T05: Economics, costs, benefits and justice across the two Late Lessons reports#

Strand A thematic synthesis. Written 26 September 2026.

Sources. Digests for all 47 sections; full or targeted reading of the audited notes for the 33 sections that carry this theme; hindsight files for every section relied on; external/critiques.md and external/context.md; spot checks against working/text/chunks/.

Conventions. LL1 = EEA 2001, Late lessons from early warnings: the precautionary principle 1896–2000; LL2 = EEA 2013, Late lessons from early warnings: science, precaution, innovation. Pages are report pages (“LL2-08, p. 187”). “Hindsight LL2-08” means hindsight/LL2-08.md (developments to September 2026); where it flags a source as secondary or unverified, the flag is carried. Each pattern separates Reports (what the reports say), Evidence and hindsight (what the cases and later record show) and Analysis (my own reading), and counts the cases that support it and cut against it.

Evidence-strength scale. Strong: several cases, contemporaneous or independent support, not overturned later. Moderate: a few cases, or mainly protagonist or secondary sources, or right in direction with contested magnitudes. Suggestive: one or two cases, or an inference beyond them. Asserted: stated without supporting evidence in the text.

1. Summary of findings#

Economics is where the reports are most openly incomplete, yet it holds some of their most transferable insights.

• They set out to answer “who bore which costs and benefits, and when?” (LL1-00, p. 11), then conceded it was “the most difficult question”, its general analysis “beyond the scope” of the report (LL1-16, p. 168). LL2’s Part on “Costs, justice and innovation” has one methods chapter on costs (LL2-23).
• The best-supported findings are structural and technology-neutral: costs of acting are concentrated, visible and near-term while costs of not acting are diffuse and deferred; appraisals count only what is already quantified; valuation conventions move results by large factors and carry ethics; the evidential threshold decides who bears the cost of error; latency, persistence and insolvency turn private gains into public and future costs; compensation arrives late and partially, decided more by procedure than by science.
• The weakest material is numerical. Nearly every headline figure checked in hindsight was overstated, understated, misattributed or not like-for-like (vinyl chloride compliance, asbestos delay costs, hormone sanctions, invasive-species costs, the Deepwater Horizon cap, vaccine-compensation arithmetic).
• Ex ante overestimation of compliance costs is right in direction but modest in size. It holds in about eight cases, but the reports often compare a worst-case forecast for a stricter proposal with the measured cost of a weaker rule; the wider literature finds only a slight tendency to overestimate.
• Justice is raised more than analysed. Minamata (LL2-05) is the one full environmental-justice case, and its mechanisms (hardened case definitions, relief without recognition, payer-adjudicated counting, the state as co-financier of the polluter) have a strong 13-year out-of-sample record. Elsewhere distribution appears case by case, consistently: harm falls on workers, neighbours, the poor, other countries and future generations; benefits flow to producers and users. LL2’s conclusion puts unequal political power “beyond the scope” of the report (LL2-28, p. 672).
• The reports underweight the costs of precaution, including several they document, and barely address the critics’ strongest economic point: precautionary costs can be regressive.

2. What the reports tried to do on costs, and what they did not#

Reports. - LL1’s Preface sets the frame. The costs of preventive action are “usually tangible, clearly allocated and often short term”, while those of failing to act are “less tangible, less clearly distributed and usually longer term, posing particular problems of governance”. Weighing them involves “ethical as well as economic considerations” (LL1-00, pp. 3–4). LL2 quotes this passage to open its costs chapter (LL2-23, p. 565). - LL1 invites readers to judge whether earlier action would have come “at a lower overall cost to society” (LL1-00, p. 3), but the synthesis could not show this systematically: case authors were not economists, and pros and cons are incommensurable and unevenly distributed (LL1-16, p. 168). Several chapters give no costs at all: radiation (LL1-03), DES (LL1-08), PCBs (LL1-06), TBT beyond oyster tonnage (LL1-13). - LL2 treats costs through a methods review arguing that “precautionary action can be justified by using credible estimates of the costs of inaction” (LL2-23, p. 577), a legal-policy essay on compensation (LL2-24) and an analysis of why firms did not act (LL2-25). Its working definition of the principle makes distribution explicit: “the pros and cons of action and inaction and their distribution” “across groups, regions, and generations” (LL2-27, p. 649).

Analysis. The EEA’s founding regulation asks it to “stimulate the development of methods of assessing the cost of damage to the environment” (context.md, Art. 2(j)), yet its flagship report on precaution put general cost-benefit analysis out of scope. The reports document well how costs and harms were mis-seen and mis-allocated, but not how large they were on either side of the ledger.

3. Patterns#

3.1 The founding asymmetry: concentrated costs of action, diffuse costs of inaction#

Reports. The asymmetry (LL1-00, pp. 3–4) recurs throughout. Lead was an “unequal contest”: benefits accruing “to particular and powerful minorities” against “unproven, general, future threats” (LL2-03, pp. 52–53, 69–70). Compliance costs “usually fall on specific, often powerful actors”, against “long-term diffuse benefits” (LL2-27, p. 659); diffuse interests “have difficulty making their voices heard” (LL2-23, pp. 564–565, 577).

Cases supporting (about 10): lead (LL2-03); northern cod, where Harris recommended 190,000 t, not the roughly 125,000 t the target required, to avoid “drastic” social and economic repercussions (LL1-02, pp. 21–22); EU fisheries, with TACs averaging 47% above advice after 2003 (LL2-17, pp. 421–422); Great Lakes, with remediation costs itemised and benefits almost never estimated (LL1-12, pp. 130–131); BSE, where export and spending concerns dominated the first 20 months (LL1-15, pp. 159–160); benzene, limited at 1 ppm on economic feasibility (LL1-04, p. 41); nitrate derogations (LL2-23, p. 570); acid rain (LL1-10, pp. 105, 107); climate, where concentrated short-term losers lobbied against diffuse gains (LL2-14, pp. 337–338); neonicotinoids, where maize, with far larger stakes than sunflower, was restricted five years later (LL2-16, p. 382).

Cases cutting against (4). Where a signature harm made damage vivid and a cheap fix existed, action was fast despite concentrated costs: vinyl chloride within months of the 1974 disclosure (LL2-08, pp. 186–187); DBCP, about two months to an emergency standard (LL2-09, pp. 206–207); the US DES contraindication within about seven months (LL1-08, pp. 84, 86). Swedish farmers themselves requested the 1986 growth-promoter ban (LL1-09, p. 95). The asymmetry is a tendency, not a law.

Evidence and hindsight. It keeps reappearing: Canada more than doubled its 2025 cod quota despite advice that no removal level gave a high probability of growth, and the EU missed its 2020 sustainable-yield target (hindsight LL2-17, Claims 5 and 8); the US benzene limit has stayed at 1 ppm since 1987 while other bodies moved to 0.02–0.2 ppm (hindsight LL1-04). A legal target plus annual public reporting reduced override without ending it (81% of EU 2026 Atlantic catch limits in line with advice; hindsight LL2-17).

Rating: strong as a description across many cases. Moderate as a causal explanation: the reports illustrate rather than measure it, and fast-action cases show it can be overridden.

Analysis. The asymmetry concerns the legibility and allocation of costs as much as their size: costs of action land on identifiable parties with standing and lobbying capacity; costs of inaction on people who are dispersed, unborn, abroad or unaware.

3.2 Undercounting: the boundaries of an appraisal decide its answer#

Reports. Costs of inaction “have often been grossly underestimated” because whole pathways are omitted: lead’s adult cardiovascular effects (LL2-23, p. 568), biodiversity in air-pollution estimates (p. 573), the climate effect of ozone-depleting substances (p. 576). Because cost-benefit analysis leans “toward conservative values”, the known error runs one way (p. 568). In LL1, benzene cost-of-delay estimates left out diseases without dose-response data (LL1-04, p. 41); Norway argued that acid-rain appraisal must count damage in emitting countries, where materials damage alone was “comparable” to abatement costs (LL1-10, pp. 103–104); the Great Lakes chapter itemises remediation costs but no health, fishery or wildlife losses (LL1-12, pp. 130–131). LL2-27 adds lost trust, sometimes “as significant as the economic costs” (p. 658).

Cases supporting (6): lead, ozone-depleting substances, SO2/air pollution, benzene, Great Lakes, invasive species in Europe (LL2-20, p. 491). Cutting against (3): the upper end of California’s MTBE treatment estimate (USD 340–1,480 million a year; LL1-11, p. 119) looks high in hindsight, partly because the phase-out removed the source (hindsight LL1-11, Claim 6); the Dutch asbestos counterfactual rested on projections later cut by 44% (hindsight LL1-05, Claim 8); the global invasive-species figure of USD 1.4 trillion was misattributed (hindsight LL2-20, Claim 3).

Evidence and hindsight. Strongly vindicated for the named pathways. Lead-attributable deaths rose from about 0.9 million (GBD 2019) to 3.5 million (GBD 2023) as cardiovascular effects were modelled (hindsight LL2-23, Claim 3). For acid rain, the decisive benefit was avoided fine-particle deaths, which the 2001 chapter never counted: US Acid Rain Program benefits of over USD 100 billion a year against about USD 3 billion in costs (hindsight LL1-10, Claim 4). Protecting the land carbon sink added 0.5–1.0 °C of avoided warming to the ozone case; only about 1% of EU invasive species have any reported cost (hindsight LL2-23, LL2-20). Qualifications: lead estimates span about six-fold with model structure; a high-profile climate-damage estimate was retracted in December 2025; a 2026 US repeal entered unmonetised air-toxics benefits as zero (hindsight LL2-23, Claims 3 and 9).

Rating: strong for the mechanism: estimates count what is already quantified, and newly quantified pathways tend to raise them. Moderate for any particular revision. Asserted for “often grossly”.

Analysis. The error is not that estimates are always too low but that omission is systematic in one direction for long-latency, multi-pathway harms, while costs of action are itemised from engineering quotes. Where an appraisal draws its boundary (which effects, countries, years) is a distributive decision, usually made before anyone argues about numbers.

3.3 Ex ante overestimation of compliance costs#

Reports. - Vinyl chloride. Compliance cost “USD 278 million” against industry forecasts of “up to USD 90 billion and 2 million jobs” (LL2-08, p. 187); the lesson restates the forecast as “USD 60 million annually” (p. 192). - Lead. Industry claimed phase-out would waste “one million barrels of oil a day” against the EPA’s 30,000 (LL2-03, p. 60); German cost claims were “significantly biased, due to the vested interests” (p. 65). - DES as a feed additive. A predicted USD 500 million-a-year consumer cost was “probably groundless” because substitutes existed (LL1-14, pp. 149–150). - Beryllium. A 1977 claim that classification would mean “no longer a viable industry” was not borne out (LL2-06, pp. 135, 145). - SO2. “A million-dollar problem with a billion-dollar solution” (LL1-10, p. 103; LL2-23, p. 572); Japan’s 1975 abatement had “practically negligible” macroeconomic effect and yielded patents (LL2-23, p. 573). - Substitutes. An asbestos-free Thai roof cost USD 65 extra, 0.5% of the total (unsourced; LL2-A3, pp. 724–725); growth promoters showed a “net loss for producers” on the annex author’s reanalysis of an industry trial (pp. 732–733); wet cleaning saved one shop USD 2,749 a year (LL2-04, p. 87). - Regulation “can stimulate innovation” (LL2-08, p. 192, hedged in fn 7 as holding “in many but only specific circumstances”; LL2-27, p. 659).

Cases supporting: about 8 (vinyl chloride, lead, DES feed use, beryllium, SO2, asbestos substitutes, growth promoters, PCE).

Cases cutting against or complicating (5). Swiss micropollutant treatment cost CHF 12–55 per connected person against the chapter’s EUR 5 (LL2-13, p. 295 fn 10; hindsight LL2-13, Claim 4). Olkiluoto 3 and Flamanville 3 ran to about four times budget (LL2-18, pp. 443–444; hindsight LL2-18). French sugar-beet yields fell from 851 to 649 q/ha in 2020 after the neonicotinoid ban, during a virus-yellows outbreak (hindsight LL2-16). Danish post-weaning diarrhoea and therapeutic antibiotic use rose after growth promoters were withdrawn (hindsight LL1-09). Norway’s cod buyout cost over EUR 50 million and fishers’ incomes fell by up to 40% (LL2-17, p. 414).

Evidence and hindsight. - The reports’ comparisons are often not like-for-like. The vinyl chloride USD 90 billion probably costed a stricter “no detectable level” proposal assuming plant closures; the like-for-like comparison (over USD 1 billion forecast against USD 228–278 million actual) is about four-fold, not 300-fold (hindsight LL2-08, Claim 3). The lead “one million barrels” was a newspaper advertisement (hindsight LL2-03, Claim 6). The DES claim has no ex post evaluation (hindsight LL1-14, Claim 7). - The general literature is guarded. Harrington et al. (2000) found total costs overestimated for 14 of 28 rules and underestimated for 3, unit costs of incentive-based rules “consistently overestimated”, often through unanticipated innovation; Simpson (2014) could not reject unbiasedness; Morgenstern (2018) found “a slight tendency to overestimate both costs and benefits” (hindsight LL2-03, LL2-08). - Where it held best. The US Acid Rain Program cost less than half the 1990 estimate (hindsight LL1-10); the WHO panel put the net cost of the Danish growth-promoter ban at “just over 1%” of pig production costs (hindsight LL1-09). - The argument recurs. US proclamations in 2025–26 exempted chemical plants from a fenceline-monitoring rule because the technology “is not available”, the 1974 argument, this time successful (hindsight LL2-08, Claim 7).

Rating: moderate. The direction is supported in several cases and by the broader literature; the reports’ magnitudes are unreliable, and projections for novel remedies can be too optimistic.

Analysis. The best-supported version is conditional. Regulated parties’ cost forecasts tend to be high when a rule leaves room for technical change, when substitutes already exist within the product class, and when the forecaster costs a worst case rather than the rule adopted. Restricting one agent within a class of substitutes is cheap; restricting the whole class is not (hindsight LL1-14). Forecasts are evidence produced by interested parties, the reports’ own point about hazard evidence, and should be tested ex post rather than accepted or dismissed by analogy.

3.4 Valuation conventions: discounting, the value of life, and the conservative-bound test#

Reports. Technical choices “decisively” drive results (LL2-23, p. 567). Lead damage per gram ranges from 8 US cents at a 10% private rate to EUR 5.90 undiscounted (p. 568 fn 1; the spread also mixes studies, years and currencies). In Stern’s figures the costs of inaction fall from 14.7% to 4.2% of GDP as the discount rate rises from 1.3% to 2.8% (Table 23.1, p. 574). Life-year (VOLY) and statistical-life (VSL) valuations differ by about 1.5 times, and an age adjustment proposed in the US was branded a “senior death discount” (p. 571). Converting IQ loss to lifetime income makes losses “highest for females” (p. 567). “Many non-economists” see discounting as “at odds with the core idea of intergenerational equity”, and corporate rates up to 10% are applied “without reflecting” (p. 574); the chapter recommends a social rate of about 1.4%. Its decision rule: report ranges and check whether even the conservative bound justifies action, as it did for SO2 (EUR 5–9 damage per kg against abatement “from below EUR 1”, pp. 571, 573). LL2-14 concedes on Stern’s low rates that “Not all economists agree” (p. 319); a Niagara study ranked options at different discount rates (LL1-12, p. 130). The Peach Council’s view that worker sterility “may be bad, it is not necessarily so” (LL2-09, p. 206) amounts to informal discounting of harm by who suffers it.

Cases supporting: 4 (lead, SO2/air, climate, Niagara). Cutting against: none directly, though calling Stern “broadly accepted” (p. 565) understated the dispute (Nordhaus 2007; hindsight LL2-23, Claim 7).

Evidence and hindsight. The conservative-bound rule became EU practice: the 2022 impact assessment for the air-quality recast reported benefit-cost ratios of 6:1 to 28:1, and the 2024 directive tightened the PM2.5 limit and gave individuals a right to compensation for breaches (hindsight LL2-23, Claim 4). Discounting proved politically contingent: the US set a 2% default in 2023, revoked it in 2025, withdrew the social cost of greenhouse gases, and by 2026 the EPA was “refraining” from primary estimates of PM2.5 and ozone benefits, while the UK and EU held steady (hindsight LL2-23, Claims 7 and 10). More than three-quarters of 200-plus surveyed experts found a 2% social rate acceptable, above the chapter’s 1.4% (Drupp et al. 2018; hindsight LL2-23).

Rating: strong that valuation conventions move results several-fold and embed ethical judgements (the chapter’s own numbers, confirmed by later practice). Moderate that reporting ranges and testing the conservative bound builds durable decisions (one domain, now institutionalised). Asserted/advocacy for the 1.4% rate.

Analysis. Hindsight supplies the qualification the chapter lacked: credible estimates do not “carry” precaution on their own. Whoever controls the accounting conventions (discount rate, which endpoints are monetised, whether unquantified effects enter as ranges or as zero) decides what counts. Frameworks pairing monetisation with statutory health standards proved sturdier than ones resting on a single monetised test. Discounting is also the formal mechanism by which future victims weigh less than present payers (§3.13).

3.5 The level of proof is a rule for allocating the cost of error#

Reports. Choosing the level of proof “can radically shift the size, nature and distribution of the costs of being wrong”; it is a political and ethical choice set by the harm, the claimed benefits, the alternatives and “the potential costs of being wrong in both directions” (LL1-17, p. 193). The Swedish commission asked who “would bear the costs of waiting … the risk-maker or the risk-taker?” (LL1-09, p. 96). Precaution should reverse the burden “so that risk makers, not risk takers” show safety, and Box 8.1 quotes Justice Marshall’s benzene dissent on “the burden of medical uncertainty squarely on the shoulders of the American worker” (LL2-08, p. 187). “Not established” judgements seldom say who bears the error (LL2-27, p. 658); tort’s burden of proof and post-Daubert gatekeeping “asymmetrically hamper plaintiffs” (LL2-24, pp. 586–593); waiting for a monetised figure is itself a burden of proof on warners (LL2-23, p. 564).

Cases supporting: about 9: benzene, where the 1980 Supreme Court required a quantified “significant risk” finding (LL1-04, p. 40); Minamata’s 1957 refusal to apply food law for lack of “clear evidence that all fish … are poisoned” (LL2-05, pp. 98–99); the nitrate guide value dropped for “absence of scientific proof” (LL2-23, p. 570); neonicotinoids, where the regulator answered “sole cause, everywhere”, a question “never asked” (LL2-16, p. 379); asbestos causation law (LL2-24, Panel 24.2); growth promoters (LL1-09, pp. 95–96); DBCP, vinyl chloride and cod (LL2-17, p. 413).

Evidence and hindsight. Law now reflects the claim, in both directions. Towards claimants: Pfizer v Council (2002); UK mesothelioma law from Fairchild to the Compensation Act 2006 and Sienkiewicz (2011); Camp Lejeune’s “at least as likely as not” standard (2022); causation presumptions in the EU’s 2024 Product Liability Directive. Away from claimants: the tightening of US Federal Rule of Evidence 702 (2023) and a 2025 US executive order disfavouring “overly precautionary assumptions” (hindsight LL1-17, LL2-24, LL2-27).

Rating: strong as a conceptual and legal point; the allocation effects are visible in the litigation record (§3.11).

Analysis. Every evidential standard is also a distributive rule: it decides who carries the cost of scientific uncertainty while it lasts. Because uncertainty about latent harm lasts decades, the standard can decide the whole outcome.

3.6 Who gains, who bears: distribution of benefits, costs and risks#

Reports. Distribution runs through almost every case, but mostly as description. - Non-consenting third parties bear harm while producers and users benefit: second-hand smoke exposure of partners, children and hospitality workers (LL2-07, pp. 151, 153); DES daughters and sons (LL1-08, pp. 84–88); asbestos workers’ families and neighbours (LL1-05, pp. 55–56); resistance borne by patients and health systems (LL1-09, p. 96); MTBE costs “borne by society” (LL1-11, p. 119); beekeepers’ uncompensated research and legal costs (LL2-16, p. 383). LL2-04 states the principle: acceptable risk belongs to “the use and its beneficiaries”, not the substance (pp. 80, 83). - Harm falls on the poor and less powerful: Minamata’s fishing families (LL2-05, pp. 96, 99); legacy lead and e-waste among “children in poor countries” (LL2-03, pp. 66, 69); climate, where “the societies that have contributed most … are generally least affected” (LL2-14, p. 309); DDT spray residents, “the largest non-occupationally exposed community in the world” (LL2-11, pp. 247–249); GM benefits that “largely bypass the poor” (LL2-19, pp. 460, 476); poor flood-zone migrants (LL2-15, pp. 359, 362); place-bound fishing communities while mobile capital “mines” the resource and leaves (LL1-02, pp. 19–20; LL2-17, p. 419). - Workers first: benzene (LL1-04, pp. 42–43, 45), beryllium, vinyl chloride, DBCP (LL2-06, -08, -09), dial painters and radiologists (LL1-03, pp. 32–33). - Harm displaced elsewhere: tall stacks moving SO2 damage to Scandinavia (LL1-10, pp. 101–103; LL2-23, p. 572); Greenland bearing USD 59 million a year of mercury damage with “hardly any local emissions” (LL2-23, p. 569); Chisso’s outfall move (LL2-05, p. 100); exports after domestic restriction, such as DBCP to plantations until 1985–86 (LL2-09, pp. 207–209), asbestos growth in Asia “offsetting” bans (LL2-A3, pp. 724–726) and leaded petrol (LL2-03, pp. 67–68); Turkey’s shoe industry switching to benzene in 1961 because it was cheaper (LL1-04, p. 39). - Precaution’s costs are distributed too: hormone sanctions hit exporters of unrelated foods (LL1-14, p. 153; §4.2–4.3).

Cases supporting: more than 15. Cutting against: none on the direction of harm. The pattern is complicated where public authorities, not firms, controlled exposure (§4.6) and where the “victims” of precaution are poor (DDT and malaria; aflatoxin standards, §4.3).

Evidence and hindsight. Lead’s legacy is larger and more concentrated on the poor than the chapter implied: 632 million children in 34 low- and middle-income countries exceed 5 µg/dL, and 48% of English cases in 2024 lived in the most deprived areas (hindsight LL2-03, Claims 2 and 5). Arctic mercury exposure is “among the highest worldwide” (hindsight LL2-23, Claim 5). DBCP use abroad is uncontested in court, yet no controlled study of plantation workers exists, and DBCP was never listed under the Rotterdam Convention (hindsight LL2-09, Claims 5 and 8). Asbestos use continues mainly in Asian construction, 930,000 t in 2025 (hindsight LL1-05).

Rating: strong as a descriptive pattern across both reports. Suggestive as quantified distribution: the reports almost never measure who bears what. Even LL2-23, in a Part titled “justice”, raises equity (Poland’s 15–22% of GDP, Greenland, the “senior death discount”) without analysing it (notes LL2-23).

Analysis. Three technology-neutral dynamics recur: 1. Consent and benefit are decoupled. Those exposed are rarely those who chose or gain. 2. Displacement relieves the visible, local indicator while harm grows elsewhere (tall stacks, outfalls, exports). 3. Evidence follows power. Harm to people without standing, money or proximity to researchers is studied least.

3.7 Distribution shapes political will and timing#

Reports. Whether more evidence is “paralysis” or “prudent and careful evaluation” depends on how pros and cons “impinge upon” each group; early action is likely when benefits are large, harms small, and both evenly spread (LL1-16, p. 182; Box 16.1, p. 170). In acid rain, positions followed perceived costs: Scandinavia found action “relatively easy” to argue for, the UK moved “arguably” only when costs “closer to home” became clear, and Eastern Europe could not pay (LL1-10, p. 107; LL1-16, p. 176). Commercial salience triggers first action: TBT harm was tolerated while the victim was a pest species and acted on when Arcachon oyster production fell from 10,000–15,000 t to 3,000 t by 1981, about USD 147 million in losses (LL1-13, p. 136; LL2-12, pp. 266–267). Invasive-species policy may have been “helped” by monetised costs (LL2-20, p. 491).

Cases supporting: about 6 (SO2, TBT, neonicotinoid sunflower-before-maize, climate as a negative case, invasive species, growth promoters where farmers broke ranks). Cutting against: 2. UK and global TBT action later rested on harm to non-commercial species once documented (LL1-13, pp. 137, 139–142). And the US 1977 aerosol ban preceded observed harm (LL1-07, p. 80).

Evidence and hindsight. Hindsight confirms the UK’s acid-rain reversal and the later health-benefit accounting that made abatement a domestic gain (hindsight LL1-10). Great Lakes remediation gained durable support once packaged as bounded local projects with visible economic returns (“USD 3.35 per USD 1 invested”), not health benefits (hindsight LL1-12, Claim 6). Beekeepers were vindicated through courts once they had standing (hindsight LL2-25, Claim 2).

Rating: moderate. One strong worked example (SO2) plus theory and several consistent cases.

Analysis. Political will is itself a function of distribution: harms become actionable when they land on a party with standing, market value or domestic political weight. A lens should ask not only “how bad is the harm?” but “does it fall on anyone able to make it count?”

3.8 Prices that exclude harm confer advantage and lock incumbents in#

Reports. Because the prices of asbestos, halocarbons and PCBs excluded health and environmental costs, they gained “an unjustifiable advantage in the marketplace”, keeping “technically superior substitutes” out longer than was socially optimal; internalisation is “essential” for efficiency and equity (LL1-16, pp. 176–177). Asbestos substitutes existed “for most uses” by the 1970s but were slowed by cartels and prices that excluded health costs (LL1-05, p. 58). Late action “consolidated technological monopolies … at unrealistically low prices” (LL2-27, p. 659; LL2-28, p. 673). Lead alternatives were denied and forgotten (LL2-03, p. 55), reportedly because they “were not patent-protected” (LL2-23, p. 567; one secondary source). DDT is cheap, long-lasting and fitted to routine (LL2-11, pp. 243–244, 251–252). At Minamata, mercury “was taken off the list on the assumption that such an expensive material would never be thrown away in the sea” (LL2-05, p. 100): price beliefs shaped the search for the cause itself.

Cases supporting: about 8 (asbestos, halocarbons, PCBs, lead, DDT, benzene in Turkey, growth promoters, neonicotinoids through prophylactic bundling, LL2-16, pp. 383–385). Complicating: substitutes were not always “smarter”; regrettable substitution recurs (MTBE for lead, LL1-11; HCFCs/HFCs, LL1-07; booster biocides, LL2-12).

Evidence and hindsight. Lock-in is strongly confirmed. The US EPA’s 2024 chrysotile rule found eight chlor-alkali plants, 42–83 years old, still using asbestos diaphragms; conversion would cost USD 2.8–3.4 billion but save energy (hindsight LL2-27, Claim 7). Mercury use in China’s PVC production reportedly doubled from 2004 to 2023 because the coal route is 7–17% cheaper (hindsight LL2-08). The Porter-hypothesis literature supports induced innovation but not net gains to firms (hindsight LL2-08, LL2-27).

Rating: strong for lock-in through long-lived capital and cheap incumbents. Moderate for the price-exclusion mechanism. Suggestive/advocacy for “early action produces smarter substitutes”.

Analysis. Unpriced harm is a subsidy that compounds. Each year of use embeds the technology in capital stock, skills and supply chains, so later restriction costs more and the incumbent’s price advantage looks like efficiency. The costs of delay include unwinding lock-in, which appears in no early appraisal.

3.9 The cost of delay: cheap early action against expensive late remediation#

Reports. About 1,200 clinical BSE cases could have been removed for about £1.5 million in 1987–88; the later bill was a forecast £4.2 billion (1996–2001) plus a lost £700 million-a-year export trade (LL1-15, pp. 158, 164). Weed eradication costs rise “at least 40 times” with delay; Caulerpa was detected early in France in 1984 but tackled late, while California began eradication 17 days after detection (LL2-20, pp. 487, 498). Asbestos deaths cost EUR 400 billion at EUR 1 million per life, and a Dutch estimate held that a 1965 ban would have saved 34,000 victims and NLG 41 billion (LL1-05, p. 58). The cod collapse cost “in excess of several billion” CAD (LL1-02, p. 22). For BSE and cod the pros and cons of early action were foreseen, but uncertainty and cost limited it, leading to “far greater costs (many of which were anticipated)” (LL1-16, p. 176).

Cases supporting: about 7 (BSE, invasive species, asbestos, cod, Great Lakes, lead, MTBE). Cutting against or complicating (3): most BSE infectivity eaten came from cattle without clinical signs, so the £1.5 million step would have averted only a small share of exposure (hindsight LL1-15, Claim 4); Caulerpa later collapsed across much of its Mediterranean range (hindsight LL2-20, Claim 4); the 1990 cod counterfactual is unproven, and restraint was needed across the 1980s (hindsight LL1-02, Claim 3).

Evidence and hindsight. Asbestos ban-effect studies now support the direction: 8,341–21,981 Italian mesothelioma deaths avoided in 1992–2020, and a Swedish cohort relative risk of 0.16 (hindsight LL1-05, Claim 8). Invasive-species management lagged damage by about 11 years on average, adding about USD 1.2 trillion (hindsight LL2-20). BSE public spending reached about £4.8 billion by 2003, some of it buying back credibility and trade access rather than safety; some markets stayed closed until 2019–20 (hindsight LL1-15, Claim 6). The cod collapse cost about 30,000 jobs (hindsight LL1-02, Claim 1). The reports’ asbestos figures erred in both directions (§4.4).

Rating: moderate. The direction is well supported, now with quasi-experimental evidence for asbestos; the specific counterfactual costings are weak, often not like-for-like, and made with hindsight. The one direct juxtaposition of a cheap step and the later bill (BSE) overstates what the step would have averted.

Analysis. Two things compound over time: the physical stock of harm (exposure pipelines, dispersed persistent agents) and the institutional cost of reversal (lock-in, lost credibility, trade restrictions). The second is less often counted. BSE shows that much of the late bill paid for restoring confidence, not for reducing risk.

3.10 Who pays for action: intervention point, public purse and polluter pays#

Reports. For ethinyl oestradiol, the point of control decides who pays: the costed UK option was end-of-pipe sewage treatment, with costs on water companies and the public, while the authors float redesign of the drug at source as “a constructive precautionary approach” and call Rio’s “cost effective” wording precaution’s “Achilles heel” (LL2-13, pp. 290–291, 294–296). Taxpayers fund Great Lakes clean-up (LL1-12, pp. 130, 132). Harmful species stay on sale while public money pays to contain them (LL2-20, p. 492). Most BSE spending was compensation to industry (LL1-15, p. 164). The ozone Multilateral Fund had disbursed over USD 1 billion to help developing countries transition (LL1-07, pp. 80–81), and Norway bought out fleet capacity (LL2-17, p. 414). Depleted resources cannot fund their own recovery (LL1-02, p. 26).

Cases supporting: about 8. Complicating: public transition funding (ozone fund, BSE compensation, fleet buyouts) buys compliance and speed at the cost of polluter pays.

Evidence and hindsight. The EU’s 2024 urban wastewater recast requires producers of medicines and cosmetics to fund at least 80% of new micropollutant treatment, which turned the cost dispute into litigation by the pharmaceutical federation and Poland (hindsight LL2-13, Claim 3). The European Court of Auditors (2021) found public budgets fund more than 42% of soil remediation; in 20 of 42 EU-funded projects the polluter could not be made to pay (hindsight LL2-25, Claim 5). US federal tank funds put USD 200 million of public money into MTBE leaks even as courts assigned producer liability (hindsight LL1-11, Claim 9).

Rating: strong that the intervention point allocates costs and that public budgets absorb them by default. Moderate for the effectiveness of producer-pays instruments, which are recent and contested.

Analysis. “Who pays for the fix” is decided largely by where it is applied and who is still solvent and identifiable when it is. Source-level measures place costs on producers; end-of-pipe and remediation measures place them on utilities, ratepayers and taxpayers. Transition funds can legitimately share costs with parties who did not cause the problem, but they weaken the price signal the reports want restored.

3.11 Compensation, liability and insurance: late, partial, and weak as deterrence#

Reports. - Tort is slow and costly, “a poor legal model for providing rapid and adequate compensation” (LL2-24, p. 589). Claimants carry the burden of proof, some courts demand a doubled relative risk, damages are capped, and cases drag: Milward ran from 2007 to past 2012; Graham v ReChem took 198 trial days and was lost on causation (pp. 587–591). - Compensation as closure. Chisso’s 1959 “sympathy money” carried a waiver of future claims, voided in 1973; the 1995–96 and 2009 settlements paid “relief money (not compensation)” without recognising patients (LL2-05, pp. 107–110). DBCP settlements averaged about USD 1,500 per worker in 1997 (LL2-09, pp. 209–210). - Insolvency and latency. Manville’s 1982 bankruptcy was “a means of dealing with” asbestos claims (LL2-25, p. 612; LL1-05, p. 56). Firms may not survive until long-tail harm appears (LL2-24, p. 596). Invasive-species lag phases mean “traceability … may be lost”, making liability “very difficult” (LL2-20, p. 497). - Caps socialise tail risk. European nuclear operator caps (EUR 700 million; EUR 1.5 billion in total) stand against Fukushima liabilities of EUR 76–152 billion (LL2-18, pp. 445–446); Deepwater Horizon damages of USD 34–670 billion against a cap given as USD 50 million (LL2-24, pp. 602–603). - Liability deters admission. Monsanto’s 1969 plan ruled out stopping PCB production lest “profits … cease and liability … soar” (LL1-06, p. 65; a paraphrase via one secondary source). Brush Wellman called its limit “fundamental to our product liability defense” (LL2-06, p. 137). Guidotti argues firms need an exit route (“forgiving past liability”) to change course (LL2-06, pp. 149–150). - Liability rebounds and discloses. Asbestos bankrupted Manville, nearly collapsed Lloyd’s, and led Turner to set aside up to £1 billion (LL1-05, pp. 56, 58). Discovery revealed “until then secret” conduct (LL2-08, p. 179). Insurers refused cover for asbestos workers in 1918, “early precautionary action” that protected insurers, not workers (LL1-05, p. 54). - Alternatives. No-fault schemes: workers’ compensation, the US vaccine scheme (VICP), the 9/11 fund, New Zealand’s scheme, and the UK radiation scheme (CSRLD), which pays from 20% probability of causation (LL2-24, pp. 594–600; LL1-03, pp. 35–36). Anticipatory compensation “to both increase the incentives to prevent further harm and to improve … accurate exposure histories” (LL1-05, pp. 60–61). Assurance bonds sized to worst-case damage (LL2-24, pp. 600–603; LL2-25, pp. 611–612, 616; LL2-28, pp. 679–680). LL2-24 concedes that compensation tables need “a history of previous diseases” and that the feedback to prevention is “modest” (pp. 599, 603).

Cases supporting “late, partial, procedurally decided”: about 10 (asbestos, Minamata, DBCP, MTBE, PCBs, vinyl chloride, beryllium, nuclear, oil spill, neonicotinoids). Cutting against: mesothelioma law moved toward claimants (Panel 24.2); MTBE liability eventually bit.

Evidence and hindsight. - Late and partial. The Manville Trust now pays 5.6% of scheduled claim value (hindsight LL2-25, Claim 5). More than half of USD 70 billion in US asbestos spending went on litigation (hindsight LL1-05). Milward ended in 2016 with no recovery from the remaining defendant (hindsight LL2-24, Claim 3). DBCP claims filed in the 1990s were still being decided in 2026, mostly on limitation and forum rather than merits; the Nicaraguan judgments were never enforced, and the Ninth Circuit disciplined plaintiffs’ lawyers (hindsight LL2-09, Claim 6). Camp Lejeune paid USD 879 million by September 2026 for exposures in 1953–87 (hindsight LL2-24, Claim 6). - Latency against limitation, and backstops. The European Court of Human Rights held in Howald Moor (2014) that a limitation period expiring before mesothelioma could be diagnosed violated access to court (hindsight LL1-05, Claim 9). The UK’s Diffuse Mesothelioma Payment Scheme (2014), funded by an insurer levy, covers untraceable defendants (hindsight LL2-24, Claim 6). The EU’s 2024 Product Liability Directive adds causation presumptions and a 25-year long-stop for latent injury, but keeps the development-risk defence that shields producers where the state of knowledge could not reveal the defect (hindsight LL2-24, LL2-25, LL2-28). - Caps. Fukushima official costs are about ¥21.5 trillion, roughly 100 times the European cap; TEPCO’s liability was uncapped, yet the state still had to take control of it (hindsight LL2-18, Claim 5). The Deepwater cap was actually USD 75 million plus removal costs and fell away for gross negligence; BP paid about USD 67 billion, so Deepwater is a weak test of the cap argument (hindsight LL2-24, Claim 9). - Deterrence. The French Senate (2005) found pooled asbestos funding favoured “la mutualisation du risque” (risk pooling) at the expense of prevention and employer accountability (hindsight LL1-05, Claim 9). The CSRLD has 202 successes in 43 years, and its “success” is self-reported (hindsight LL1-03, Claim 9). No assurance bond for an uncertain hazard was found anywhere; the US EPA declined financial-responsibility rules in 2018 and 2020 (hindsight LL2-25, Claim 8). - When liability bit. US courts held MTBE producers liable on failure-to-warn theories grounded in their own early knowledge (verdicts of USD 423 million, 236 million and 105 million), and Congress refused a safe harbour (hindsight LL1-11, Claim 9). - Exit routes. Beryllium’s producer co-drafted a tenfold-tighter limit only after its legacy liability had largely run off, partly because federal compensation diverted tort claims; this fits the exit-route thesis, but interest alignment explains it equally well (hindsight LL2-06, Claim 8).

Ratings. Strong: compensation is late and partial, decided by procedure (limitation, forum, proof, solvency) more than by science; tables and presumptions need a history of prior victims, so work for mature hazards, not novel ones; caps and insolvency socialise tail risk (moderate as to deterrence effect). Asserted or contradicted: that compensation schemes improve prevention incentives or exposure records. Untested: assurance bonds for uncertain hazards.

Analysis. - Liability is two-edged. It is the reports’ main hope for internalising costs, and also their main documented reason firms deny, delay and fund doubt. - After-the-fact liability works best with documentary evidence of early knowledge and a solvent defendant, exactly what long latency and concealment erode. - Pre-emptive compensation for novel hazards trades one error for another. Tables built in 2013 for mobile-phone brain tumours would probably have compensated a harm the best current evidence suggests does not exist at population level (hindsight LL2-24, Claim 7), moving the cost of false positives onto producers or levy-payers: a legitimate value choice that should be named.

3.12 Victims, late victims and environmental justice: Minamata as the paradigm#

Reports. LL2-05 is the reports’ only full environmental-justice case, framed as a “democratic deficit” (pp. 92, 114). - Economic embeddedness. Chisso’s “castle town” employed 3,811 of 19,819 city workers and paid half the local taxes (p. 96). The trade ministry insisted effluent restrictions “should never be implemented” (p. 99). A single-sourced anecdote has an official contrasting his own earlier action on polio: “Chisso never existed behind the polio outbreak” (p. 99). - Power and hierarchy. Fishermen “lacked political power” (p. 99); “Centre” scientists were trusted over “hick” universities (p. 103); patients were “shunned” (p. 98). - Counting and defining victims. Accreditation is passive, on application, adjudicated by a committee whose members do not examine applicants; one took 25 years (p. 107). A presumptive 1971 standard was reversed by strict 1977 criteria after claims surged (pp. 108–109). From 1978 the prefecture issued debt to fund Chisso’s payments, a conflict between “duty to the patients and their financial situation” (p. 109). Closed expert opinions defending the criteria were later exposed as government positions (pp. 109–110). 2,273 recognised patients stand against “several tens of thousands” unrecognised (p. 94). - Late victims. The congenital disease was missed because experts believed the placenta protected the foetus; a mother saw it first (pp. 105–106). Moderate in-utero effects were “disregarded” for lack of proactive study (pp. 106–107). - Wider claims. Minamata victims waited “almost fifty years” for equitable compensation (LL2-24, p. 581); “justice lags harm by decades” (LL2-28, p. 679).

Other cases with justice features (7): DBCP plantation workers (LL2-09); lead and e-waste in poor countries (LL2-03); DDT spray residents (LL2-11); Love Canal (LL1-12, p. 127; told only from the residents’ side); Greenland mercury (LL2-23, p. 569); climate (LL2-14, p. 309); asbestos in Asia (LL2-A3).

Evidence and hindsight. The Minamata mechanisms now have a 13-year out-of-sample record (hindsight LL2-05): - Recognition is frozen. Kumamoto and Kagoshima had about 2,284 certified patients in 2022, and Kumamoto’s count has not moved since. About 55,000 people received relief off-register, an official measure of the unrecognised population. - The 1977 criteria survived a 2013 Supreme Court ruling that made them non-exclusive (not “invalid”, as the chapter says), with the ministry citing “uniform operation”. - “Final” settlements did not end litigation. About 1,800 new plaintiffs followed the 2009 settlement. Courts split: Osaka 2023 recognised all 128 plaintiffs; Kumamoto 2024 found 25 of 144 had the disease but time-barred them. - The survey has barely started. The statutory residents’ survey, required “promptly” in 2009, had reached a 32-person pilot by 2026, framed by the minister as “dispelling health anxiety”. Process failures continue (the scripted microphone cut-off, 2024). - Balancing evidence: apologies and inflation-linked allowances from April 2026; the diagnostic dispute remains live, and two courts accepted the state’s arguments.

Rating: strong for the Minamata mechanisms: case definitions harden when the definer also pays; relief without recognition produces false closure; passive, claim-based counting hides scale; the state becomes co-financier of the polluter; claimants meet procedural disrespect. Moderate as a generalisation: the reports analyse justice in depth only once, and the authors are protagonists (Harada testified in the Minamata cases). Asserted for “for the most part … irresponsible corporations” (LL2-00, p. 11), which later cases weaken (§4.6).

Analysis. The core insight concerns counting: whoever defines and counts victims, especially if they also pay, controls the apparent size of harm, and settlements that pay without recognition buy quiet, not closure. Justice failures are also procedural (who is heard, examined, believed) and recognitional (who is named as harmed), not only distributive.

3.13 Intergenerational effects and legacy costs#

Reports. Persistence and latency commit costs to the future: - Stocks in use and in the environment. Asbestos harm peaks decades after use, and removal “will cost further billions” (LL1-05, pp. 52, 58). Installed PCBs outlast production bans, with harm to “the unborn” and “future capabilities and those of their offspring” (LL1-06, pp. 66–72). CFC banks keep leaking, CFC-12 stays above 37% of its 2001 level in 2100, and a 60-year skin-cancer latency locks in 14,000 extra cases a year in north-western Europe even with action (LL1-07, p. 77; LL2-23, p. 576). Lead in soils persists for centuries (LL2-03, p. 66; LL2-23, p. 568). - Inherited clean-up. Great Lakes containment “leaves the problem to another generation” (LL1-12, pp. 130–131); TBT sediments burden dredging authorities (LL1-13, p. 141); DBCP persists in groundwater (LL2-09, p. 210); DDT left 213 Superfund sites, with remediation “not as effective as anticipated” (LL2-11, pp. 249–250). - Bodies and resources. DES has second- and possibly third-generation effects (LL1-08, pp. 87–88), and in-utero exposure passes Great Lakes contaminant injury between generations (LL1-12, pp. 130–131); depleted fisheries fall into a rebuilding trap (LL1-02, pp. 22, 26). - Decision rules. Waiting for observed climate harm locks in more (LL2-14, pp. 309, 314, 337); flood policy should avoid “committing future generations to inappropriate options” (LL2-15, p. 362); persistent agents “cannot be recalled” (LL2-26, p. 636); discounting weights future harm less (LL2-23, p. 574).

Cases supporting: more than 13. Cutting against: none on the physical facts. Recovery is sometimes faster than feared: Great Lakes delistings accelerated and MTBE attenuated (hindsight LL1-12, LL1-11).

Evidence and hindsight. EU asbestos-containing materials likely exceed 100 million tonnes, with 4.1–7.3 million workers currently exposed (hindsight LL1-05). A July 2026 USGS study projects California DBCP well exceedances to about 2080 (range 2048–2109) (hindsight LL2-09). TBT sediment half-times run up to 33 years, and the treaty did not settle who pays for dredging (hindsight LL1-13). Antarctic ozone recovers around 2066, and legacy ocean mercury drives Arctic uptake despite falling deposition (hindsight LL2-23). The July 2026 earthquake liquefied reclaimed land holding Minamata’s mercury sludge (hindsight LL2-05). DES third-generation effects remain unresolved after 25 years (hindsight LL1-08).

Rating: strong for the physical mechanism: persistence plus latency commits future harm and cost regardless of later action. Suggestive for intergenerational ethics: beyond LL2-23’s paragraph on discounting and LL2-27’s “across … generations” (p. 649), the reports document obligations to future people but do not analyse them.

Analysis. The legacy is not only contamination but infrastructure that locks in exposure (asbestos diaphragms, water pipes, building stock), institutions that must be sustained for decades (monitoring, containment, compensation), and evidence gaps later generations inherit (unmonitored exposures, closed cohorts). A decision’s time horizon should be set by the agent’s persistence, not the decision-maker’s term of office.

4. Counter-evidence, complications and critiques#

4.1 The evidence base is one-sided by construction. - All 14 LL1 cases were confirmed hazards; the editors found no usable false positives (LL1-00, pp. 12–13). - LL2’s false-positive chapter reviewed 88 alleged cases, mostly from critics’ lists, and confirmed four (LL2-02, p. 25). Its definition excludes market, liability and risk-risk costs, so it cannot register the critics’ main economic concern (critiques.md, §3.2, §5). - LL2-23 weighs costs of action only where they are small (Stern’s 1%, SO2 below EUR 1/kg, Japan) and never asks how accurate cost-of-action estimates have been (notes LL2-23). - LL2-25’s finding that “virtually all” firms found continuation profitable (p. 607) describes a sample chosen for failure; hindsight adds a refiner that declined MTBE and a beryllium producer that co-drafted a tighter limit (hindsight LL2-25, Claim 1).

4.2 The reports document costs of precaution themselves, then underweight them. - Hormones. The EU ban was “in reality, a political risk assessment” with “no good evidence” of health benefit, and led to retaliation authorised at US$116.8 million plus C$11.3 million a year (not the chapter’s EUR 160 million, and a ceiling rather than a measured loss) (LL1-14, pp. 153–154; hindsight LL1-14, Claim 9). It ended in market-access deals, with costs on third-party exporters and consumers. - Swine flu. The 1976 programme cost about USD 124 million plus about USD 100 million from over 4,100 lawsuits, with Guillain-Barré cases and deaths (LL2-02, pp. 28–29). The FDA pause on food irradiation lasted about 20 years (p. 30). - DDT. South Africa’s withdrawal was followed by pyrethroid-resistant reinvasion and outbreaks (LL2-11, p. 243), with no quantified malaria-versus-exposure trade-off (notes LL2-11). LL2-13 asks whether precaution’s price “may be a price too high to pay” (p. 296). - Fukushima. 2,351 disaster-related deaths from evacuation, more than direct deaths: protective measures produced the largest measurable harms, which the chapter did not foresee (hindsight LL2-18). - Germany’s accelerated nuclear phase-out. Jarvis, Deschenes and Jha (2022) put its social cost at €3–8 billion a year, mostly air-pollution mortality from replacement coal and imports (hindsight LL2-02; the “USD 12 billion” in hindsight LL2-25 is the 2019 working-paper figure, superseded by the published abstract, checked via Crossref). LL2-25 hedges only that vivid events “may” bring over-reaction (p. 613). - Transitional costs of bans: growth promoters and neonicotinoids (§3.3; hindsight LL1-09, LL2-16). - LL1-17 concedes “over-precaution can also be expensive” (p. 194) but analyses none of this.

4.3 Precaution’s costs can be regressive. Critics’ strongest economic point is distributive: a ban suited to rich countries may cost lives or livelihoods in poor ones (Sunstein and Goklany on DDT and malaria; Majone on EU aflatoxin standards and African exporters; critiques.md, §§3.1–3.3). LL2 classifies both as “real risks” (LL2-02, pp. 35–36), which answers whether there is a hazard but not who pays for the response. LL2-27’s definition (p. 649) requires the justice lens to apply to precautionary measures too; the chapters rarely do so.

4.4 Cost errors run both ways, including in the reports’ own figures.

Monetised totals travel into policy faster than their caveats, and an advocacy chapter can be rigorous about opponents’ numbers while uncritical about helpful ones (hindsight LL2-20).

4.5 Standpoint and self-citation. Several economic claims come from the chapter authors’ own work or interested sponsors: LL2-23’s lead and nitrate values are the lead author’s own; Cranor was the plaintiffs’ methodology expert in Milward, the case he praises (hindsight LL2-24); the Dutch asbestos estimate was Greenpeace-commissioned; the nuclear full-insurance estimate was commissioned by a renewables body (LL2-18, p. 446); the growth-promoter “net loss” is the annex author’s reanalysis of a non-randomised trial.

4.6 “Irresponsible corporations” is too narrow a villain. Public authorities controlled exposure or information in several cases: the US Department of Energy on beryllium (LL2-06, p. 132); the Japanese state on Minamata, whose liability the Supreme Court found in 2004; public-health DDT programmes; and, after 2013, Flint and Camp Lejeune (hindsight LL2-00, Claim 4). The better-supported generalisation concerns the actor with control over exposure and information, whatever its ownership, and a remedy structure skewed against those harmed.

4.7 Compensation may not deter, and “no-fault” is not “fast”. Beyond the French finding (§3.11), about 80% of compensated VICP claims since 2006 were negotiated settlements, with about 3.5 years to adjudication (hindsight LL2-24). LL2-24’s concession that deterrence feedback is “modest” (p. 603) sits in tension with LL2-28’s incentive claims (pp. 679–680).

4.8 Accounting conventions and protections are politically reversible. US valuation practice moved toward the reports’ preferences in 2023 and reversed in 2025–26 (§3.4); France’s alert commission lasted 2013–2026 (hindsight LL2-24). “Credible numbers win” only where institutions accept monetised health benefits.

4.9 Benefits of the hazardous technologies are rarely weighed. Lesson 6 says to scrutinise claimed benefits as well as risks (LL1-16, p. 176), yet DDT’s malaria control, PCBs’ fire safety (LL1-06), asbestos’s fire protection (LL1-05, second-hand), lead’s octane and the pill’s contraceptive value (LL2-13, which does credit it) go largely unquantified. Only DES, with zero benefit, makes the weighing easy (LL1-08): the reports’ cleanest economic case, and the least informative.

5. Technology-neutral diagnostic questions#

Each question is tied to the patterns above and can be asked of any emerging technology and of those developing, financing and regulating it.

1. Who carries the costs of acting, and who the costs of not acting? Are the latter dispersed, future, foreign or unaware? If so, expect delay regardless of evidence (§3.1; LL1-00, pp. 3–4; LL2-27, p. 659; LL2-17, pp. 413, 421).

2. What does the appraisal leave out (pathways, populations, regions, time periods), and is any omission systematic in one direction? Omitted long-latency pathways have repeatedly turned out to dominate (§3.2; LL2-23, pp. 564, 568, 576; hindsight LL2-23, LL1-10).

3. Who produced the compliance-cost forecasts, what exactly did they cost, and has anyone compared them with outcomes? Does a forecast cost the rule proposed or a worst case? Do substitutes exist within the product class? Does the rule leave room for technical change? (§3.3; LL2-08, p. 187; hindsight LL2-08, LL2-03, LL1-14.)

4. Which valuation conventions (discount rate, time horizon, value of life, which endpoints are monetised, how unquantified effects are entered) drive the result, and who sets them? Does the preferred decision survive the most conservative credible bound? (§3.4; LL2-23, pp. 568, 571, 573–574; hindsight LL2-23, Claims 4, 7, 10.)

5. What evidential standard must be met before action, and who bears the cost of uncertainty while it is unmet? Does “not established” leave the loss with those exposed or those creating the exposure? (§3.5; LL1-17, p. 193; LL2-27, p. 658; LL1-09, p. 96.)

6. Do those exposed consent to, or benefit from, the exposure? If harm is displaced (downstream, abroad, to later users or to the end of the product’s life), who will study it? (§3.6; LL2-04, pp. 80, 83; LL2-09, pp. 207–209; LL1-10, pp. 101–103; hindsight LL2-09, Claim 5.)

7. Does the technology’s price include its expected harms? If not, what capital, skills and supply chains will accumulate around that underpricing, and how costly would reversal be in twenty years? (§3.8; LL1-16, pp. 176–177; LL2-27, p. 659; hindsight LL2-27, Claim 7.)

8. If harm appears after decades, will the responsible party still exist and be solvent, and is money set aside now? Do liability caps, limitation periods or state backstops shift tail risk to the public? (§3.11; LL2-24, pp. 596, 600–603; LL2-18, pp. 445–446; LL2-25, p. 612; hindsight LL2-25, LL2-18.)

9. Does liability exposure give the developer a reason to avoid learning about, or admitting, harm? Could it change course without ruinous admission? (§3.11; LL1-06, p. 65; LL2-06, pp. 137, 149–150; hindsight LL2-06, Claim 8.)

10. Who will define and count victims, and does that body also pay? Is counting active (surveys, registries) or passive (applications)? Are the exposure records later claimants will need being kept? (§3.12; LL2-05, pp. 107–110; LL1-05, pp. 60–61; hindsight LL2-05.)

11. Where along the causal chain will control be applied, and does that put the costs on the producer, on intermediaries (utilities, ports, health systems) or on the public? (§3.10; LL2-13, pp. 290–291, 296; LL2-20, p. 492; hindsight LL2-13, LL2-25.)

12. How persistent is the technology’s footprint (materials, infrastructure, contamination, institutional commitments), and is the decision horizon as long? Which options does deployment close for later generations? (§3.13; LL1-12, p. 130; LL2-15, p. 362; LL2-26, p. 636; hindsight LL2-09, LL1-13.)

13. What would the precautionary response itself cost, and who would bear it (countervailing risks, forgone benefits for the least advantaged, third parties)? Is there a graduated or reversible option that lowers the cost of being wrong? (§4.2–4.3; LL2-02, pp. 25, 28–29; LL1-14, pp. 153–154; LL2-11, pp. 243, 250; critiques.md, §3.)

14. What are the claimed benefits, who receives them, and have they been tested as rigorously as the risks? Unproven benefit justifies no risk; benefit flowing only to producers justifies little uncertainty (§4.9; LL1-08, pp. 86, 90; LL1-11, p. 111; LL1-14, pp. 153–154; LL2-A3, pp. 732–733).

15. Which comparable actors have chosen differently, and what happened to them? Firm and jurisdiction heterogeneity is the best check on claims that action is unaffordable or inaction inevitable (hindsight LL2-25, Claim 1; LL1-09, pp. 95–96; LL2-12, pp. 271–273).