T06: Governance, institutions, law and participation#

Strand A cross-cutting synthesis across both EEA Late lessons from early warnings reports. Written 26 September 2026 from the section digests, the audited notes, the hindsight files and the two external notes (external/critiques.md, external/context.md). Direct quotations from the reports were checked against working/text/chunks/ on the same day; quotations marked [H: …] come from the hindsight files.

How to read this document#

Sources and page references. LL1 is the 2001 report (report page = PDF page). LL2 is the 2013 report (report page = PDF page − 2). Citations take the form “LL1-15, pp. 161–162” (project section id, report pages). “[H: LL1-15]” marks a point from that section’s hindsight file (developments to September 2026), not from the report.

Three voices, kept separate. Within each pattern, The reports say records the authors’ claims with their hedges; Evidence and hindsight records what the cases document and what later evidence shows; Analysis is my own reading and should not be cited as the reports’ view.

Strength ratings apply to the pattern as a transferable claim, not to any single case: - Strong: documented in several independent cases, with primary or contemporaneous evidence, and not overturned by later evidence. - Moderate: documented in a few cases, or well documented in one case with plausible but untested generalisation. - Suggestive: one or two instances, or evidence that is circumstantial, retrospective or from interested parties. - Asserted: stated by the reports without supporting case evidence, or normative.

A standing caveat. Both reports select cases in which harm materialised. Authors were chosen because they had taken part in the histories (LL1-00, p. 12; LL2-00, pp. 9–10). Governance lessons drawn from such a set can show that a mechanism operates. They cannot show how often it operates, or what the same institutions did in cases where warnings proved false (see section 12).

Summary of findings#

1. What “precaution” means: forms, definitions and labels#

The reports say. - LL1 insists that “precaution” covers very different legal commitments. Table 1.2 runs from the 1990 North Sea formulation, which allows action on persistent, toxic, bioaccumulative substances “even where there is no scientific evidence to prove a causal link”, to Rio Principle 15, which is negative (“lack of full scientific certainty shall not be used as a reason for postponing cost-effective measures”) and qualified by “capabilities” (LL1-00, p. 14). - The German Vorsorgeprinzip was a programme, not a trigger. It included monitoring, burden reduction, clean production and innovation, proportionality and stakeholder cooperation (LL1-00, p. 13). - LL1’s own definition is “an overarching framework of thinking that governs the use of foresight” under uncertainty and ignorance (LL1-17, p. 192). Table 17.1 pairs risk with “prevention”, uncertainty with “precautionary prevention” and ignorance with “precaution” (p. 192). - LL2 offers an EEA working definition: precaution “provides justification for public policy and other actions in situations of scientific complexity, uncertainty and ignorance … using an appropriate strength of scientific evidence, and taking into account the pros and cons of action and inaction and their distribution” (LL2-27, p. 649). It is deliberately affirmative, against Rio’s “triple negatives” (p. 649). LL2 adds ambiguity, answered by “participatory precaution” (Table 27.1, p. 656). - Several authors separate precaution from prevention. Infante says precaution “is not relevant” in a workplace where harm was known (LL1-04, p. 46). Gee says that once evidence “proves” a causal link “there is no need for the PP as the issue is then firmly in the ‘prevention principle’ area” (LL2-27, p. 657). The Minamata authors locate the failure in not using existing powers, not in a missing doctrine (LL2-05, pp. 99, 114).

Evidence and hindsight. - The label is repeatedly detached from practice. The fisheries chapter calls EU and ICES reference points a “precautionary gloss” on an unchanged regime (LL1-02, p. 24). UK ministers later described their BSE policy as ultra-precautionary (LL1-15, p. 157). The ethinyl oestradiol authors argue Rio’s “cost effective” wording is precaution’s “Achilles heel”, making it “a perfect excuse for inaction” (LL2-13, p. 296). - The reports themselves relabel. “Precautionary prevention” enrols accepted public-health successes, such as John Snow’s pump handle, into precaution (LL1-00, pp. 12–15). Ch17’s range of proof levels omits the lowest rung in Ch16’s own Table 16.1 (LL1-16, p. 184; LL1-17, p. 193). - Hindsight shows the definitional contest continuing, and law settling on narrower versions than the reports’. The EU General Court in Pfizer (2002) upheld action before harm is proven but required a risk “adequately backed up by the scientific data available at the time”, not “purely hypothetical” (paras 143–144) [H: LL1-09]. The European Parliament’s research service found institutions “do not agree on the method” for deciding when to apply precaution [H: LL1-17]. The ICJ’s 2025 climate advisory opinion treated precaution as an element of due diligence [H: LL2-14]. An industry-promoted “innovation principle” entered the Horizon Europe Regulation (2021) as a procedural counter-definition [H: LL2-27; critiques.md §7].

Analysis. The reports’ strongest contribution here is diagnostic: they show that the word “precaution” can conceal the four choices that actually matter (the trigger, the strength of evidence, whether costs qualify the duty, and who must act). Their own definitions push these choices into procedural placeholders (“appropriate strength of evidence”, “pros and cons … and their distribution”), which answers the “incoherence” critique but moves the hard decisions elsewhere (critiques.md §9.1). Most of the historical failures the reports document are failures of prevention (acting on known harm), not of precaution under uncertainty. That matters for the lens: the governance lessons from asbestos, benzene, vinyl chloride, Minamata or tobacco are lessons about institutions failing to act on strong evidence.

Strength: strong for the variation and the label–practice gap (Table 1.2 plus fisheries, BSE, EE2, climate and Minamata). The reports’ own relabelling is a caution against taking their definitions as neutral.

2. Evidence thresholds and the burden of proof as governance choices#

The reports say. - The level of proof needed to act “can radically shift the size, nature and distribution of the costs of being wrong. This is a key political decision with profound ethical implications.” It should depend on the potential harm, the claimed benefits, the alternatives and “the potential costs of being wrong in both directions” (LL1-17, p. 193). - Legislatures can split the burden between parties. Sweden’s 1973 chemicals law demands evidence “beyond all reasonable doubt” of safety 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). - Gee draws on Bradford Hill’s call for “differential standards before we convict” (LL2-27, pp. 656–657). He argues that “not established” judgements “seldom” say who bears the error, “risk takers or risk makers”, or for what purpose the evidence is being judged (pp. 657–658). - Whether a demand for more information is “paralysis” or “prudent and careful evaluation” depends on how pros and cons fall on each group (LL1-16, p. 182).

Evidence and hindsight. The cases show thresholds being set, raised or reframed by whoever controls the question: - Courts raising the bar. The US Supreme Court’s 1980 Benzene decision required a quantified “significant risk” finding, which the author calls a “straightjacket” (LL1-04, pp. 40–41; LL1-16, p. 181). - Question framing. The UK government asked its Medical Research Council panel whether evidence “prove[d]” that lead was doing British children “obvious harm”, not at what level effects were detectable (LL2-03, Millstone panel, p. 68). The French pesticide commission answered whether Gaucho was “solely responsible, at national level, for all” bee losses, “a question that had never been asked” (LL2-16, p. 379). - Proof drift. At Minamata the required proof moved from route to agent to mechanism. The health ministry refused to apply food law because there was no “clear evidence that all fish and all shellfish are poisoned” (LL2-05, p. 99). - Burden paradoxes. Under US chemicals law as it stood, the regulator had to show risk before it could demand data (LL2-22, p. 537). Policy-makers “often” want monetised costs of inaction before acting (LL2-23, p. 564); in my reading, that leaves the burden on warners until someone prices the harm. - Asymmetric bars in the same system. Reassurance for drinking water was given quickly on “this small set of studies”, while action on wildlife waited decades (LL2-13, p. 283). - Hindsight. Pfizer held that a scientific committee has “neither democratic legitimacy nor political responsibilities”, so the Council could depart from it with adequately reasoned grounds (paras 199–201) [H: LL1-09; H: LL1-17]. EU pesticide law now bars substances in hazard categories 1A and 1B, and the EU added “suspected” persistence and endocrine categories in 2023; the amended US Toxic Substances Control Act (2016) requires risk findings “without consideration of costs” [H: LL1-17; H: LL2-27]. Moving the other way, US Executive Order 14303 (2025) confines “overly precautionary assumptions” to cases required by law [H: LL2-02; H: LL2-27]. The 2023 EU glyphosate renewal went through with neither member-state committee delivering an opinion: the threshold was set by default [H: LL1-17].

Analysis. This is the reports’ most durable governance insight, and it is not specific to precaution. It holds whether one favours more or less precaution: the evidential threshold is a device for allocating the cost of error. The historical record since 2013 has not moved towards the reports’ preferred low thresholds. It has made the choice openly political in both directions. The reports give no procedure for deciding who sets the threshold or how the factors are weighed (LL1-17 notes, §7).

Strength: strong as analysis (12+ instances across LL1-04, LL1-09, LL1-12, LL1-15, LL1-16, LL1-17, LL2-03, LL2-05, LL2-13, LL2-16, LL2-22, LL2-23, LL2-27; later legislation and case law treat it the same way). Partly held up as prescription: explicit, two-way weighing remains uncommon [H: LL1-17].

3. Proportionality, costs and who pays#

The reports say. - LL1 treats proportionality as part of the original Vorsorge programme (LL1-00, p. 13). Its conclusion, however, speaks of balancing “proportionate and precautionary public policies”, recognising that “over-precaution can also be expensive, in terms of lost opportunities for innovation” (LL1-17, p. 194). That sets the two against each other, as the notes observe. - LL2’s definition requires “pros and cons … and their distribution” across “groups, regions, and generations” (LL2-27, p. 649), wider than cost–benefit analysis. Costs and benefits are “skewed towards the tangible short-term compliance costs … and against the long-term diffuse benefits” (p. 659). - The costs-of-inaction chapter accepts that policy-makers often wait for monetised damage estimates and argues these can be used in precaution’s favour, reporting ranges and testing the conservative bound (LL2-23, pp. 564, 571, 577).

Evidence and hindsight. - Cost logics that delayed action. The BSE minister refused to slaughter and compensate about 1,200 clinical cases (about £1.5m) citing “offsetting savings”, consistency with an unrelated sugar-beet claim, and exports (LL1-15, pp. 158–159). PCB restrictions proceeded from cheap to costly uses (LL1-06, pp. 66, 72). Officials in the Great Lakes demanded causal proof before “massive” spending (LL1-12, p. 128). Ethinyl oestradiol regulation “stall[ed] in the still waters of cost-benefit analysis” (LL2-13, p. 293). - Inflated cost forecasts by regulated parties recur (vinyl chloride, LL2-08, pp. 187, 192; leaded petrol, LL2-03, p. 60; a US DES ban, LL1-14, p. 150), though the vinyl chloride like-for-like overestimate is about fourfold, not the headline contrast [H: LL2-08]. - Where precaution cost more than it protected. The EU hormones ban was taken against both its own expert committee and JECFA; the authors find “no good evidence” it protected health and report sanctions of about EUR 160m a year (LL1-14, pp. 150, 153–154). Hindsight corrects the sanction to a ceiling of about EUR 117–139m, and shows the dispute was settled by beef quotas, with costs falling on third-country exporters [H: LL1-14]. The false-positive chapter’s own cases show that measures can outlast their basis: saccharin labelling took 23 years to lift and irradiation approvals were frozen for 20 years [H: LL2-02]. - Proportionate de-escalation. The UK Food Standards Act 1999 requires account of risks “including any uncertainty” and the costs and benefits of acting or not (s. 23). The Food Standards Agency’s open review found the over-thirty-months rule cost about £2bn per death prevented, and it was replaced by testing in 2005 [H: LL1-15]. - Reallocating cost rather than weighing it. Where the EE2 authors wanted public debate about acceptable cost (LL2-13, p. 297), the EU instead required producers of medicines and cosmetics to fund at least 80% of new wastewater treatment [H: LL2-13].

Analysis. The reports are right that the timing of costs biases decisions towards delay. They are weakest on proportionality in the other direction: what the critics call “over-regulation of minor risks” is defined out of LL2’s false-positive analysis (critiques.md §5.3), and Majone’s aflatoxin argument about trade costs to poor exporters is answered by classifying aflatoxin as a “real risk”, which does not engage it. Hindsight adds a lesson the reports did not draw: openness about costs served proportionality in both directions, including relaxing measures.

Strength: moderate. Delay-by-cost is well documented in about eight cases. The claim that cost-effectiveness clauses cause delay is untested (LL2-13 notes). The counter-cases (hormones, saccharin, irradiation, BSE measures later relaxed) come partly from inside the reports.

4. Risk assessment versus risk management: scope, framing and the role of advisers#

The reports say. - LL1 rejects the idea that appraisal, management and communication are sequential stages. Stakeholder involvement must “begin at the beginning” (LL1-16, p. 186). This departs from the Commission’s 2000 Communication, which places precaution in risk management (context.md §2.7). - The hormones authors conclude that because committees were never asked to characterise uncertainty, the original EU ban was “in reality, a political risk assessment” (LL1-14, p. 154). JECFA assessed only authorised use, single substances and manufacturers’ data (p. 150). - The BSE authors argue that ministers should articulate and justify framing assumptions, and that advisers erred in acquiescing to present mixed scientific and non-scientific advice “as if it was purely scientific” (LL1-15, p. 165). - Gee argues that assessment committees “may need to consider the consequences of their judgements as well as just causation” (LL2-27, p. 658). The GM chapter wants precaution inside risk assessment, through choices about error types (LL2-19, p. 470). - LL2’s conclusion asks committees to explain their “choice of paradigms, assumptions, criteria for accepting evidence, weights … and how uncertainties were handled” (LL2-28, p. 677).

Evidence and hindsight. - Committees diverge on shared evidence. 29 trichloroethylene cancer assessments reached four types of conclusion, split by type of institution (LL2-04, Rudén panel, pp. 84–85). BPA guidance values diverged by orders of magnitude on largely shared studies, driven by evidence-admission rules (LL2-10, pp. 221–223). MTBE carcinogenicity verdicts ranged from “not classifiable” to “carcinogenic potential” on the same data (LL1-11, p. 113). The same scrapie evidence led the US but not the UK to exclude affected animals (LL1-16, p. 181; the US decision is unverified [H: LL1-16]). - Advisers pre-filter. Southwood’s BSE committee thought a brain ban might not be “politically feasible” (LL1-15, p. 160; LL1-16, p. 179). This rests on one retrospective remark. - Legacy methods. Acute, spray-based bee tests were applied to systemic seed-dressings (LL2-16, pp. 377, 389). Probabilistic safety assessment bounded by listed scenarios missed common-cause cascades (LL2-18, pp. 447–448). Remit decides what a review can find: security was excluded from post-Fukushima stress tests (LL2-18, p. 444). - Hindsight. The EU formally separated risk assessment (EFSA) from management (the Commission) in 2002. But the UK Food Standards Agency combined both and kept public confidence near 80%, while EFSA’s independence became its own controversy (the European Court of Auditors found in 2012 that none of four agencies adequately managed conflicts of interest) [H: LL1-15; H: LL1-16]. EFSA made uncertainty analysis “unconditional” in 2018, with a carve-out for standardised assessments, exactly the kind the hormones chapter criticised [H: LL1-14]. Divergence persisted (BPA, glyphosate, titanium dioxide, aspartame); explanation of it by committees remains rare, though WHO’s 2023 aspartame communication is a partial example [H: LL2-28]. EU member states blocked EFSA’s 2013 bee guidance for over a decade: choosing an assessment method was itself a political act [H: LL2-16].

Analysis. The reports’ key move is that framing is a management decision disguised as a scientific one. Later evidence supports that, but not the institutional fix most associated with it. Separating assessment from management was neither necessary for trust (FSA) nor sufficient for independence (EFSA). What seems to matter more is who writes the question, whether the scope and evidence rules are published, and whether divergence is explained.

Strength: strong for divergence on shared evidence and for scope determining verdicts (10+ instances across LL1-11, LL1-14, LL1-15, LL1-16, LL2-04, LL2-10, LL2-16, LL2-18, LL2-21, LL2-28). Moderate for the remedies.

5. Regulatory independence, sponsorship and information dependence#

The reports say. - Lesson 10: interested parties should argue their claims and independent, accountable institutions adjudicate (LL1-16, pp. 178–179). Appraisal “frequently fails” because it depends on information “produced and owned by the very actors whose products are being assessed”; independent information is “necessary, if not sufficient” (p. 179). The editors concede that “not all” cases show this distorting effect (p. 179). - The BSE authors say genuine precaution “would have necessitated, firstly and most fundamentally”, separating regulation from sponsorship (LL1-15, p. 165). - The business chapter lists five promote-and-protect mandates (USDA, DFO, MAFF, the French agriculture ministry, US DOE) as sources of bias towards demanding proof (LL2-25, p. 612).

Evidence and hindsight. Sponsor-regulator and dependence patterns recur across about ten cases: - MAFF, which was “responsible first to the industry and only second to consumers” (LL1-16, p. 179), told the Department of Health about BSE 17 months late (LL1-15, pp. 159–160). The Commission’s internal-market and agriculture directorates held the BSE brief (p. 163). - USDA “tended to align itself” with manufacturers until DDT registration moved to EPA and litigation broke the deadlock (LL1-12, p. 128). - The US Energy Secretary said beryllium worker safety had been the “last priority”; DOE was sponsor, customer and regulator (LL2-06, p. 132). - Japan’s pre-Fukushima regulators; the Diet commission’s “regulatory capture” (LL2-18, pp. 441–443). - The US nanotechnology initiative promoted the technology and oversaw its risks (LL2-22, pp. 546–548). - A California governor replaced state fisheries experts in 1939 (not independently verified [H: LL1-02]), and a federal bureau opposed state precaution as a drag on commerce (LL1-02, p. 20; LL1-16, p. 180). - Three French officials handled 20,000 pesticide applications a year, with risk assessment “performed jointly with the industry” (LL2-16, pp. 382–383). The same ministry told public researchers not to measure below Bayer’s 10 ppb detection limit (p. 373). - Standard-setting bodies with producer members: benzene TLVs (LL1-04, pp. 43, 46), vinyl chloride limits reflecting “what the industry felt was achievable” (LL2-08, p. 182), tobacco-shaped ISO and ASHRAE standards (LL2-07, pp. 162–163). Chisso funding at Kumamoto University (LL2-05, p. 104).

Complications inside the reports. - Industry was not uniform. Pet-food and meat-products firms moved to remove offal before the BSE regulator (LL1-15, p. 160). Swedish and Danish farmers asked for, or adopted, growth-promoter restrictions before governments (LL1-09, pp. 95–96). - Guidotti’s beryllium panel argues for auditing interested science rather than discounting it (LL2-06, p. 148).

Hindsight. - Remedies adopted. The FSA (2000), EFSA (2002); the EU Transparency Regulation 2019/1381 (burden stays on applicants, plus pre-notification of commissioned studies, disclosure and verification studies); Blaise (2019), telling authorities not to give “preponderant weight” to applicant studies. An independent re-analysis of raw chlorpyrifos data found neurodevelopmental effects an industry summary had not reported [H: LL1-16]. - Complications. Phillips found the Department of Health, which had no sponsorship role, as keen to reassure as MAFF; Germany’s separated ministries also failed on BSE; the Ireland–Austria comparison is confounded by exposure (5 versus 1,656 cases) [H: LL1-15]. OSHA’s 2017 beryllium rule was “based, in part” on a joint Materion–United Steelworkers draft [H: LL2-06]. - Erosion. Japan’s post-2011 regulator held its ground, but 2023 legislation moved lifetime decisions towards the promoting ministry. A 2025 US executive order directs the nuclear regulator to speed licensing [H: LL2-18]. The nanotechnology initiative’s safety research share fell from about 10% (2016) to a requested 1% (FY2026) [H: LL2-22].

Analysis. Within cases, dual mandates visibly shaped decisions. As a comparative law (separate the functions and protection improves) the evidence is thin, and hindsight shows independence is rebuilt after disasters and then eroded when authority drifts back to promoting ministries. The durable technology-neutral form is about evidence, not organograms: who generates the evidence, whether all commissioned studies must be registered, whether raw data are open, and whether an independent body has funded capacity to verify.

Strength: strong within cases (about ten); moderate to weak as a comparative generalisation [H: LL1-15 downgrades the comparative claim to weak]; remedies partly adopted, effect on outcomes unproven.

6. Trust, reassurance and communication#

The reports say. - LL1 opens on “very low” public trust “in the politicians and scientists” protecting people, “especially in Europe”, after BSE, Belgian dioxins and contaminated blood (LL1-00, p. 16). - The BSE chapter’s signature mechanism: once beef was declared safe, any further measure, “no matter how useful or cheap”, implied the reassurance was false. Measures were refused “not because of their immediate costs” but for presentational reasons (LL1-15, pp. 161–162). - Institutional efforts to reassure by intensifying research on known uncertainties “are futile”, and the approach “only feeds public mistrust” by displaying its own “denial of ignorance” (LL1-16, p. 185). - Lost trust after false reassurance “can sometimes be as significant as the economic costs” (LL2-27, p. 658). Nuclear regulation speaks “the language of certainty” over uncertain science (LL2-18, p. 448). - Experts should not talk “with one voice” but make disagreement understandable (LL2-20, p. 499); divergent expert views should be published side by side (LL2-21, fn 12, p. 518); disagreement should not be hidden (LL2-02, p. 34).

Evidence and hindsight. - BSE. In May 1990 SEAC advised that “no risk” could not be stated categorically; in June the minister told Parliament beef was “perfectly safe” (LL1-15, p. 161). Cheap measures on mechanically recovered meat and head meat were blocked lest they invite demands for more (p. 162). In 1995, 48% of abattoirs visited failed the offal rules, and an enforcer said the message made them “a bit of window dressing” (p. 162). - Similar patterns elsewhere. Monsanto’s public denial beside private acknowledgement of PCB contamination (LL1-06, p. 65); a contract that kept a UK effluent survey unpublished until 1994 (LL2-13, p. 284); the Commission’s unpublished hormones committee conclusions (LL1-14, p. 150); the IPCC’s “delayed and defensive” response to errors (LL2-14, p. 330); a divided consortium’s “elegant and oracular” conclusion, captured by each side (LL2-21, pp. 517–518); Minamata payments without recognition (LL2-05, pp. 107–110). - Hindsight. Phillips found the reassurance policy aimed at “sedation” and contributed to “lack of diligence” in enforcement; once enforced, offal reaching renderers more than doubled [H: LL1-15]. But Phillips also found officials sincerely believed the risk was remote. The chapter’s “covertly subordinated” charge should not be carried forward as fact [H: LL1-15]. - The trust premise was wrong in part. Eurobarometer data fielded in 2001 show scientists relatively highly trusted (the most trusted source after a local disaster); politicians were not [H: LL1-00]. A 2025 68-country study found “no widespread lack of trust in scientists” [H: LL2-00]. - Experiments on uncertainty. Communicating quantified uncertainty cost little trust (van der Bles et al. 2020) [H: LL1-16]. “Futile” is overstated [H: LL1-16]. - Costs of repair. Much post-1996 BSE spending bought back credibility and market access rather than reducing risk [H: LL1-15]. At Fukushima the measurable harms came from evacuation and overdiagnosis, not radiation [H: LL2-18]. Protective communication and action carry their own costs.

Analysis. The reassurance trap is the best-documented, least hindsight-dependent mechanism in this theme: officials’ own minutes record it. Its transferable form concerns commitment. An early categorical claim raises the political price of every later correction, collapses graded options into hold the line or do everything, and tells enforcers the rules do not matter. The reports’ general claims about public trust were less accurate than their case-level mechanism.

Strength: strong for the reassurance trap and its effect on enforcement (BSE, with contemporaneous documents, plus five other cases showing public certainty over private doubt). Moderate for open disagreement preserves credibility. Weak for the reports’ premise of generally low trust in scientists.

7. Public engagement, lay knowledge and democratic legitimacy#

The reports say. - Lessons 8 and 9: use lay and local knowledge, and take account of different groups’ values (LL1-16, pp. 177–178). - Participation should be “at an early stage, broadly drawn”, and local, but “without the need for what would be paralysing indiscriminate full public participation in every single decision” (LL1-16, p. 188). Consultation can “degenerate” into an exercise “driven by the sponsor’s agenda” (p. 188). - Early stakeholder involvement “may also improve public trust … without necessarily stifling innovation” (LL1-17, p. 193). - LL2 proposes a participatory, precautionary risk-analysis cycle with stakeholder review at the hub, most intense at framing and option choice (LL2-27, Fig. 27.2, p. 660). Such recommendations “do not appear to be reflected in most existing international and European arrangements” (p. 659). - Pathway decisions are “made by a few people on behalf of many” (LL2-28, pp. 671, 678). Minamata shows a “fundamental democratic deficit” (LL2-05, pp. 92, 114). The EE2 public were “silent witnesses” to “a poorly understood, closed process”; deciding on costs alone “loads the dice before they are thrown” (LL2-13, p. 297). - The 2013 preface calls for “the power structures of knowledge” to change (LL2-00, p. 8), while also saying questions should be framed around which decisions to delegate to experts (p. 8).

Evidence and hindsight. - Lay and frontline detection. Harm was first seen outside the expert core by women factory inspectors (asbestos, LL1-05, p. 53), inshore fishers (LL1-02, p. 21; LL2-17, p. 412), DBCP workers comparing notes at lunch and “their wives” (LL2-09, pp. 204, 211), a Minamata mother (LL2-05, pp. 105–106), beekeepers (LL2-16, pp. 372–373), environmental health officers (LL1-15, p. 162) and Love Canal residents (LL1-12, p. 127). Lay knowledge was also positional: fishers’ dissent tracked their interests (LL1-02, pp. 17, 20–21), and the reports credit public intuition mainly when it proved right (LL1-16 digest caveats). - Legitimacy constraints. A Great Lakes mirex fish ban that restricted individuals rather than sources collapsed (LL1-12, p. 128). An animal-cruelty prosecution ended Italy’s grey squirrel eradication (LL2-20, p. 495). - Participation by crisis. UK unleaded petrol was accepted “within half an hour” because of a public campaign and an imminent election (LL2-03, Haigh panel, p. 63). German acid-rain action followed Green Party pressure (LL1-10, pp. 104–105, 107). - Hindsight. The UK’s GM Nation? debate (2003) was flawed in design and unrepresentative. Deliberative bodies raise perceived legitimacy mainly when their recommendations are honoured (Germann et al. 2022). Engagement advocates later conceded “over-promising” [H: LL1-16; H: LL1-17]. EU law opened GMO cultivation to member-state opt-outs on socio-economic and policy grounds (Directive 2015/412) [H: LL1-17]. The Brent Spar outcome rested partly on a Greenpeace oil estimate that was wrong and withdrawn [H: LL1-17]. What advanced most was not participatory framing but transparency and legal standing: EU study disclosure (2019), NGO rights to request internal review (2021), and PAN Europe (2025) [H: LL2-27].

Analysis. Two different claims need separating. That people close to a hazard often notice harm first is well supported, though the selection of cases inflates it. That participation in framing improves decisions or builds trust is largely normative in the reports and only conditionally supported later. The reports under-theorise the legitimacy problem their own cases expose: in the Nunavut fisheries case, a minister’s weighing of socio-economic factors was lawful even as it overrode scientific advice (LL2-17, pp. 419–420), and the reports treat such weighing mainly as capture.

Strength: moderate for lay and frontline early detection (10+ cases, selected on outcome). Suggestive for participation improving outcomes. Moderate (hindsight) for the conditional form: participation adds legitimacy only when it visibly changes outcomes.

8. Implementation and the gap between rule and reality#

The reports say. Adopting a rule is not the same as reducing risk. This is not one of the twelve lessons, but it runs through the case chapters, and the 2013 annex draws it out (LL2-A2, pp. 706–716).

Evidence and hindsight. - Paper rules. The world’s first asbestos regulations (1931) produced two prosecutions in 37 years (LL1-05, pp. 54, 56). The Great Lakes pledge to “virtually eliminate” persistent toxics was “not properly implemented” (LL1-12, p. 132). Radiation recommendations without legal force left misuse unchecked (LL1-03, p. 34). - Dilution. Swann’s 1969 antimicrobial recommendations were adopted, then “gradually diluted” (LL1-09, p. 94). - Conditions that lapse. Leaded petrol was approved “provided that” it was “controlled by proper regulations”, with long-term public study urged; neither followed, and for 40 years TEL research was industry-funded (LL2-03, pp. 53, 56). The UK set lead targets only once they had been met (Millstone, LL2-03, p. 68). - Frameworks without teeth. Climate framework commitments came quickly but were “not really legally binding”, and Kyoto compliance was flattered by windfalls (LL2-14, pp. 322, 324). - Open-ended exemptions. The DDT exemption under the Stockholm Convention has no end date and “no timetable” (LL2-11, pp. 244, 252). Voluntary codes for invasive species had “limited effectiveness and buy-in” (LL2-20, p. 498). - Hindsight. - EFSA’s 2013 bee guidance was blocked by member states, and the 2002 guidance “remains the basis” for approvals. Member states granted 206 emergency authorisations for restricted neonicotinoids in 2013–2019 [H: LL2-16]. - Leaded aviation fuel, exempted as “temporary” in 1996, remains the largest US source of airborne lead [H: LL2-03]. - Most member states had not delivered invasive-species pathway plans by 2021 [H: LL2-20]. The Floods Directive’s risk reduction could not be verified because its objectives were “generally not quantified or time-bound” [H: LL2-15]. The US EPA later wrote that its voluntary nanomaterial scheme yielded “limited reporting” [H: LL2-22]. - Counter-evidence. Some rules worked fast once enforced. UK BSE cases fell from 1,248 in the 1995 birth cohort to 111 in the 1996 cohort, born under the all-species feed ban [H: LL1-15]. TBT imposex fell after the global ban [H: LL1-13]. European acidity exceedance fell to 3.5% by 2024 [H: LL1-10].

Analysis. For a lens, implementation is where governance claims are most easily tested. The recurring failure modes are: - unfunded conditions; - exemptions that become permanent; - voluntary or point-of-sale controls with no measurement; - process-based laws judged on process; - rules whose stated purpose is played down.

Strength: strong. More than 12 cases across both volumes, with hindsight adding further instances. The positive counter-cases show the gap is not inevitable; they share enforcement, measurement and a limited set of regulated actors.

9. International coordination, trade and jurisdiction#

The reports say. - Japan’s continuing TBT contamination despite a national ban “clearly illustrates the transboundary nature of the problem”; “It would seem that universal, global restrictions are the only way to address the totality of the TBT problem” (LL1-13, p. 142). For sulphur dioxide, “Only when the issue was taken to the international level could significant change occur” (LL1-10, pp. 106–107). - Global problems need global institutions; EU membership rules obstructed Sweden’s national antimicrobial policy (LL1-16, pp. 180–181). - Waiting for EU-level coordination became an “excuse for inaction” on invasive species (LL2-20, Box 20.4, p. 501). - Precaution has mostly been applied where “those implementing policies were generally also the ones benefitting”; climate breaks that alignment (LL2-14, p. 337). - LL1 aimed at “improved transatlantic agreement” and credited the 2000 Communication with helping avoid trade disputes (LL1-00, pp. 3, 13).

Evidence and hindsight. - Institutional reach. - TBT controls began nationally (France 1982; UK 1985–87). The IMO first said no further controls were needed (1994), then reversed after North Sea ministers pressed it (1995–97) (LL1-13, pp. 136–141). Hindsight: the convention took effect in 2008 and covered 95.77% of tonnage by 2025 [H: LL1-13]. - Acid rain moved through joint monitoring (EMEP), a convention (CLRTAP 1979) and critical loads. The UK and Poland did not sign the 1985 protocol (LL1-10, pp. 104–107). - Montreal moved by consensus-paced ratchets, with a fund exceeding USD 1bn (LL1-07, pp. 78, 80–81). Hindsight shows real adaptive capacity (2007 HCFC acceleration, Kigali 2016, illegal CFC-11 detected and reversed), with exemptions as the main residual leak [H: LL1-07]. - Unilateral leakage. UK meat-and-bone-meal exports to the EU doubled after the 1988 domestic feed ban (LL1-15, p. 163). DBCP was banned for US use but exported for years (LL2-09, pp. 207–209). DES remained in use in Europe and elsewhere after US action (LL1-08, p. 89). The US claimed its unilateral aerosol ban cost it CFC market share (LL1-07, p. 80). Hindsight: the EU extended its growth-promoter ban to imports from September 2026, closing a gap the manufacturer had pointed to in 1999, and the first month suspended several Brazilian product categories [H: LL1-09]. - Trade law. The WTO upheld the French asbestos ban, accepting qualitative evidence and minority scientific opinion (LL1-05, Box 5.1, p. 57). It found the EU hormones ban not based on a risk assessment (LL1-14, p. 153). Hindsight: the 2008 Appellate Body accepted measures resting on “divergent or minority views” from “qualified and respected sources” but never ruled on the ban; the dispute was settled by quota deals; the Appellate Body has been unable to hear appeals since 2020 [H: LL1-14]. LL1’s claim that precaution would avoid trade disputes “did not hold” [H: LL1-00]. - Consensus vetoes. Chrysotile asbestos has been recommended for listing under the Rotterdam Convention since 2006 and remains unlisted, blocked by producer states [H: LL1-05; H: LL2-A3]. DBCP was never listed [H: LL2-09]. - Small jurisdictions sometimes lead. Bermuda funded its own coral toxicology and banned booster biocides faster than the EU (LL2-12, pp. 271–272). Germany led the EU on lead (LL2-03, p. 63). LL1 itself found precaution varied by institution, not continent (LL1-16, p. 168), which later random-sample work confirmed [H: LL1-16].

Analysis. Where hazards are carried by mobile sources, trade or shared media, protection tracks the reach and ratification of the weakest necessary institution. Consensus rules give producers a veto, and unilateral standards leak until extended to trade, at a trade cost. The positive regimes (acid rain, ozone, TBT) share jointly produced monitoring and a mechanism for tightening over time.

Strength: strong for reach must match the hazard and unilateral action leaks (10 cases: TBT, SO2, CFCs, BSE, antimicrobials, DBCP, DES, asbestos, DDT, climate). Moderate for shared monitoring enables agreement, which was necessary but not sufficient (LL1-10 digest).

10. Monitoring, adaptive governance, review and exit#

The reports say. - Lesson 2 calls for monitoring and research for early warnings (LL1-16, pp. 171–173). The radiation chapter’s one explicit recommendation is to fund long-term epidemiological databases “even when an immediate need is not perceived” (LL1-03, p. 36). The TBT authors call baselines “frequently underrated” (LL1-13, p. 142). - Monitoring without predetermined action thresholds becomes an “academic pursuit” (LL2-12, p. 274). Societies should agree in advance “which diagnostic criteria and metrics will be used to elicit action” (LL2-17, p. 423). - Decisions should build in re-evaluation (LL2-02, p. 35) and remain provisional (LL2-26, pp. 624, 639). LL2’s cycle includes “effectiveness evaluation of actions/inactions” (LL2-27, Fig. 27.2, p. 660). - Vigilance fades after a crisis in a “homo-illogical cycle” (LL2-28, p. 680, extending the flood chapter’s “hydro-illogical cycle”, LL2-15, pp. 360–361). The flood author credits codifying preparedness in law with overcoming it “in some countries” (LL2-15, p. 361). - Standards become “sticky”, so continuous improvement is preferable (LL2-06, Guidotti panel, p. 150).

Evidence and hindsight. - Monitoring as precondition. Warnings came from long-term monitoring built for other purposes: Swedish precipitation chemistry (LL1-10, p. 102), Antarctic ozone (LL1-07, p. 82). Absence of monitoring produced absence of evidence: no random BSE survey of food-chain cattle (LL1-15, pp. 163–164), no antimicrobial resistance monitoring after Swann (LL1-09, p. 94), MTBE not routinely monitored in EU groundwater in 2001 (LL1-11, p. 112). - Hindsight on monitoring. - EU active testing from 2001 found BSE where it had been assumed absent [H: LL1-16]. - Monitoring built after the growth-promoter bans (DANMAP, Svarm) produced the evidence that later judged them [H: LL1-09]. - Atmospheric monitoring caught illegal CFC-11 production [H: LL1-17]. - Radiation surveillance units closed when no need was perceived [H: LL1-03]. - Review and exit. After a precautionary decision, neither side may produce the evidence that could settle it (hormones, LL1-14, p. 153). Hindsight: - The hormones case has not been reassessed since 2007 [H: LL1-14]. - Reversal of mistaken measures took decades (saccharin, irradiation) [H: LL2-02]. - BSE measures were relaxed through open, costed review (“precaution needs exit criteria”) [H: LL1-15]. - The Commission’s 2025 proposal to make most pesticide approvals unlimited in duration would weaken scheduled re-examination [H: LL2-16]. - Revised reference points can redefine recovery: northern cod’s 2023 upgrade came from a lowered limit point, not more fish [H: LL1-02; H: LL2-17]. - The forgetting cycle. Post-Fukushima independent regulators kept their vigilance while political direction drifted [H: LL2-28]. A pre-set adaptive rule did appear: EU provisional prohibition of high-risk plant imports pending assessment (2019) [H: LL2-20].

Analysis. Monitoring is the least contested governance lesson in the reports and one of the best supported. The reports are much thinner on the adaptive machinery that should connect monitoring to action (pre-agreed triggers, scheduled review, exit criteria); these are mostly asserted in LL2. Hindsight adds the missing half: measures in both directions persist by inertia, so the lens should ask for exit rules as well as entry rules. The “homo-illogical cycle” is an extrapolation from floods. The better-supported form is narrower: vigilance decays unless it is lodged in institutions with legal mandates [H: LL2-28].

Strength: strong for monitoring as a precondition (10+ cases). Asserted for pre-agreed triggers in the reports, supported logically and by a few later designs. Hindsight lesson (moderate) for exit criteria and review.

11. Courts, liability and compensation#

The reports say. - Cranor’s chapter argues that ex ante regulation comes “first and foremost”, backed by protection for early warners and compensation that does not require proving fault (LL2-24, pp. 582–587). Tort is “a poor legal model” (p. 589). Post-Daubert admissibility rules “asymmetrically hamper plaintiffs” (p. 588). Pre-funded assurance bonds would create incentives to research risks (pp. 600–603). - The asbestos chapter calls for speedy, anticipatory compensation (LL1-05, pp. 60–61). The vinyl chloride authors pair no-fault compensation with ways to extract corporate information (LL2-08, p. 179). The PCE author makes liability the “acid test” of whether a mechanism makes a company care (LL2-04, pp. 86, 88). - LL2’s conclusion proposes pre-funded no-fault schemes and liability bonds, and notes that accountability has “relied mainly upon” document discovery in civil cases (LL2-28, pp. 679–680).

Evidence and hindsight: courts as enablers. Ethyl Corp. v EPA (1976) upheld lead phase-down under a “precautionary statute” before the decisive low-dose evidence (LL2-03, p. 60). A US appeals court upheld the 1974 vinyl chloride standard on “the frontiers of scientific knowledge” (LL2-08, Box 8.1, p. 187). The WTO upheld the French asbestos ban (LL1-05, p. 57). France’s highest administrative court held laboratory and field bee studies both legitimate and twice told the minister to reconsider: “it was not a scientific institution” that decided (LL2-16, pp. 378–380). UK asbestos case law relaxed causation and extended foreseeability to classes of harm (LL2-24, pp. 591–593). Hindsight adds Pfizer and Alpharma (2002), Bayer v Commission (2021: “an exhaustive risk assessment cannot be required”), PAN Europe (2023: emergency authorisations cannot override a treated-seed ban) and French Constitutional Council rulings that allowed only narrow, time-limited neonicotinoid derogations (2020, 2026) and struck down a broader one (2025) [H: LL1-09; H: LL2-16].

Evidence and hindsight: courts as constraints. The 1980 Benzene decision (LL1-04, p. 40). The 1985 EU hormones ban was annulled on procedure, and the 1974 US DES reinstatement followed a procedural ruling (LL1-14, pp. 149–150) [H: LL1-16]. In Nunavut a court would not “become an academy of science to arbitrate conflicting scientific predictions” (LL2-17, p. 420). The French criminal route against Gaucho stalled (LL2-16, p. 380), ending in dismissal on causal uncertainty according to a party source [H: LL2-16]. Hindsight adds Japanese rulings that the Fukushima tsunami was not legally foreseeable (acquittals final 2025) [H: LL2-18] and the plaintiffs’ loss in Milward (2016) [H: LL2-24].

Liability and compensation. - In the cases. Liability deterred admission rather than prompting protection: Monsanto’s plan feared “liability to soar” (LL1-06, p. 65), and Brush Wellman called its standard “fundamental to our product liability defense” (LL2-06, p. 137). Insolvency shifted costs to society (Manville; LL2-25, p. 612). At Minamata, the prefecture issued debt to support the polluter’s payments, and settlements paid relief without recognition (LL2-05, pp. 107–110). DBCP settlements averaged USD 1,500 per worker in one case; litigation disclosed export records and was also used as a SLAPP (LL2-09, pp. 208–210). Nuclear operator liability is capped far below accident costs (LL2-18, pp. 445–446). Litigation discovery was the main window onto internal knowledge (LL2-07, p. 152; LL2-08, p. 179; LL2-28, p. 680). - Hindsight. Fukushima’s official cost of ¥21.5 trillion is about 100 times the European cap, and uncapped TEPCO still needed a state rescue [H: LL2-18]. The Manville Trust pays 5.6% of scheduled value [H: LL2-25]. The UK needed a levy-funded scheme (2014) for mesothelioma victims with no solvent defendant [H: LL2-24]. The revised EU Product Liability Directive (2024) extends the long-stop for latent injury but keeps the development-risk defence [H: LL2-28]. No pre-funded no-fault scheme or worst-case bond for a novel technology was found [H: LL2-24; H: LL2-28]. US states settled with Monsanto over PCBs decades later [H: LL1-06], and MTBE producers were held liable [H: LL1-11].

Analysis. Courts do not have a consistent precautionary or anti-precautionary tilt. They act on the legal standard they are given. Where a statute is precautionary and the regulator gives reasons, courts have upheld action under uncertainty. Where the standard demands quantified risk or individual foreseeability, courts have constrained or acquitted. Liability, by contrast, arrives late, is defeated by latency and insolvency, and in several cases gave firms a motive to deny rather than a motive to prevent. Its strongest documented function is disclosure.

Strength: strong that courts cut both ways and that the legal standard decides the outcome (15+ episodes). Moderate that liability deters weakly and late. Asserted or untested for no-fault schemes and bonds (Cranor’s own chapter concedes deterrence is “modest”, pp. 599–600).

12. The reports’ governance recommendations and what became of them#

Analysis. Reforms about information (disclosure, registration, uncertainty statements, standing to challenge) advanced furthest, usually after a specific scandal. Reforms that reallocate money or power (no-fault funds, bonds, tax shifts, participatory framing, research rebalancing) barely moved. Several adopted reforms were later diluted or reversed (the French alert commission, the NNI safety share, Japan’s lifetime rules, the proposed EU move to unlimited approvals). The reports cannot be credited with causing most adoptions [H: LL1-16].

13. Counter-evidence, complications and critiques#

1. Selection on outcome and hindsight. All LL1 cases and most LL2 cases are confirmed harms. Governance failures are therefore visible, and governance that correctly resisted false alarms is invisible. LL2’s attempt to fill the gap (4 genuine false positives in 88 alleged cases) used an asymmetric threshold, a critics’-list denominator, and excluded market and de facto effects (critiques.md §5.3). Some warnings the reports amplified have weakened, notably mobile phones and brain tumours [H: LL2-21] and the GM health claim in LL2’s conclusion [H: LL2-28]. The EEA itself withdrew from the IARC mobile-phone meeting (LL2-21, fn 15, p. 520), placing the agency inside a controversy it later narrated.

2. The comparative claims are weak. Relative precaution varies by risk, not jurisdiction [H: LL1-16]. The Ireland–Austria institutional comparison is confounded [H: LL1-15]. Dual-mandate bias is shown in only two of the five mandates the business chapter lists (LL2-25 digest).

3. The reports’ own internal counterpoint. The hormones chapter describes an EU precautionary ban taken against two expert committees, driven “principally” by public concern, whose expert advice was left unpublished, and which produced trade sanctions without demonstrated health benefit (LL1-14, pp. 150, 153–154). LL1’s synthesis did not use it (LL1-17 digest caveats). It is the clearest case in either report of precaution’s legitimacy costs.

4. Risk–risk trade-offs and regrettable substitution appear in the reports’ own cases: MTBE as a mandated substitute (LL1-11), HCFCs and HFCs (LL1-07), booster biocides (LL2-12) and withdrawal of DDT before substitutes were proven (LL2-11, p. 250). Hindsight adds rebounding therapeutic antibiotic use after the growth-promoter ban [H: LL1-09] and Fukushima evacuation deaths [H: LL2-18]. LL2’s false-positive analysis files trade-offs as “mistaken” false positives rather than precautionary error (critiques.md §3.2).

5. Participation and values cut both ways. Value-driven outcomes can rest on factual error (Brent Spar), and formal recognition of values can channel conflict without resolving it (GMO opt-outs) [H: LL1-17]. Participation that is ignored may lower trust [H: LL1-17].

6. Openness can be weaponised. The tobacco industry lobbied for laws making data from publicly funded studies available through freedom-of-information requests (LL2-07, p. 164). Grandjean warned that openness cuts both ways (LL2-26, p. 639). Hindsight confirms it: the US EPA “transparency” rule (2018–21, vacated) and Executive Order 14303 (2025) use transparency and rigour language to discount epidemiology [H: LL2-07; H: LL2-26].

7. Courts’ precaution is narrower than the reports’. Pfizer requires a non-hypothetical, scientifically grounded risk and reasons of commensurate scientific quality for departing from experts [H: LL1-09]. The reports’ “reasonable grounds for concern” is legally workable. Action on suspicion alone is not.

8. Legal vagueness. Marchant’s critique that open-textured principles produce arbitrary application is not answered by the EEA definition, whose key terms are placeholders (critiques.md §§3.3, 9.1).

9. The innovation counter-frame. The reports’ claim that precaution stimulates innovation holds only in the weak Porter sense [H: LL1-16; H: LL1-17]; the “innovation principle” and the 2024 Draghi report frame precaution as a drag. Neither side has counterfactual evidence (critiques.md §7) [H: LL2-27].

10. Author positionality. Many governance chapters were written by protagonists: Michaels headed OSHA (LL2-06); Bingham ran OSHA during the DBCP rulemaking (LL2-09); Bridges advised the EU at the WTO (LL1-14); the invasive-species authors run institutions they describe (LL2-20); a Commission litigator wrote the hormones update (LL2-A3, pp. 736–737); Cranor was an undisclosed plaintiffs’ expert in Milward [H: LL2-24]. Charges of intent (“covertly subordinated”, “manufactured doubt”) should be carried forward only where contemporaneous documents support them.

11. Political will versus information. LL1’s preface judges lack of political will “an even more important factor” than trusted information (LL1-00, p. 4), yet the remedies are mostly informational, and LL2 puts unequal political power “well beyond the scope” (LL2-28, p. 672). The reports diagnose power but prescribe information. Later asbestos governance supports the political-will reading [H: LL1-00], but it was never tested comparatively.

14. Technology-neutral diagnostic questions#

Each question can be asked of any emerging technology and the organisations developing, promoting and overseeing it. The evidence behind each is in the section indicated.

1. Which version of “precaution” (or of “responsible” or “safe”) is actually in force? What triggers action, how strong must evidence be, do cost qualifiers apply, and is the label matched by any change in practice? (Section 1: Table 1.2, LL1-00, p. 14; “precautionary gloss”, LL1-02, p. 24; LL2-13, p. 296.)

2. Who sets the evidential threshold for acting, and who bears the cost if it is set wrongly? Is the choice explicit and reasoned, or made by default (no opinion, no decision)? (Section 2: LL1-17, p. 193; LL2-27, pp. 657–658; glyphosate default [H: LL1-17].)

3. Who wrote the question the assessors are answering, and what did it exclude? Were intended use, single agents, sponsor data, average subjects or one endpoint written into the scope? Would a differently framed question give a different answer? (Section 4: LL1-14, p. 150; LL2-03, p. 68; LL2-16, p. 379.)

4. Where assessors disagree on the same evidence, do they explain why? Are evidence-admission rules, weights and uncertainties published? (Section 4: LL2-04, pp. 84–85; LL2-10, pp. 221–223; LL2-28, p. 677.)

5. Does any body both promote the technology and oversee its hazards, formally or through budgets, careers or national strategy? Has independence won after an earlier crisis been drifting back? (Section 5: LL1-15, pp. 157, 165; LL2-06, p. 132; LL2-18, pp. 441–443; LL2-22, pp. 546–548; [H: LL2-18].)

6. Who generates the evidence regulators rely on, and can anyone else check it? Must all commissioned studies be registered in advance? Are raw data open? Is there funded capacity for independent verification? (Section 5: LL1-16, p. 179; Transparency Regulation and Blaise [H: LL1-16].)

7. Have developers or officials made categorical safety claims that would make later, graded precaution look like an admission? Are private caveats stronger than public statements? (Section 6: LL1-15, pp. 161–162; LL1-06, p. 65.)

8. Who is close enough to the technology’s effects to notice harm first (workers, users, frontline inspectors, affected communities), and is there a channel that treats their reports as data rather than noise? (Section 7: LL1-05, p. 53; LL2-09, pp. 204, 211; LL2-05, pp. 105–106; LL1-15, p. 162.)

9. Do participation processes influence framing and outcomes, or only communication? Is there evidence that participants’ recommendations were acted on? (Section 7: LL1-16, p. 188; LL2-27, pp. 659–660; [H: LL1-17].)

10. Are protective commitments backed by enforcement, measurement, funding and deadlines, or do they rest on voluntary codes, conditional approvals, open-ended exemptions and process targets? (Section 8: LL1-05, pp. 54, 56; LL2-03, pp. 53, 56; LL2-11, p. 252; LL2-20, p. 498; [H: LL2-15].)

11. Does the governing institution’s reach match the reach of the technology’s effects? Can activity move to less regulated jurisdictions, and can any single party veto coordinated action? (Section 9: LL1-13, p. 142; LL2-09, pp. 207–211; Rotterdam chrysotile [H: LL2-A3].)

12. What independent monitoring exists now, before any need is perceived, and is it tied to pre-agreed thresholds for action? Would the absence of evidence of harm reflect the absence of a search? (Section 10: LL1-03, p. 36; LL2-12, p. 274; LL2-17, p. 423; LL1-15, pp. 163–164.)

13. Are there scheduled reviews and stated exit criteria in both directions? What would lift a protective measure, and what would trigger one? (Section 10: LL2-02, p. 35; LL1-14, p. 153; BSE exits and saccharin reversal [H: LL1-15; H: LL2-02].)

14. If harm emerges decades from now, who will pay? Are there liability caps, insolvency routes, development-risk defences or state backstops that move tail costs onto the public? Would the legal standard (foreseeability, proof of causation) let anyone be held responsible? (Section 11: LL2-24, pp. 588–603; LL2-18, pp. 445–446; LL2-25, p. 612; [H: LL2-18].)

15. Are the proposed protective actions themselves appraised for substitute risks, forgone benefits, distributional effects and reversibility? Are the alternatives appraised with the same rigour as the incumbent? (Sections 3 and 13: LL1-11; LL2-12; LL1-14, pp. 153–154; [H: LL2-02; H: LL2-18].)