rose sharply: USD 185/tCO2 at a 2% rate [S55], and a “major increase” in DICE-2023 [S56]. - Discount rates. - A survey of more than 200 experts found “more than three-quarters finding the median risk-free [social discount rate] of 2 percent acceptable” [S57]. That is close to, but above, the chapter’s 1.4%. - United States. - The 2023 revision of Circular A-4 (88 FR 77615, 13 November 2023) set a default social rate of time preference of 2.0% and endorsed declining long-term rates [S58]. - Executive Order 14192 (31 January 2025) ordered OMB to “revoke OMB Circular No. A-4 of November 9, 2023 … and … reinstate” the 2003 version, which uses 3% and 7% [S59]. - Executive Order 14154 (20 January 2025) withdrew the federal social cost of greenhouse gas estimates. It called the calculation “marked by logical deficiencies, a poor basis in empirical science, politicization” [S60]. - EPA’s February 2026 MATS repeal analysis reports only 3% and 7% rates [S41]. - EPA also rescinded the 2009 greenhouse-gas endangerment finding (18 February 2026). It did so on statutory-authority grounds, and expressly did not base the action “on a new finding” about the science [S61]. - UK. The Green Book (2026) keeps 3.5% for years 1–30, declining to 3.0% and then 2.5%. It uses 1.5% for health effects because the “wealth effect” does not apply [S62]. - EU. The Better Regulation Toolbox (December 2025) recommends 3% real, noting that “for discounting health impacts or environmental projects, it is a common practice … to choose a lower rate” [S63]. - The 10% rate. No government uses 10% as a social rate today. The US 7% (opportunity cost of capital), reinstated in 2025, is the nearest official analogue to the chapter’s “corporate” rate.

Verdict: contested. - The directional claim, that the benefits of action exceed its costs, has medium-confidence support from IPCC. - The magnitude of damages and the choice of discount rate remain disputed. - In the US, official discounting practice moved toward the chapter’s view in 2023 and was reversed in 2025.

Implication for weight. The robust lesson is the one the chapter puts in Table 23.1: the discount rate alone can move the estimated cost of inaction severalfold (from 14.7% to 4.2% of GDP in Stern’s figures, p. 574). That choice is ethical and political, not technical. The chapter’s normative preference for about 1.4% is a defensible position, not an established consensus. It should carry weight as advocacy, not as settled fact.

Claim 8. In January 2011 only six countries still allowed leaded petrol; remaining EU lead damage is under 0.1% of GDP; there is no safe blood lead level; urban playgrounds should be resurfaced (pp. 567–568)#

Subsequent developments. - Leaded petrol ended worldwide. UNEP announced on 30 August 2021 that “when service stations in Algeria stopped providing leaded petrol in July, the use of leaded petrol ended globally”. UNEP cited estimates that the ban prevents “more than 1.2 million premature deaths per year” and saves “USD 2.45 trillion” a year [S26]. - That figure is itself a cost-of-inaction estimate used in advocacy. - Lead remains in piston-engine aviation fuel. The US EPA made an endangerment finding for aircraft lead emissions on 20 October 2023 [S27]. - No safe level: affirmed and tightened. - WHO (10 June 2026): “There is no level of exposure to lead that is known to be without harmful effects.” Levels “as low as 3.5 µg/dL may be associated with decreased intelligence” [S25]. That figure reflects the US CDC reference value, lowered to 3.5 µg/dL [cited in S25]. - The EU lead-in-PVC restriction (Regulation (EU) 2023/923) treats lead as “a non-threshold toxic substance”. It notes EFSA’s finding that current exposure “still exceeds the tolerable exposure levels and leads to adverse neurodevelopmental effects in children” [S29]. - That restriction took five years from the scientific committee’s opinion (March 2018) to adoption (May 2023) [S29]. - Is residual EU damage under 0.1% of GDP? I found no update of the chapter’s estimate, which covered new emissions only. Total-burden estimates that include legacy exposure are far larger. From Larsen and Sánchez-Triana’s abstract, roughly 540,000 of 5.5 million lead-attributable cardiovascular deaths in 2019 occurred outside low- and middle-income countries [S21] (my subtraction). The chapter’s figure and the total burden measure different things. - Soil and playgrounds. - US EPA’s residential soil lead directive (16 October 2025) keeps a regional screening level of 200 ppm, with a 600 ppm removal management level. It superseded January 2024 guidance [S28]. - The EU halved its drinking-water lead value to 5 µg/L by 2036 [S1]. - I found no EU-level playground-resurfacing measure, but did not search national programmes.

Verdict: held up. The phase-out trajectory, “no safe level” and continued legacy harm are all confirmed. The <0.1% of GDP figure is neither confirmed nor refuted.

Implication for weight. The lead case supports two transferable lessons: - Staged, incentive-aligned transitions reduce resistance (p. 567). The global phase-out finished only in 2021, about 50 years after the first national bans. That shows how long the tail of adoption can be even after the evidence is settled. - Persistent legacy stocks keep generating harm (p. 568). The same sources support this.

Claim 9. Better science and modelling have shifted benefit-cost ratios toward regulation; credible, interdisciplinary estimates of the costs of inaction can justify precaution and strengthen diffuse interests (pp. 564, 577)#

This is a general prediction about how cost-benefit analysis works in practice.

Evidence that it held. - EU air quality. Monetised health benefits (VSL and VOLY, with ranges) underpinned the 2024 recast, which explicitly invokes precaution and creates compensation rights [S67, S68]. - Nitrate (Denmark and the EU). - The Danish ministry’s 2023 re-commissioning of DTU followed “an economic analysis … [suggesting] substantial health and economic benefits from lowering the nitrate standard” [S12, Background; S15]. - The resulting expert group proposed a precautionary limit [S12]. - The Commission’s 2026 evaluation produced the first EU-wide benefit-cost estimate of nitrate regulation [S13]. - This is close to the sequence the chapter recommended (p. 577). - Lead. The EU PVC restriction rested on a socio-economic committee opinion that explicitly used non-threshold reasoning [S29]. UNEP used a monetised benefit (USD 2.45 trillion a year) in promoting the leaded-petrol phase-out [S26]. - Better modelling raised estimates. Lead cardiovascular burden [S21, S23], the social cost of carbon [S55, S56] and ozone co-benefits [S42, S46] all rose with improved models, as the chapter predicted.

Evidence that it did not, or not reliably. - United States, 2025–26. The same kind of estimates were set aside by changing the accounting conventions rather than disputing the science: - the 2023 Circular A-4 revoked [S59]; - the social cost of greenhouse gases withdrawn [S60]; - EPA “refraining in providing primary estimates” of PM2.5 and ozone benefits, citing uncertainty “especially at low concentrations” [S41]. - In the 2026 MATS repeal, the unmonetised hazardous-air-pollutant benefits counted as nothing against USD 69–78m a year in compliance-cost savings [S41]. - Nitrate standards still unchanged 13 years later, despite the estimates [S1, S12]. - Credibility risk. The retraction of a high-profile damage estimate [S54] shows that high-end estimates can backfire if they fail scrutiny.

Verdict: partly held up. - Where institutions accept monetised health and environmental benefits as legitimate, which is now standard EU practice [S63], better estimates did shift decisions toward precaution. - Where the legitimacy of the valuation conventions was itself contested, better estimates did not protect regulation.

Implication for weight. Carry the chapter’s mechanism, that requiring a money figure for harm effectively sets the burden of proof, and add a qualification hindsight supplies: - The accounting rules (discount rate, which endpoints count, whether uncertain benefits are entered as ranges or as zero) are themselves a site of political contest. - Whoever sets them can neutralise even strong evidence.

This is the digest’s insight 13 (“a tool’s political legitimacy decides whether its results matter”), now well illustrated.

Claim 10. Political acceptance of valuing mortality through VSL is falling in the US (the 2011 withdrawal of ozone restrictions; the “senior death discount” backlash); Europe remains cautious about formal cost-benefit analysis (pp. 571, 574, 577)#

Subsequent developments. - US ozone standard. - After the White House returned the draft ozone rule on 2 September 2011, EPA revised the standard to 70 ppb in October 2015 (80 FR 65292). It retained that level in December 2020 (85 FR 87256) [S64, S65]. - The 2015 rule restates that in setting air quality standards “the EPA may not consider the costs of implementing the standards” (Whitman v. American Trucking, 2001) [S64]. - The chapter’s use of the 2011 episode as evidence about acceptance of VSL-based analysis therefore partly mislabels it. The episode was political intervention in a legally cost-blind standard. - US fine particles. EPA lowered the annual PM2.5 standard from 12.0 to 9.0 µg/m³ in March 2024 (89 FR 16202) [S66]. That fits a recovery of acceptance between 2015 and 2024. - US reversal, 2025–26. The sharpest decline in official acceptance of monetised mortality benefits came after 2025: - EPA’s February 2026 repeal of the 2024 MATS amendments stated it was “refraining” from primary PM2.5 and ozone benefit estimates [S41]; - the 2023 Circular A-4 was revoked [S59]. - I found no sign that an explicit age-based “senior death discount” was revived. The retreat took the form of declining to monetise benefits at all. - In Michigan v. EPA (2015) the Supreme Court held that the “appropriate and necessary” test for power-plant air toxics requires considering cost relative to benefits [cited in S41]. That made the counting of co-benefits a legal battleground. - EU. - The Better Regulation Toolbox (December 2025 edition) sets out monetisation of mortality using VSL and VOLY. It cites an OECD-derived EU adult VSL range of USD 1.8–5.4m (2005), with a base of USD 3.6m. It recommends a 3% social discount rate [S63]. - The 2022 air quality impact assessment reported monetised benefit-cost ranges [S68], and the 2026 Nitrates evaluation monetised benefits and costs [S13]. - Formal cost-benefit analysis still informs, rather than decides, EU legislation, and the Toolbox gives precaution its own section [S63]. - UK. The Green Book (2026) uses a Value of a Prevented Fatality, a value of a life year, QALYs, and a separate 1.5% health discount rate [S62].

Verdict: partly held up. - The US trajectory was not a steady decline. It fell (2011), recovered (2015–2024) and then fell sharply (2025–26). - Europe has become more, not less, routine in monetising mortality within impact assessment, though it still stops short of a binding net-benefit test.

Implication for weight. The transferable lesson is not “VSL is losing legitimacy”. It is that political acceptance of monetised harm swings with administrations. Decision frameworks that depend on a single monetised test are fragile. Frameworks that pair monetisation with statutory health standards or precautionary provisions, as the EU does, are sturdier.

Cross-cutting notes for the analytical lens (technology-neutral)#

Sources#