T02: Early warnings and why they were missed or ignored#

Cross-cutting thematic synthesis, Strand A. Written 26 September 2026.

Sources. LL1 = EEA, Late lessons from early warnings: the precautionary principle 1896–2000 (2001); LL2 = EEA, Late lessons from early warnings: science, precaution, innovation (2013). Built from all 47 digests, the audited notes (LL1-16 and LL2-A2 directly; the rest searched by term), the hindsight files (to September 2026) and the external/ files. Quotations were checked against working/text/chunks/. Pages are report pages.

How to read this document#

Each pattern has three registers: Reports (what the chapters say or show); Evidence and hindsight (what the record, notes and post-publication checks show, including where they cut against the reports); Analysis (my inference).

Strength ratings. Strong: several cases, primary or contemporaneous records, consistent with later independent evidence. Moderate: several cases with gaps (protagonist sourcing, hindsight risk, mixed later record). Suggestive: one or two cases, or thin. Asserted: no supporting evidence. Case counts illustrate a curated set, not frequencies.

Three caveats apply throughout (see §8): 1. Selection on outcome. LL1’s 14 cases and LL2’s nine Part A cases were chosen because the harm is now known (LL1-00, pp. 11–12; LL2-A2, p. 701; LL2-00, pp. 9–10). They show how warnings were mishandled, not how often comparable warnings were right. 2. Insider authorship. Most case authors took part in the histories they wrote (LL1-00, p. 12; LL2-00, pp. 9–10); several were protagonists whose own studies are key evidence (Infante, LL1-04; Needleman, LL2-03; Michaels, LL2-06; Hardell, LL2-21). The synthesis chapters are by the project’s editors (LL1-16; LL2-27; LL2-28). 3. Hindsight. The reports ask to be judged by “the spirit of the times” (LL1-00, p. 11; LL2-A2, p. 701), yet some dated “first warnings” read as warnings only once the outcome is known.

Summary of findings#

• Sources. Warnings came early, from the edges of the expert system (inspectors, workers, fishers, beekeepers, a mother, wildlife, dissenting scientists) and just as often from inside producing firms, often by chance. Strong for the cases; moderate as a generalisation.
• How warnings were lost. Some never reached those able to act; others arrived and were discounted, contested or trimmed. Litigation documents later strengthened the “discounted” strand; for BSE the official inquiry found over-reassurance, not producer bias. Strong that both occur; frequency unverified.
• Lags. Table A2.1’s “years of substantial inaction” run from 5–30 to over 100, median about 40 (LL2-A2, p. 702), but its dating rules are inconsistent. Counted from the first specific warning, lags shrink to months for vivid, attributable harms and stay at decades for chronic, diffuse ones. “Effective action” is usually a decades-long sequence. Strong as description; weak as a measure.
• Standard and burden of proof. In at least 12 cases the evidence demanded governed timing. Universal demands (“all fish”, “sole cause”, proof “that will satisfy everyone”) defaulted to inaction; lower standards brought fast action. The durable lesson: the evidential threshold decides who bears the cost of error. Strong.
• False negatives and positives. Absence of evidence from an unperformed or underpowered search is not evidence of absence, and definitions decide how many errors of each kind are counted. The “4 of 88” rate and the claim that false positives are short-lived both fail. Erring towards protection is best justified when harm is persistent, latent or irreversible and the benefit forgone is modest. Strong for the logic; contested for frequency.
• Status quo bias was institutional: grandfathering, provisional numbers that hardened, lapsed conditions, feasibility-based limits, research in place of action, dual mandates, credibility traps. Strong for mechanisms; moderate as a general tendency.
• Knowing is not acting. Accepted knowledge often failed to produce action because costs were concentrated, harm was “elsewhere”, or rules went unenforced. That political will outranks information was never tested; later evidence suggests will follows harm documented in a jurisdiction’s own population. Strong as description; moderate as explanation.
• Counter-evidence. The reports’ own forward warnings have a mixed record, swine flu shows an over-weighted warning, genuine ignorance limited foresight, and acting on warnings created new risks.

1. What counted as a warning, and where warnings came from#

1.1 Kinds of warning#

Reports. Neither report defines “early warning”. LL1 asked authors for “the first credible scientific warning” (LL1-00, p. 11); Table A2.1 dates very different kinds of signal (LL2-A2, p. 702).

Evidence (LL2-A2 notes; case digests):

Strength: strong (the reports’ own dating).

Analysis. The type decides what “ignoring” means. Leaving observed harm unaddressed for decades (asbestosis after 1930, benzene leukaemia after 1928, DES after 1953) is a failure of response; missing a related-substance or property signal is at most a failure of foresight, and only if someone was charged with foresight. Mixing the two produces the century-long headline lags (§3). Trust in the data matters too: Callendar’s CO₂ work was disbelieved partly because colleagues distrusted his measurements (LL2-14, p. 311).

1.2 Who raised warnings#

Reports. Warnings often came from outside the research establishment (LL1-16, pp. 177–178; LL2-28, p. 675). Individuals, not the ICRP, curbed radiation misuse (LL1-03, p. 34). Economically harmed parties (inshore fishers, oyster growers, beekeepers) pressed warnings, though such cases “seem to be a minority” (LL2-25, p. 609).

Evidence (across both reports): - Front-line officials and affected people (at least 8 cases): women factory inspectors (LL1-05, p. 53); DBCP workers comparing notes at lunch, and their wives (LL2-09, pp. 204, 211); a Minamata mother who recognised congenital poisoning before the experts, Harada among them (LL2-05, pp. 105–106; LL2-28, p. 675); inshore fishers behind the Keats Report (LL1-02, p. 21); beekeepers (LL2-16, pp. 372–373); Love Canal residents (LL1-12, p. 127); Swedish farmers (LL1-09, p. 95); slaughterhouse workers “apparently better informed” than officials (LL1-16, p. 178). - Dissenting or independent scientists (at least 10): Patterson and Needleman (LL2-03, pp. 57–61); Stewart (LL1-03, p. 34); Odén (LL1-10, p. 102); Molina, Rowland and Farman (LL1-07, pp. 79–80); Jensen (LL1-06, p. 64); the Boston DES clinicians (LL1-08, p. 84); Swann (LL1-09, p. 94); academic BPA researchers (LL2-10, pp. 219–221); Callendar (LL2-14, p. 311); the Hardell group, whose warning was later largely not borne out (LL2-21). - Inside the producing industry, or commissioned by it (at least 8): Dow’s toxicologist and a Union Carbide consultant on vinyl chloride, 1959 (LL2-08, pp. 182–183); the consultant Hine on DBCP, 1958 (LL2-09, pp. 204–205); the American Petroleum Institute calling zero “the only absolutely safe level” of benzene, 1948 (LL1-04, p. 39); Drinker’s Halowax-commissioned 1937 study, presented to Monsanto and General Electric (LL1-06, p. 64); Chisso’s doctor’s cat experiment (LL2-05, p. 101); a Japanese beryllium producer’s below-limit cases and the limit co-author’s private 1989 admission (LL2-06, pp. 134, 136); Ethyl’s paid consultant, who voiced health concerns privately a week before publicly backing leaded petrol (LL2-03, pp. 52–53); Doll’s company-commissioned 1955 asbestos study, which directors tried to suppress (LL1-05, p. 54). - Chance, or monitoring built for another purpose (at least 7): PCBs found during DDT analysis (LL1-06, p. 64); BPA traced to autoclaved labware (LL2-10, p. 217); Irgarol found in a herbicide survey (LL2-12, p. 267); the DES cancer cluster (LL1-08, p. 86); intersex roach on a routine check (LL2-13, p. 282); acid rain via a plant-nutrient network (LL1-10, p. 102); Antarctic ozone loss, “essentially serendipitous” to the editors (LL1-16, p. 172) but from “systematic long-term measurements” to its discoverer (LL1-07, p. 82).

Strength. Strong for these cases; moderate as a generalisation, since the reports cannot show how many peripheral warnings proved wrong, and LL1-16 itself warns of lay fallacies such as the “pensioners’ party” (p. 178).

Analysis. In at least seven cases (vinyl chloride, DBCP, PCBs, beryllium, Minamata, benzene, lead) the best-informed warner sat inside, or was paid by, the producing industry: the gap was disclosure, not detection (§2.5). Chance detection recurs so often that the supply of warnings depended on monitoring nobody was charged with.

1.3 Measurement and monitoring decide whether a warning can exist#

Reports. LL1’s lesson 2 is to monitor and research for early warnings (LL1-16, pp. 171–173); the annex treats monitoring as a precondition for any warning (LL2-A2, pp. 704, 707, 713–714). The radiation chapter’s one recommendation is to fund long-term databases “even when an immediate need is not perceived” (LL1-03, p. 36).

Evidence. - Analytical limits: imposex seen in 1970–71 stayed unexplained for a decade (LL1-13, p. 136); chronic beryllium disease was masked until a blood test existed (LL2-06, pp. 135, 139–140). - Detection limits as “no evidence”: Bayer’s studies could not detect residues below 10 ppb, and the ministry told public researchers not to go below that level (LL2-16, p. 373). - Monitoring never set up, or stopped: antimicrobial resistance after Swann (LL1-09, p. 94); symptomless BSE cattle (LL1-16, p. 172); MTBE in EU groundwater (LL1-11, p. 112); men exposed to DBCP (LL2-09, p. 205); UK child blood lead after 1987 (LL2-03, p. 68). - Monitoring without power: a 50% decline over 15 years would go undetected in 72–90% of whale stocks (LL2-26, p. 634). - Assuming harm would announce itself: in 1925 Ethyl’s paid consultant held that 27 months of use “should have sufficed” to reveal poisonings (LL2-03, p. 52).

Hindsight. EU active testing from 2001 found about 7,000 BSE cases in about 50 million cattle and showed that clinical surveillance had “a poor capacity to detect cases” (LL1-16 hindsight). Atmospheric monitoring caught unreported CFC-11 production (LL1-17 hindsight).

Strength: strong. More than 10 cases, sound statistical logic, later confirmation. Cuts against: more research can widen uncertainty (Great Lakes; LL1-16, pp. 172–173), and monitoring not tied to action “can easily become an essentially scientific or academic pursuit” (LL2-12, p. 274).

1.4 The first recognised harm is rarely the decisive one#

Reports. Acute effects preceded chronic ones in six LL1 cases, while BSE and CFCs gave no acute warning at all (LL1-16, p. 171). Harms “expand” to new effects and lower doses (LL2-28, p. 672; LL2-A3, p. 717).

Evidence. The 1925 radiation tolerance dose was about 700 mSv a year, against 20 mSv now (LL1-03, p. 33). The 1931 UK asbestos rules covered asbestosis in manufacturing only (LL1-05, pp. 54, 56). The 1948 beryllium limit eliminated acute disease and bred false confidence about chronic disease (LL2-06, pp. 133–134). In 1925 Emerson called deaths an unsound indicator of lead hazard (LL2-03, p. 52). Hindsight supports harm expansion for lead, asbestos, PCBs and radiation (LL2-28 hindsight), partly as a selection effect (LL2-A3 hindsight).

Strength. Strong for at least six cases; moderate as a general expectation.

Analysis. Controlling the visible harm created a sense that the hazard was handled. Asbestos claims that past disease belonged to superseded conditions go back to 1906; after each improvement, the risks of the new conditions took further decades to show (LL1-16, p. 173).

2. How warnings were treated#

2.1 Not delivered, or delivered and discounted#

Reports. In many cases information “was either not brought to the attention of the appropriate decision-makers early enough, or was discounted for one reason or another” (LL1-16, p. 168). Some “loud and late” warnings (asbestos, PCBs, Great Lakes, SO₂) were ignored “because of short-term economic and political interactions” (p. 168).

Evidence. - Not delivered. MAFF told the Department of Health about BSE some 17 months after it was alerted, and only because health’s assent was needed to keep clinically affected cattle in the food chain (LL1-16, p. 180; LL1-15, pp. 159–160). Drug regulators missed published reports of hormones in water (LL2-13, p. 284). DBCP makers’ knowledge did not reach user firms (LL2-09, p. 211). The designers of a solvent-leaching pipe lining did not consult toxicologists (LL2-04, pp. 82, 84, 88). PCB evidence from 1937 “did not gain the wider attention of policy-makers” (LL1-06, p. 64). - Delivered and discounted: the 1953 DES trial (LL1-08, p. 86); the Keats and Alverson fisheries reviews (LL1-02, p. 21); the fisheries department’s own 1988 call to halve the cod quota (LL2-17, p. 413); SEAC’s May 1990 advice that “no risk” could not be stated categorically (LL1-15, p. 161); a 1990 MTBE warning the Danish EPA dismissed because petrol components were “rarely found in groundwater at the time”, though nobody monitored for MTBE (LL1-11, p. 114); Swann’s recommendations, “gradually diluted” (LL1-09, p. 94); a 2001 paper on a roughly 1,000-year tsunami recurrence on the Sendai coast that never reached plant design bases (LL2-18, p. 438).

Hindsight. Litigation documents strengthened the discounting strand: PFAS producers knew of toxicity by 1970, “forty years before the public health community” (LL1-16 hindsight); MTBE’s groundwater threat, flagged by an Exxon engineer (1984–85), a state paper (1986) and the US EPA (1988), was “considered and not acted on” (LL1-11 hindsight); TEPCO’s own 2008 study estimated a possible 15.7 m tsunami (secondary source; LL2-18 hindsight). Against it, Phillips found no producer bias in MAFF’s BSE policy, and health officials “as eager as MAFF” to avoid alarm (LL1-15 hindsight).

Strength. Strong that both modes occur (at least 12 cases). LL1-16’s claim that non-delivery marks “most” cases (p. 171) is unverified (LL1-00 hindsight).

Analysis. Non-delivery calls for duties to share information; discounting calls for independence, disclosure and explicit standards of proof. The reports blur the two: LL1-16 files MTBE under “institutional” ignorance although the information “was available” and was “essentially disregarded” (p. 174).

2.2 How warnings were contested#

Reports. A recurring repertoire: calls for more research, alternative explanations, replication demanded of inconvenient findings, “sound science” and sometimes deliberate doubt (LL1-16, p. 179; LL2-25, p. 610; LL2-27, pp. 645–646).

Evidence. - More research instead of action. Monsanto publicly called for “extensive research, on a worldwide basis” while privately acknowledging contamination (LL1-06, p. 65). Politicians “asked for more evidence” on cod (LL2-17, p. 413). The beryllium producer sought more research when it was “beyond doubt” that its limit failed (LL2-25, p. 610). Twenty years on, “many governments still call for more information as a substitute for action” on nanomaterials (LL2-22, p. 547). Swann’s counter-rule: “the cry for more research should not be allowed to hold up our recommendations” (LL1-16, p. 181). - Alternative causes and shifting reassurance. Chisso, the trade ministry and a sponsored committee promoted rival theories (LL2-05, pp. 102–103); benzene leukaemias were called a random cluster (LL1-04, pp. 42–45). Growth-promoter reassurance moved from resistance confined to one clone, to no transfer from gram-positive bacteria, to the claim that low doses cannot select resistance, each overturned (LL1-09, pp. 94–95). The beryllium defence moved from blaming every case on overexposure to saying not enough was known (LL2-06, pp. 137–138). - Replication demanded of the inconvenient finding. Stewart’s finding was “disbelieved” until repeated, while the practice continued (LL1-03, p. 34). - Deliberate doubt. Tobacco research was the “antidote”, to “keep the controversy alive” (LL2-07, p. 154). A vinyl chloride report was accepted only once it said the “etiologic agent of the disease is unknown”, and a producer then refused further research so industry could “act as if the cause was ‘unknown’” (LL2-08, p. 184). - Unanswerable questions. A French committee asked whether Gaucho was “solely responsible, at national level, for all” losses, “a question that had never been asked” (LL2-16, p. 379).

Hindsight. Documentary evidence of doubt-making grew: US tobacco judgments (2006, 2009), cross-sector syntheses, the 2025 retraction of a ghostwritten glyphosate review (LL2-07, LL2-26, LL2-27 hindsight). It comes largely from one research tradition, and the “manufactured doubt” label is disputed in specific cases (LL2-27 hindsight). Industry-linked science also produced protection: a 2012 producer–union draft shaped OSHA’s 2017 beryllium rule (LL2-06 hindsight).

Strength. Strong that the repertoire recurred (at least 10 cases, internal documents for tobacco, vinyl chloride, beryllium, PCBs). Moderate on intent where there are no documents; purposeful doubt over PCE is asserted (LL2-04, p. 88). Suggestive as a predictor.

2.3 How warners were treated#

Reports. “Shooting the messenger” rarely, “if ever, promotes societal welfare” (LL1-16, p. 179). Early warners face retaliation and are vindicated late (LL2-03, pp. 70–71), so protection should turn on “objectively reasonable belief”, not on being proved right (LL2-24, pp. 582–584).

Evidence. - Replaced or labelled unscientific. California replaced its fishery experts, and reports moved from an “imperative” cut (1938) to “no reason to be concerned” (1942) (LL1-02, p. 20; single secondary source). The Keats Report was called “biased pseudoscience” (LL1-02, p. 21); inshore data were excluded as “messy and often anecdotal” (LL2-17, p. 414); Minamata’s lay and sentinel signals were dismissed (LL2-05, pp. 96–97, 105). A cod cut was rejected because “the scientists had been wrong before” (LL2-17, p. 413). - Pressure. Patterson was attacked professionally (LL2-03, pp. 57–58; partly his own account). Bayer sued three beekeeper leaders and lost all three (LL2-16, p. 380). Dole’s suit against a documentary was ruled a SLAPP (LL2-09, p. 208). - Containment. A 1937 plea to avoid “mob hysteria” among workers (LL1-06, p. 64); a vinyl chloride producers’ secrecy agreement (LL2-08, pp. 184–186); a contract that kept a national effluent survey unpublished until 1994 (LL2-13, p. 284); Chisso’s suppressed cat experiment (LL2-05, p. 101). - Trimming inside advisory processes. The Southwood committee judged a ban on cattle brains justifiable but not “politically feasible” (LL1-16, p. 179); a member’s call to keep seeking alternatives to leaded petrol was omitted from the 1925 report (LL2-03, p. 53); a SEAC draft had its “most potentially inflammatory” wording edited out (LL1-15, p. 161).

Hindsight. At Flint, regulators met independent testers and a paediatrician with “aggressive dismissal, belittlement, and attempts to discredit” (task force, 2016; LL2-03 hindsight). Legal protection advanced mainly for reporting breaches of law, not warnings about lawful products; France’s alert commission was abolished in 2026 (LL2-24 hindsight).

Strength: moderate. At least 10 cases, but most accounts are the warners’ own, and the warners were selected because they were vindicated (LL2-24 digest). Strong where documents exist (secrecy agreement, contract, suppressed study).

Analysis. The costliest treatment was probably the quiet trimming or non-disclosure of warnings produced inside firms and advisory bodies, more than the persecution of outsiders. Whistleblower regimes reach only part of this.

2.4 Reassurance that forecloses later action#

Reports. Once beef had been declared safe, any further step, however cheap, implied the reassurance had been false, so measures were refused for presentational reasons (LL1-15, pp. 161–162). Reassurance cited the absence of evidence “when no evidence was actually being sought” (LL1-16, p. 172). Both prefaces blame misplaced “certainty” about the absence of harm (LL1-00, p. 4; LL2-00, p. 6).

Evidence. In June 1990 the minister called beef “perfectly safe”, a month after SEAC’s contrary advice (LL1-15, p. 161); of a cheap measure, “It was agreed not to raise it” (p. 162); about 48% of abattoirs visited in 1995 failed the offal rules (p. 162). Parallels: “No peril to public” on leaded petrol in 1924 (LL2-03, p. 51); TBT’s partial recovery cited as having “solved” the problem (LL1-13, p. 138); a nuclear “language of certainty” (LL2-18, p. 448).

Hindsight. Phillips found the BSE approach’s “object was sedation”, that it “did not set out to deceive”, and that the “campaign of reassurance was a mistake” (LL1-15 hindsight). The IAEA (2015) found Fukushima’s safety assumption “was not challenged by regulators or by the Government” (LL2-18 hindsight).

Strength. Strong for BSE (contemporaneous minutes, official inquiry); moderate across at least five cases. Best framed as over-reassurance, not deception.

2.5 Producers often knew first#

Reports. Part A shows “slow and sometimes obstructive behaviour by businesses” (LL2-00, p. 10); “in virtually all reviewed cases it was perceived to be profitable for industries to continue” (LL2-25, p. 607).

Evidence. At least seven cases rest on internal records: vinyl chloride (warnings from 1959, bone disease shared only among firms in 1966, cancer results withheld from NIOSH in 1973; LL2-08, pp. 182–186); DBCP (a Shell representative called protective advice “impractical”; LL2-09, p. 205); beryllium (the limit was “fundamental to our product liability defense”; LL2-06, p. 137); PCBs (Monsanto’s 1969 plan, via one secondary source; LL1-06, p. 65; hindsight confirms the public–private split from primary documents but finds the “profits to cease” line a journalist’s paraphrase); tobacco (LL2-07); Chisso (LL2-05, pp. 100–102); the API on benzene (LL1-04, p. 39). Hindsight adds PFAS, sugar and oil-company climate research (LL2-27 hindsight). Litigation discovery is the main window, which creates observability bias (LL2-07 digest).

Cuts against. PCE showed indifference, not concealment: “nobody made them care” (LL2-04, p. 86). DES motives are asserted without documents (LL1-08 digest). Firms differed: a refiner declined MTBE, and the beryllium producer co-drafted a stricter limit (LL2-25 hindsight).

Strength. Strong for the documented cases; moderate as a general pattern.

3. The lag between warning and action#

3.1 The reports’ numbers and their limits#

Reports. LL1-16 finds gaps of “many years or decades, and in some cases over a century” (p. 168). Table A2.1 (LL2-A2, p. 702, after Gee 2009):

The median, excluding fisheries, is about 40 years (LL2-A2 notes). Other chapters add: Minamata, 12 years from identifying the cause to government acceptance (LL2-05, p. 105); synthetic oestrogen (EE2) in rivers, about 25 years (LL2-13, pp. 282, 290); invasive species, about 20 years (LL2-20, p. 501); tobacco, over 40 years (LL2-27, p. 647); nanomaterials, a lag that “will not be less than two decades”, though the chapter’s sources say “one or two decades” (LL2-22, p. 549).

Evidence and hindsight: not a consistent measure (LL2-A2 notes and hindsight). Start dates mix direct observations with generic signals; end dates mix bans, voluntary withdrawals, admittedly ineffective management and the compilation date; the PCB and Great Lakes figures do not follow from the table’s own entries; recent cases are truncated. Counted from the first specific, credible warning (the notes’ reading): benzene 10 years from the 1977 cohort study (59 from the 1928 leukaemia report); asbestos 35–39 from mesothelioma (68 from 1930); PCBs 3–10 from the 1966–69 wildlife evidence; ozone 3 years to the US aerosol ban, 13 to Montreal; DES 18 years from the no-benefit trial but 7 months from the cancer link; growth promoters 16 years to Sweden’s ban, 29–30 to the EU’s; MTBE 9–10 from 1990; TBT 0–5 nationally, about 25 globally; BSE 2–3 to partial measures, 8 to the total ban.

“Effective action” also kept arriving long after the table’s dates: - Asbestos: the EU cut its limit 10–50-fold in 2023 (LL2-A3 hindsight); the US banned chrysotile in 2024, 126 years after the first British warning, with litigation pending in 2026 (LL1-00 hindsight). - Benzene: the EU limit reached 0.2 ppm in April 2026; the US limit has stayed at 1 ppm since 1987 (LL1-04 hindsight). - PCBs: well under half of known stocks had been disposed of by 2020, against Stockholm deadlines of 2025 and 2028 (LL2-A2 hindsight). - EE2: an EU standard was adopted in 2026, with a 2039 target (LL2-13 hindsight). - Leaded aviation fuel: exempted as “temporary” in 1996, still permitted, with a US goal of ending it by 2030 and EU authorisations to 2032 (LL2-03 hindsight).

Strength. Strong as description: lags of decades were common in these 23 cases, and hindsight lengthens several. Weak to moderate for the specific figures (orders of magnitude only). Not a base rate for warnings in general.

Analysis. Keep three intervals apart: the lag in knowing (until a specific, attributable warning exists), the lag in responding, and the lag from the first binding measure to a measurable fall in harm. Counting from specific warnings removes the century-scale figures, but chronic, diffuse harms still take decades, and latent or persistent agents add more: British male mesothelioma deaths peaked only in 2016 (LL1-05 hindsight). Neither report cites Collingridge’s dilemma, but LL2 names the same bind: by the time harm is confirmed, sunk investment and the claim that “today’s technology is now safe” resist action (LL2-28, p. 672; external/context.md §2.3).

3.2 When response was fast#

Reports. “just four to seven cases of very rare cancers or sperm reduction” were enough for prompt action on DES, vinyl chloride and DBCP (LL2-27, p. 645). Once a sentinel finding appears, action can take months (LL1-09, p. 96).

Evidence. At least ten fast responses: DES, FDA action within about seven months of the cancer link (LL1-08, pp. 84, 86); vinyl chloride, OSHA to 50 then 1 ppm within months (LL2-08, pp. 186–187); DBCP, an emergency standard in about two months (LL2-09, p. 206); the Danish avoparcin ban, months after resistance findings (LL1-09, p. 96); TBT, France in 1982 on the “best information available” and the UK within about a year of a key 1986 study (LL1-13, pp. 136, 138); CFCs, the US aerosol ban three years after the theory, before observed harm (LL1-07, p. 80); DDT, Swedish bans a decade before the eagle link was “confirmed” (LL2-11, p. 242, the notes’ reading of Fig. 11.1); Love Canal relocation (LL1-12, p. 127); Caulerpa, eradicated in California 17 days after detection where France’s delay after 1984 let it spread (LL2-20, p. 498); Snow’s pump handle (LL1-00, pp. 14–15).

What they share: a legible endpoint; an affected group with a voice; independent public expertise; a concentrated industry or cheap fix (LL2-08, pp. 187, 189); low commercial stakes (sunflower seed-dressing suspended in 1999, maize only in 2004; LL2-16, p. 382); harm to something with market value (a pest snail’s loss would have prompted “little if any action”, the oyster collapse did; LL1-13, p. 136).

Cuts against. The 1952 London smog drew only “modest remedies” (LL2-A2, p. 711); Minamata’s identified route and dramatic suffering still met 12 years of inaction (LL2-05); Europe lagged the US by 3–6 years on DES after the cancer signal (LL1-08, p. 89); the US aerosol ban needed no vivid harm.

Strength: moderate. Consistent but confounded: vinyl chloride, DBCP and DES were unusually easy cases.

Analysis. Systems reacted faster to vivid harm than to absent benefit. DES was contraindicated within months of the cancer finding, after 18 years of evidence that it did not work; the FDA’s 1971 notice shows efficacy review still rated it “possibly effective” for preventing pregnancy loss (LL1-08 hindsight, 36 Fed. Reg. 21537). Strong for DES; suggestive in general.

4. Burden and standard of proof#

4.1 Demands for proof as the instrument of delay#

Reports. Choosing the level of proof “can radically shift the size, nature and distribution of the costs of being wrong” and “is a key political decision with profound ethical implications” (LL1-17, p. 193). LL2-27’s fourth barrier is the tension between “the high strength of evidence needed for scientific causality and the lower strength of evidence needed for timely public policy” (p. 645). For PCBs, “at almost every stage government action was taken only when there was a high level of scientific proof” (LL1-06, p. 72).

Evidence. Standards demanded, in at least 12 cases: - Fisheries: no restriction unless experiment had “decisively settled” the matter (1883–85; LL1-02, p. 18); proof “must be overwhelming” (1919; p. 25); no evidence of depletion “that will satisfy everyone” (sardines; p. 20); “no evidence that fishing on spawning grounds in any way harmed the stock” (cod; LL2-17, p. 411). - Asbestos, 1912 and 1917: “insufficient evidence to justify further action” (LL1-05, p. 54). - Minamata, 1957: the health ministry wanted “clear evidence that all fish and all shellfish are poisoned”, although Shizuoka had used the same law in 1950 on comparable evidence (LL2-05, pp. 98–99). - Benzene: the 1980 US Supreme Court required a finding of “significant risk” (LL1-04, p. 40), placing, in a dissent’s words, “the burden of medical uncertainty squarely on the shoulders of the American worker” (LL2-08, p. 187). - Great Lakes: causal proof before “massive” spending (LL1-12, p. 128). Nitrate: an EU guide value dropped in 1998 for “absence of scientific proof” (LL2-23, p. 570). Neonicotinoids: the sole-cause framing (LL2-16, p. 379). - Lead, 1925: “because some animals die and some do not die in some experiments, shall we give this thing up entirely?” (Ethyl’s president; LL2-03, p. 53). - CFCs: Du Pont would stop only on “reputable evidence” of harm, and denied any existed until 1986 (LL1-07, p. 80). - Growth promoters: proof needed research along a 17-step causal chain, at least 5–10 years per gene and substance, so inaction “is not a neutral position” (LL1-09, pp. 95–96).

Contrast: lower standards brought action. US aerosols on a “reasonable expectation” of harm (LL2-A2, p. 708); Sweden’s growth-promoter ban over “uncertain” long-term effects (p. 710); the French TBT ban (LL1-13, p. 136); the DOE’s 1999 beryllium trigger (LL2-06, p. 138); the vinyl chloride rule, upheld “on the frontiers of scientific knowledge” because “the factual finger” pointed to a low limit (LL2-08, p. 187).

Hindsight: thresholds were later treated as choices, in both directions. Pfizer v Council (2002): scientific committees have “neither democratic legitimacy nor political responsibilities” (LL1-17 hindsight); the CJEU (2021): protective measures need not wait until risks “become fully apparent” (LL1-16 hindsight); the EU’s 2023 hazard classes for persistent, mobile and endocrine-disrupting substances. The other way: US Executive Order 14303 (2025) confines “overly precautionary assumptions and scenarios” to where law requires them (LL2-27 hindsight). Evidence-admission rules act as standards too: BPA guidance values since 2006 span about 250,000-fold on largely shared evidence (LL2-10 hindsight; LL2-10, pp. 221–223).

Strength. Strong that the standard governed timing (at least 12 cases; primary records for the 1957 reply, the 1980 ruling and the 1883–85 inquiry). Moderate that lower standards would have produced better outcomes: counterfactuals are untested, and the cod one is “unproven” (LL1-02 hindsight).

Analysis. Three features recur. The standard was often universal (“all fish”, “all losses”, “everyone satisfied”), which nothing in a multicausal system can meet, so it defaults to inaction. It rose with the cost of the remedy (Great Lakes, BSE, cod; suggestive). And it was asymmetric: high proof for harm, little for safety (LL2-28, p. 675). DBCP’s safety rested on “authoritative assertion but without evidence” (LL2-09, p. 211); a UK health committee found enough “to discount any possibility of risk” from oestrogenic drinking water on “this small set of studies”, while action for fish took decades (LL2-13, pp. 283–284).

4.2 Who carries the burden#

Reports. Sweden’s 1973 chemicals law demanded safety “beyond all reasonable doubt” from manufacturers but only a “scientific suspicion of risk” from regulators (LL1-16, Table 16.1, p. 184; LL2-27, Table 27.2, p. 658). Under TSCA, the EPA had to show risk before demanding the data needed to show it, “a classic regulatory paradox” (LL2-22, p. 537). Tort rules “asymmetrically hamper plaintiffs” (LL2-24, p. 588). Beekeepers funded both the research and the litigation (LL2-16, p. 383). Grandfathering spared MTBE new-substance scrutiny (LL1-11, p. 116).

Hindsight. REACH put a duty of proof on manufacturers; Blaise (2019) told authorities “not to give in all cases preponderant weight to the studies provided by the applicant” (LL1-16 hindsight). Tort stayed slow: Milward ran 2007–2016 and ended in a plaintiff loss; Camp Lejeune compensation came about 35 years after exposure (LL2-24 hindsight).

Strength. Strong as a structural point; moderate as a cause in individual cases.

4.3 The economics of proof#

Reports. Policymakers often act only once “costs of inaction” have been estimated (LL2-23, p. 564). “The costs of preventive actions are usually tangible, clearly allocated and often short term”, while the costs of failing to act are diffuse and longer term (LL1-00, p. 3; LL2-27, p. 647). The Rio principle’s “cost effective” wording is precaution’s “Achilles heel” (LL2-13, p. 296).

Evidence. EE2 treatment by activated carbon was put at EUR 32–37bn for England and Wales, sand filtration two-thirds cheaper (LL2-13, pp. 291–293). Industry cost forecasts exceeded outcomes: vinyl chloride, up to USD 90bn forecast against an unsourced USD 278m (LL2-08, pp. 187, 192); lead, “one million barrels” of oil a day claimed against the EPA’s 30,000 (LL2-03, p. 60).

Hindsight. Ex ante compliance-cost estimates usually exceed ex post costs, though this “does not necessarily demonstrate” bias (LL2-27 hindsight). Whoever controls the accounting rules decides what counts: the US withdrew carbon-cost valuation in 2025 (LL2-23 hindsight).

Strength. Suggestive that monetisation acts as a gatekeeper (LL2-23’s “often” is unevidenced); moderate for concentrated versus diffuse costs (many cases, never measured).

5. False negatives and false positives#

Reports. - LL1: all 14 cases are false negatives, and industry offered no usable false positives (LL1-00, pp. 12–13); false positives are “the smaller but commonly feared risk” (p. 16); science’s bias against false positives “inevitably involves generating false negatives” (LL1-16, p. 184). - LL2 Ch. 2: of 88 alleged false positives, four are genuine: swine-flu immunisation (1976), saccharin labelling, Southern corn leaf blight and food irradiation (LL2-02, pp. 17–25). False positives are “few and far between as compared to false negatives” (p. 35); agencies lean to inaction through blame-avoidance, judicial review, cost–benefit hurdles and “ossification” (p. 34); both errors deserve “equal weight” (p. 34). - LL2 Ch. 26: ten design features push towards false negatives, three towards false positives (Table 26.4, p. 635); inferring absence of harm from absence of proof is “the greatest error” (p. 631); the upper confidence limit is “a plausible worst case” (p. 633). - LL2 Ch. 28: under irreversibility, tip policy “towards avoiding harm, even at the cost of more false alarms” (p. 673), restating 4-of-88 without Ch. 2’s caveats.

Evidence and hindsight (external/critiques.md §§4–5; LL2-02 and LL1-00 hindsight). - Asymmetric definition. A false positive needs “high confidence” (67–95%) of no harm (LL2-02, p. 18), while action may rest on evidence the EEA itself grades “weak” or “moderate” (LL2-27, Table 27.2, p. 658). Unregulated alarms and risk–risk trade-offs are excluded; the 88 came from critics’ lists, so there is no denominator; Cox (2007) argued the criteria bias cases towards “real”; LL1’s own two candidates (sludge dumping, Y2K) were never assessed. - “Jury still out” cases. Of about 18 checked, about 12 moved towards harm or regulation (BPA, phthalates, PFOA, PCE, endocrine disruptors) and about 3 towards reassurance (GM food, mobile phones, Bt pollen and monarch butterflies); GM food safety arguably now meets the false-positive bar. The checking was selective (LL2-02 hindsight). - False positives were not short-lived. Saccharin labelling lasted 23 years and a federal listing 33; the EU irradiation list has not grown since 1999; the US cyclamate ban has lasted over 55 years; MMR, excluded as an “unregulated alarm”, did lasting harm; Germany’s nuclear phase-out costs an estimated €3–8bn a year, mostly in air-pollution deaths (LL2-02 hindsight). - The one-way statistical claim needs qualifying. Low power also inflates false positives, as LL2’s own mobile-phone warning shows; and several independent null lines, followed long enough, can cap large risks (LL1-16, LL2-26, LL2-21 hindsight). - What still holds. For data-poor chemicals, the dominant regulatory risk remains a default false negative (LL2-26 hindsight). - What counts as a false alarm. The floods chapter separates a warning whose risk was real but whose disaster did not arrive (the 1995 Dutch evacuation, when the levees held) from a false alarm, and notes that fear of error makes officials hesitate to warn (LL2-15, p. 354).

Strength. “Absence of evidence from an unperformed or underpowered search is not evidence of absence”: strong. “Definitions and thresholds decide how many errors of each kind are found”: strong. “False positives are rare relative to false negatives”: asserted and contested. “False positives are brief and narrow”: weakened. “Irreversibility makes missed harm costlier than unnecessary restriction”: moderate; sound as a conditional, but it fails where precaution creates serious countervailing risks, as with DDT and malaria (external/critiques.md §§3.2, 6) or Fukushima’s 2,351 disaster-related deaths among evacuees against no discernible radiation disease (LL2-18 hindsight).

Analysis. State the asymmetry as a conditional: a missed harm probably costs more than an unnecessary restriction when the agent is persistent, latent or irreversible, exposure is widespread, and the benefit forgone is modest or substitutable. Elsewhere it is an empirical question. Since unwarranted reassurance and unwarranted restriction both persist for decades, re-evaluation matters as much as the first decision (LL2-02, lesson 7, p. 35).

6. Status quo bias and institutional inertia#

6.1 The status quo as default#

Reports. Institutions tend “to favour the status quo and the short term perspective” (LL2-00, p. 11). The belief that “today’s technology is now safe” and sunk investment prolong exposure (LL2-28, p. 672). Governments “opted to retain the regulatory status quo” for nanomaterials (LL2-22, p. 541).

Evidence. Six mechanisms made the status quo the default: 1. Grandfathering and inherited categories. MTBE predated the new-substance cut-off (LL1-11, p. 116); nanomaterials inherited rules through their chemical identity numbers (LL2-22, pp. 537–541); the DES chapter asserts that legacy drugs escaped re-evaluation (LL1-08, pp. 89–90). 2. Provisional numbers hardened. The 1948 beryllium limit, reportedly chosen in a taxi, was adopted “tentatively” and then made permanent (LL2-06, p. 133); the 500 ppm vinyl chloride limit rested on one 1930s guinea-pig study (LL2-08, pp. 182, 193); the first UK TBT target was about tenfold too high (LL1-13, pp. 137–138). Hindsight adds the US benzene limit (unchanged since 1987) and the “temporary” aviation-fuel exemption. 3. Conditional approvals whose conditions lapsed. Leaded petrol was cleared in 1925 “provided that” it was controlled by “proper regulations”; the urged long-term study never happened, research for about 40 years was industry-funded, and voluntary compliance pre-empted binding rules (LL2-03, pp. 53, 56). 4. Committees and research in place of action (LL2-A2, pp. 703, 711; LL2-22, p. 547). The counter-rule: research “should supplement other risk-reducing regulatory measures and not be seen as a regulatory measure in itself” (LL2-02, p. 35). 5. Feasibility-based limits: “easily achievable” (benzene; LL1-04, p. 43); “what the industry felt was achievable” (vinyl chloride; LL2-08, p. 182). 6. Precautionary relabelling. Fishery reference points were a “precautionary gloss” (LL1-02, p. 24); northern cod’s later upgrade came partly from a lowered reference point, not more fish (LL1-02, LL2-17 hindsight).

Strength. Strong for mechanisms 2, 3 and 5 and the legal facts in 1 (primary records in at least 8 cases). Moderate for status quo bias as a general tendency: illustrated, never tested comparatively.

6.2 Dual mandates and dependence on the regulated#

Reports. MAFF was “responsible first to the industry and only second to consumers” (LL1-16, p. 179); the USDA “tended to align itself” with manufacturers and farmers (LL1-12, p. 128); the US DOE was beryllium’s sponsor, customer and regulator (LL2-06, pp. 132, 135); the US nanotechnology initiative both promoted and oversaw (LL2-22, pp. 546–548). The French pesticide bureau had three officials for 20,000 applications a year, with assessment “performed jointly with the industry” (LL2-16, pp. 382–383). Appraisal “frequently fails” through dependence on “information produced and owned by the very actors whose products are being assessed” (LL1-16, p. 179).

Hindsight. The European Court of Auditors (2012) found that “none” of four EU agencies adequately managed conflicts of interest; a re-analysis of industry chlorpyrifos studies found brain effects the sponsor’s summary had not reported (LL1-16 hindsight). Against the dual-mandate thesis: Phillips rejected producer bias in MAFF’s BSE policy (LL1-15 hindsight), and LL2-25 shows biased behaviour for only two of five dual-mandate examples (p. 612).

Strength. Strong for dependence on applicant data; moderate for dual mandates as a cause.

6.3 Credibility traps#

Reports and evidence. A regulator can become “part of the issue through its past decisions” (LL1-16, p. 180). Canada’s fisheries department stayed loyal to its model; its scientists were “lulled by false data signals” (LL2-17, pp. 413–415). As uncertainty fell, the stakes of admitting error rose, so there must be “room for them to turn around” (Guidotti, LL2-06, pp. 148–150). The IPCC’s response to criticism was “delayed and defensive” (LL2-14, p. 330). EU catch limits matched advice for only 8% of stocks in 1987–2006 (LL2-17, pp. 421–422); the 2026 North Sea cod limit was 14,034 t against zero-catch advice (LL1-02 hindsight).

Strength: moderate. Strong for cod and BSE; suggestive for the liability trap.

6.4 Silos and remits#

Appraisal “captive” to one discipline produces “institutional” ignorance (LL1-16, p. 174). Drug regulators had “limited expertise in environmental issues” (LL2-13, p. 284); wildlife warnings about hormones from 1970–73 were ignored until the late 1980s (LL1-14, p. 152); MTBE was appraised as an air measure and harmed water (LL1-11, p. 114); Swann’s cross-sectoral committee was never fully implemented (LL1-09, p. 94); turf disputes and deferral to EU coordination stalled invasive-species action (LL2-20, pp. 496–497). Strength: moderate to strong (at least seven cases).

6.5 Short horizons, transitions and forgetting#

Reports. A new administration withdrew draft rendering rules in 1979 as “an unnecessary burden on industry” (LL1-16, p. 180). A 1980s political shift raised the proof demanded before the Great Lakes clean-up (LL2-A2, p. 713). Memory fades between floods, the “hydro‑illogical cycle” (LL2-15, pp. 360–361), generalised as a “homo‑illogical cycle” (LL2-28, p. 680).

Hindsight. Protective measures have been reopened: the EPA moved to reconsider the US asbestos ban within fifteen months before withdrawing that motion; PFAS drinking-water limits face proposed rollback (LL1-00, LL2-26, LL2-27 hindsight). Post-Fukushima regulators largely held their ground, but Japan’s 2023 legislation moved reactor-lifetime decisions towards the promoting ministry (LL2-18, LL2-28 hindsight). Strength: moderate.

7. The information-to-action gap#

Reports. “The absence of political will … seems to be an even more important factor in these histories than is the availability of trusted information” (LL1-00, p. 4), yet the twelve lessons are mostly about information (LL1-17, p. 194). On PCE the science “has not been hidden. It has been ineffective in guiding and catalysing action”, and “uncertainty favours the side of inaction” (LL2-04, pp. 76, 86). In the Great Lakes, proven causation did not produce remediation, a “cognitive dissonance” between science and government (LL1-12, p. 130). At Minamata the failure was not using existing powers (LL2-05, pp. 99, 114). On acid rain, positions followed perceived costs, and “only when the issue was taken to the international level” did significant change come (LL1-10, p. 106); Norway accepted the science in 1976, the UK only in 1985 (LL1-16, p. 180). On climate, a framework came before human influence was detected, but effective measures did not follow (LL2-14, pp. 321–328). On floods, officials hesitate for fear of error, people at risk avoid warnings to protect house prices, and local authorities may lack legal power to raise the alarm (LL2-15, pp. 353–354, 360).

Evidence and hindsight. - Asbestos. Listing chrysotile under the Rotterdam Convention, recommended in 2005, was blocked under the consensus rule through 2025, with blockers presenting a dispute about evidence: contesting settled information can be how a lack of will shows (LL1-00 hindsight). - Local evidence drives action. Across 198 countries, a country’s own identification of mesothelioma raised the odds of an asbestos ban 14.1-fold (Chimed-Ochir et al. 2022; LL1-00 hindsight). - Floods. In 2021 forecasts showed the extreme rainfall at least two days ahead, yet 29–35% of surveyed residents in the worst-hit German states got no warning and 190 people died; in Valencia (2024) an official’s doubts about her powers “generated a paralysis at the decisive moments”, in the investigating judge’s words (LL2-15 hindsight). - Neonicotinoids. EFSA quickly confirmed that the bee assessment was blind to seed-dressings, but Member States blocked the 2013 guidance, and its revision still awaited a vote in September 2026 (LL2-16 hindsight). - Adoption is not reduction: two asbestos prosecutions in 1931–68 (LL1-05, p. 56); BSE non-compliance (LL1-15, p. 162); rated strong in the annex (LL2-A2, pp. 706, 708, 713, 715–716). - Warnings from elsewhere discounted (moderate). Florida bald-eagle declines were published in 1952 and 1958, but Great Lakes research began only after most of the local population was gone (LL1-12, p. 126); reasoning from the absence of harm elsewhere misled at Minamata, and the disaster repeated at Niigata (LL2-05, pp. 102, 105); the UK took nine years to accept Scandinavian acid-rain science (LL1-16, p. 180).

Strength. “Knowing is not acting”: strong as description (at least 10 cases). “Political will outranks information”: moderate, never tested comparatively; the two interact.

Analysis. The gap has at least four layers, each with its own remedy: knowledge not delivered (§2.1); delivered but contested (§2.2); accepted, but cost distribution blocks action (§4.3, §6); action formal but not implemented. Treating it as one “information problem” repeats LL1’s mismatch between diagnosis and remedy (LL1-17 digest). Two chapters propose a bridge, untested but aimed at the third layer: agree in advance “which diagnostic criteria and metrics will be used to elicit action” (LL2-17, p. 423), and link monitoring to “predetermined threshold level(s)” (LL2-12, p. 274).

8. Counter-evidence, complications and critiques#

8.1 Selection and the missing denominator. The strongest structural objection (external/critiques.md §§4, 9.1). Cases were chosen for known harm (LL1-00, pp. 11–12). Marchant (2003) adds that harm is easier to prove than its absence, so false positives stay provisional and undercounted. A random-sample comparison found no significant overall US–Europe difference in precaution over 1970–2004; precaution was selective, risk by risk (Hammitt et al. 2005). Mazur’s (2004) design, judging a whole period’s alarms true or false, would answer the objection; LL2 used Mazur only as a source of candidates. Every pattern above is evidence about mechanisms, not frequencies.

8.2 Hindsight in dating. For asbestos, Bartrip argues there was “no compelling medical or scientific evidence” before the late 1920s; the chapter cites him, misdated, as support, and Deane herself judged decisive statistics “practically unattainable” (LL1-05, p. 60; notes and hindsight). The “first warnings” for PCBs, DES and fisheries are generic. Outcome knowledge inflates judged foreseeability (Fischhoff 1975, in LL1-16 hindsight). Hindsight sometimes strengthened a warning, as with MTBE’s 1984–88 flags (LL1-11 hindsight).

8.3 Warnings that were wrong, over-weighted or overstated. - The reports’ own emerging warnings, the nearest thing to a prospective test, have a mixed record (LL2-27 hindsight): mobile phones largely not borne out (LL2-21 hindsight); the GM health claim unsupported (LL2-28, p. 674; LL2-19 hindsight); nanomaterials, specific signals only (LL2-22 hindsight); BPA and neonicotinoids moved towards the chapters. - Swine flu (1976): a warning that fitted prevailing theory was over-weighted (“Perhaps too much faith was placed on the ability of science to foresee”), with options and dissent sidelined (LL2-02, pp. 27–31). The same institutional pathologies can produce false positives. - Hormones: the EU overrode, disbanded and left unpublished its own expert committee (LL1-14, p. 150), suppression in the precautionary direction; the dispute was settled by trade quota, not science (LL1-14 hindsight). - Over-claims inside true-warning cases: PCB paediatric attributions (LL1-06 hindsight); DDT and adult breast cancer (LL2-11 hindsight); MTBE and asthma (LL1-11 digest); resistant enterococci as the “most convincing” antibiotic example (LL1-09 hindsight); Chernobyl toll figures (LL2-18 hindsight); the “irreversible” cod collapse (LL2-17 hindsight). - Warning-side actors contested credibility too: the EEA withdrew from the IARC mobile-phone meeting, citing publication delays and the chair’s “intellectual bias” (LL2-21, p. 520). - A risk of unfalsifiability: the GM chapter treats “the persistence of the same institutional patterns” as itself the emerging late lesson (LL2-19, p. 470). A lens that reads institutional patterns alone as a warning cannot be checked against outcomes.

8.4 Genuine ignorance. Ozone depletion before 1974, mesothelioma, DES’s transgenerational effects and BSE could not have been foreseen at deployment (LL1-16, p. 171; LL1-17, p. 192); a 1965 CFC assessment would have found “no known grounds for concern” (LL1-07, p. 82). Marchant’s point stands: precaution cannot prevent the unanticipated, so the fair charge is slow response once evidence emerged. The reports’ answer is property screening and monitoring, not prohibition.

8.5 Delay was not always bad faith. Phillips found BSE officials sincerely believed the risk remote (LL1-15 hindsight). Guidotti suspected “cognitive dissonance and denial rather than cupidity” over beryllium (LL2-06, p. 145). LL2-25 leaves calculated doubt-making and self-serving bias unreconciled and warns against blame “with hindsight” (pp. 607, 616). The 1883–85 fisheries inquiry found most complaints unfounded (LL1-02, p. 18). Assessors genuinely diverge on shared data: BPA, glyphosate, PCE (LL2-04, pp. 84–85; LL2-27 hindsight). Two Japanese courts accepted the state’s Minamata criteria (LL2-05 hindsight).

8.6 Acting on warnings has costs. Responses created new hazards: MTBE, itself a response to the lead warning and an air mandate (LL1-11; Goldstein, external/critiques.md §3.2); tall stacks (LL1-10); HCFCs and HFCs (LL1-07); BPS and BPF (LL2-10 hindsight); booster biocides (LL2-12). South Africa’s move off DDT failed when mosquitoes resisted the replacement (LL2-11, pp. 243, 250). Bans brought real costs: sugar-beet losses (LL2-16 hindsight), a rebound in therapeutic antibiotic use (LL1-09 hindsight). The reports file such cases as “risk–risk trade-offs”, outside the error count (external/critiques.md §5.3).

8.7 Inertia is symmetric. Labels and bans persist after concern fades: saccharin, cyclamate, irradiation (LL2-02 hindsight). Status quo bias is a general institutional property, not only a bias against protection.

8.8 The world changed after 2001. EU precaution became less scarce (court rulings, REACH, neonicotinoid restrictions, the BPA ban), so the “scarcity” diagnosis fits 1970–2000 best (LL1-16 hindsight); the counter-current is real too: the EU “innovation principle” and US deregulatory orders of 2025 (external/critiques.md §7; LL2-27 hindsight).

8.9 Authorship. Later evidence has been least kind where authors were most involved: mobile phones, the GM health claim, the nuclear casualty figures (external/critiques.md §9.1). The strongest evidence on warning-handling comes from contemporaneous documents: vinyl chloride, tobacco, beryllium, the Minamata ministry reply, the BSE minutes.

8.10 The reports’ own asymmetry. They set a low bar for a warning to count as “credible” and a high bar for a false positive. A lens built from them must apply one standard in both directions.

9. Technology-neutral diagnostic questions#

1. What kind of warning is this, and what would a specific, attributable version look like? Who is responsible for producing it, and by when? (§1.1, §3.1.)
2. Who is positioned to notice harm first (workers, users, neighbours, downstream operators, in-house scientists), and is there a protected channel to someone with authority to act, covering concerns about lawful products as well as breaches of law? (§1.2, §2.3.)
3. What does the developer know internally that overseers and the public do not, and what would surface it before litigation does: study registration, raw-data disclosure, independent verification? (§2.5; LL1-16 hindsight.)
4. Can current monitoring detect the harms that matter, with the right endpoint, power, baselines and detection limits? How large an effect could the evidence have missed? (§1.3; LL2-26, pp. 633–634.)
5. Is oversight calibrated to the first, most visible harm while slower, chronic, diffuse or indirect harms go unmeasured? (§1.4.)
6. What standard of proof is demanded before any protective step, of whom, and could the system ever meet it? Are there universal, sole-cause or “satisfy everyone” framings? Is the same standard applied to claims of safety? (§4.1.)
7. Who must produce the evidence? Can overseers require data without first proving risk, and are incumbent versions exempt from scrutiny that newcomers face? (§4.2, §6.1.)
8. Which limits, exemptions or approvals are “provisional”, “temporary” or conditional? What forces their review, and who pays for the follow-up they assume? (§6.1.)
9. Does any body both promote the technology and oversee its risks, or depend on the developer for information, staff or funding? (§6.2.)
10. Has anyone publicly committed to a safety claim that would now be costly to revise, and is there a route to change course without catastrophic liability or loss of face? (§2.4, §6.3.)
11. Is “more research” offered in place of interim action? Are the question, timeline, funder and independence stated, and have the criteria that would trigger action been agreed in advance? (§2.2, §7; LL2-17, p. 423.)
12. Would the harms be persistent, latent, dispersed or hard to reverse, and is the benefit forgone by restraint modest or replaceable? What new risks would a restriction or substitute create? (§5, §8.6.)
13. Who bears the costs of acting and of not acting, over what horizon, and do those raising concerns also carry the cost of proving them? (§4.3, §6.5.)
14. Would a warning be discounted because the harm has so far appeared elsewhere: another jurisdiction, population, or part of the system nobody values commercially? (§3.2, §7.)
15. If the warning proves wrong, how fast could the response be reversed? If right, how fast could harm actually fall? Are re-evaluation and exit built in on both sides? (§3.1, §5, §8.7.)