T04. Innovation, technological trajectories, lock-in and alternatives#

Cross-cutting thematic synthesis, Strand A. Sources: the two EEA Late lessons from early warnings reports (LL1 = EEA 2001; LL2 = EEA 2013), read through the project’s digests, audited notes, hindsight files (checked to September 2026) and the external critiques and context files. Page numbers are report pages. Section ids follow the project scheme (for example LL1-11 = LL1 Chapter 11).

How to read this file#

Three voices are kept apart throughout. - Reports: what the chapters say or show, with section id and page. - Evidence / hindsight: what the case record and post-publication developments show, taken from the hindsight files. The hindsight files are the project’s own checks, several made without general web search; their caveats carry over. - Analysis: my synthesis across cases. It is marked as such and should not be cited as a report finding.

Strength ratings apply to each pattern as a transferable claim: - Strong: several cases across different domains, well documented, and confirmed or reinforced by later evidence largely independent of the authors. - Moderate: several cases, but with gaps, hindsight risk, protagonist sourcing, or mixed later evidence. - Suggestive: one or two cases, or a plausible mechanism with thin support. - Asserted: stated as a principle or prediction without evidence in the reports.

Standing caveats for this theme. The case sets are selected on the outcome: every LL1 case and nine LL2 Part A cases were chosen because harm happened (LL1-00, pp. 11–13; LL2-00, pp. 9–10). Many chapters were written by protagonists. The synthesis chapters that carry most of the innovation claims (LL1-16, LL1-17, LL2-27, LL2-28) were written by editors whose framework (Stirling’s ESTO project, STS work on lock-in and diversity) predates the sorting of cases into lessons (LL1-16, p. 168). The innovation layer is the one most often argued and illustrated rather than measured. One disclosure bears on this file: LL2-22 (nanotechnology), used below, was co-authored by Andrew Maynard.

1. Bottom line#

Analysis, in one paragraph. The most robust contributions of the two reports to this theme are about mechanisms of trajectory: prized properties that are also hazards, substitution within the same chemistry or operating principle, and several distinct kinds of lock-in. These are documented across many cases and later events have mostly strengthened them. The reports’ more optimistic claims are weaker: that precaution stimulates innovation, that diversity insures against surprise, and that agroecology outperforms. These rest on illustration, on the editors’ own framework and on literature about environmental regulation in general, and later evidence supports only a weak “redirects, not stops” version. The reports under-weight three things later evidence makes prominent: transition costs concentrated on particular groups, the fragility of exits compared with the stickiness of lock-in, and the fact that trajectories often change for reasons unrelated to hazard.

2. What the reports claim about innovation#

Reports (LL1). - The Preface sets the aim of learning to prevent harm “without stifling innovation or compromising science” (LL1-00, p. 11) and asserts that wider precaution “can help stimulate” innovation and science (p. 4). Chapter 1 recalls that the German Vorsorgeprinzip included clean production and innovation, alongside monitoring and proportionality (p. 13). - Lesson 7 asks decision-makers to “evaluate a range of alternative options … and promote more robust, diverse and adaptable technologies so as to minimise the costs of surprises and maximise the benefits of innovation” (LL1-16, pp. 168–169, elaborated p. 177). - The synthesis chapter argues that curtailing one option can “foster and intensify innovation in other areas” and give leading countries “a competitive edge” (LL1-16, p. 182, citing asbestos, halocarbons, PCBs and antimicrobials). It adds that technologies “lock in” early, sometimes for “arbitrary reasons” of “chance and first-leader advantage”, and that surprises are smaller with several competing technologies than with “one, global, near monopoly” (pp. 186–187). It recommends constructive technology assessment and technology options analysis (pp. 186–187). Unpriced harms, it says, gave asbestos, halocarbons and PCBs “an unjustifiable advantage in the marketplace” that kept “technically superior substitutes” out (pp. 176–177). - The Conclusions put technological diversity and fewer “monopolies” in the “precaution” row of Table 17.1 (LL1-17, p. 192). They concede that “over-precaution can also be expensive, in terms of lost opportunities for innovation”, and call “maximising innovation whilst minimising hazards” “ultimately a matter of political discourse” (p. 194).

Reports (LL2). The subtitle becomes “science, precaution, innovation”. - Careful precaution can stimulate innovation (LL2-00, p. 10). Even the four genuine false positives “sparked innovation”, and Lesson 6 urges measures “that can spark innovation even if the precautionary action proves unnecessary” (LL2-02, pp. 32, 35). - Gee gives precaution a second role as “a trigger for broader debates about technological pathways” (LL2-27, pp. 644, 662), lists “potential for stimulating innovation” among criteria for action (Box 27.4, p. 653), and states that “late actions have consolidated technological monopolies for products, at unrealistically low prices, which served to keep smarter substitutes out of the markets for many years” (p. 659). Box 27.5 (von Schomberg, personal capacity) criticises treating innovation as “steerless but inherently good” (p. 661). - The concluding chapter asserts “increasing evidence that precautionary measures do not stifle innovation” (LL2-28, p. 670), names a “moving target” (“today’s technology is now safe”) and “de facto technological lock-ins” from sunk investment (p. 672), links “monopolies” of lead in petrol, asbestos, CFCs and PCBs to technological, institutional and ideological lock-in and to “artificially low market prices” (p. 673), and asks that substitutes be screened (p. 676).

Evidence / context. The reports move from judging when to act on warnings toward arguing about the direction, diversity and democratic control of innovation. Neither cites Collingridge, but the timing-and-lock-in problem runs through both (context.md §2.3–2.4).

Analysis. The innovation thesis is stated most strongly where case evidence is thinnest: prefaces, introductions and syntheses. Case chapters touch innovation in passing (LL2-08, p. 192; LL1-14, p. 150; LL1-05, p. 59; LL2-03, p. 69), and several say nothing about substitutes (LL2-09, LL2-10, LL1-12). This file treats the thesis as claims to test, not as a finding.

3. P1. The prized property is often the hazardous property#

Reports. Across cases, the feature that made a technology attractive was the same feature, or a close correlate, that later caused harm. - PCBs were adopted as less flammable, more stable and “thought to be safer” (LL1-06, p. 64); stability made them persistent and hard to destroy (p. 72). - CFCs were adopted as efficient, non-toxic, non-flammable refrigerants (LL1-07, p. 79); “Short-term safety appears to demand” inertness, which means persistence (p. 83; also LL1-16, pp. 169–170). - DDT’s persistence was presented as a practical advantage in the 1948 Nobel presentation speech; it is “today considered as one of the main environmental concerns” (LL2-11, p. 241). - Asbestos, the “magic mineral”, was valued for durability. Its fibre form and durability in tissue are what make it carcinogenic, and substitutes that “reproduce the same physical form” are likely to be carcinogenic too (LL1-05, pp. 52–53, 61). - TBT was “extremely effective and relatively economical” (LL1-13, p. 135) precisely because it killed fouling organisms. Its boosters’ “undeniable (and necessarily) toxic nature” is the point of the product (LL2-12, pp. 274, 276). - Ethinyloestradiol was valued for stability and oral potency, properties that let it pass through sewage treatment (LL2-13, pp. 279–281). - MTBE was chosen because it was cheap, blended easily and moved through existing pipelines (LL1-11, p. 110). Its high solubility and mobility are what carried it through groundwater (pp. 111–112). - Imidacloprid seed dressing was marketed as confined to the soil (LL2-16, p. 371); systemic uptake and soil half-lives of 188–249 days (p. 376) created exposure routes spray-era tests missed.

Evidence / hindsight. - Persistence as a warning sign was strongly vindicated. PFAS, which the reports flag only as emerging (LL2-26), became the main post-2001 example (LL1-16 hindsight). - The EU created hazard classes for persistent, mobile substances (PMT/vPvM) in 2023, without a toxicity criterion. That is close to the MTBE problem, and Anses (2023) found that MTBE meets the vPvM criteria (LL1-11 hindsight). - The pattern recurred in the next generation of halocarbon substitutes: HFOs degrade to the very persistent trifluoroacetic acid (TFA) (LL1-07 hindsight). - Qualification: not every later surprise turned on persistence (LL1-16 hindsight cites an acutely toxic transformation product of a rubber antioxidant), and lipid-based bioaccumulation screens missed mobile substances.

Cases supporting: eight (above), across chemicals, pesticides, pharmaceuticals and materials. Cases cutting against or outside the pattern: benzene, lead, BSE, DBCP and vinyl chloride, where the hazard was not the prized property; DES, where potency was both the rationale and the hazard but the benefit never existed (LL1-08).

Rating: strong. Many cases, and regulatory uptake independent of the authors (persistence and mobility criteria). Analysis: the lesson is about trade-offs, not rejection. CFC inertness did remove acute toxicity; PCB stability did bring fire safety (a benefit LL1-06 does not weigh, LL1-06 notes); DDT persistence made indoor spraying work. The transferable point is that a design virtue at the point of use can be a liability at system scale and over time, so the virtue itself should trigger scrutiny.

4. P2. Enthusiasm and conspicuous benefit displaced appraisal; benefits were often weaker than claimed#

Reports. - Radiation: excitement, hype and real medical value meant “caution tended to be thrown away” (LL1-03, p. 31). Recommendation-only rules allowed “ill-conceived” uses such as shoe-shop fluoroscopes (p. 34). - DES: cheap, unpatented, sold under more than 200 brand names, promoted for “routine prophylaxis in all pregnancies”, with a premium on modernity (LL1-08, pp. 84, 86, 88). Randomised trials showed in 1953 that it did not work; prescribing continued for about two decades. - Leaded petrol: industry told the 1925 conference that TEL was essential to industrial progress and that “all innovation entails risks”; Ethyl’s president called it an “apparent gift of God” (LL2-03, pp. 52–53). - Antibiotic growth promoters were “readily adopted” and became integral to industrial husbandry (LL1-09, p. 93). The 2013 update reports trials showing small or no benefit (LL2-A3, pp. 731–733). - MTBE “apparently promised a simple solution” to lead, so its water problems were “perhaps more readily overlooked” (LL1-16, p. 176). Its specific air-quality benefit was small and shrinking (LL1-11, p. 111). - Neonicotinoid seed dressings were used preventively “regardless of the presence and abundance of pests” (LL2-16, p. 384). - GM crops: broad promises narrowed to two traits for the largest markets (LL2-19, pp. 463, 466). - Nanotechnology: “nano-fever” produced “solutions looking for a problem”; the question is whether it “should” be used, not only whether it “could” (LL2-22, pp. 545–546). - The editors’ Lesson 6 generalises this: scrutinise claimed benefits alongside risks (LL1-16, pp. 175–177).

Evidence / hindsight. - Confirmed: DES never worked (LL1-08 hindsight). Danish and Swedish production grew without growth promoters, and the WHO panel put the net cost at about 1% of pig production costs (LL1-09 hindsight). Seed treatments gave “negligible” yield benefit in several crops (LL2-16 hindsight). GM traits stayed narrow (LL2-19 hindsight). - Complicated: the benefits were real and large in several cases. DDT’s malaria benefit is written into the Stockholm Convention’s acceptable-purpose exemption (LL1-17 hindsight; LL2-11). Where a seed treatment controlled a virus vector, as for French sugar beet in 2020, its removal cut national yields from 851 to 649 q/ha (LL2-16 hindsight). Publicly developed insect-resistant crops produced measurable smallholder gains (LL2-19 hindsight). The chapter on growth promoters is silent on transition costs that later proved real (LL1-09 hindsight).

Cases supporting: eight. Cases cutting against: DDT (real benefit), PCBs (fire safety unweighed), radiation (medical value), sugar-beet seed treatment, some Bt crops.

Rating: moderate. Strong for DES and growth promoters, where benefit was tested and found absent. But the reports rarely quantify benefits: LL1-16 admits a general cost-benefit analysis “lay beyond the scope” (p. 168). Several chapters give no weight to the benefits of the technology they criticise (LL1-06; LL2-10; LL2-12 on copper). Analysis: the durable lesson is procedural: benefit claims need the same independent testing as risk claims, including claims made for the preferred alternative (see P8).

5. P3. Regrettable substitution: substitutes judged against the incumbent reproduced or relocated the problem#

Reports. This is the reports’ most consistently documented innovation dynamic. The editors list “substitution traps” (LL1-16, pp. 173–177): substitutes sharing the hazardous property (asbestos-like fibres), substitutes “unduly tolerated” because they beat a worse predecessor (HCFCs), “chemical for chemical” substitution (TBT), protective measures creating new hazards (MTBE) and end-of-pipe fixes exporting harm (tall stacks). LL2-12 turns it into a five-step cycle: replace the old agent on a presumption of relative safety; monitor; find concern; ban the worst; search again (p. 273). LL2-28 calls for “hazard screening of alternatives” (p. 676).

Substitution chains in the reports, with hindsight

Mechanisms visible across the chains (Reports, with Analysis where noted). 1. Comparative-only appraisal. A substitute is judged against the incumbent’s known harm, not on its own terms (LL1-16, p. 177; LL1-11, p. 117; LL2-12, p. 265). 2. Same principle, same hazard. Swapping agents within the same functional principle (a leaching biocide, a halocarbon refrigerant, a durable fibre) tends to move harm rather than remove it (LL1-13, p. 141; LL2-12, pp. 273–276). Improvement came from a different principle: non-release coatings, hydrocarbon refrigerants, designed-out durability. 3. Incumbents choose the substitute. CFC producers preferred HCFCs and HFCs (LL1-07, p. 81); paint makers “tend to devise solutions with which they are familiar — typically more biocide” (LL2-12, p. 275). 4. Mandates scale substitutes before assessment. The US oxygenate requirement drove MTBE use (LL1-11, pp. 110–111). Hindsight: the ethanol replacement was again scaled by mandate before its multi-media assessment (LL1-11 hindsight). 5. Phase-outs that do not name a replacement. (Analysis, from hindsight.) These tend to produce substitution by the cheapest drop-in alternative (LL2-04, LL2-10, LL2-16 hindsight). 6. Neighbouring inputs absorb the function. When one input is withdrawn, adjacent ones take its role, with delayed costs of their own (LL1-09 hindsight). 7. Displacement across media and borders. Fixes that clean one place or medium can load another (LL1-10; LL1-11, p. 114; LL2-23, pp. 569, 572).

Evidence / hindsight on responses. Regulators adopted class-based rules to block like-for-like substitution (the 2024 EU bisphenols rule, the PFAS group proposal, the PFAS-foam restriction), and the US National Research Council published an alternatives-assessment framework in 2014 (LL1-16, LL2-27 hindsight). The cycle nonetheless kept running in antifoulants and insecticides (LL2-12, LL2-16 hindsight).

Cases supporting: about fourteen (table). Cases cutting against: - Substitution often did reduce harm: most asbestos substitutes are much less hazardous; hydrocarbon refrigerants now dominate domestic refrigeration; biocide-free foul-release coatings perform as well as copper in field trials (LL1-05, LL1-07, LL2-12 hindsight). - The HCFC detour’s ozone cost proved modest, and its supporters argue it enabled a faster CFC exit. That counterfactual is untested (LL1-07 hindsight). - LL2’s own false-positive chapter files risk-risk trade-offs as a mistaken false-positive claim rather than as a cost of precaution (LL2-02, pp. 24–25). Critics say this keeps substitution harms out of the ledger (critiques.md §3.2). LL2-27 does not discuss regrettable substitution at all (LL2-27 hindsight).

Rating: strong that substitutes chosen by comparison with a worse incumbent, within the same principle, and scaled quickly tend to carry their own hazards. Moderate that function-level redesign reliably avoids this: the record shows it can (coatings, refrigerants), but “non-toxic” alternatives brought chemistries of their own (fluorinated and organosilicon coatings, LL2-12 hindsight).

6. P4. Lock-in and path dependence#

Reports. The core statement is that “once a technological commitment is made, a host of institutional and market processes act to reinforce its position, even if markedly inferior to potential alternatives” (LL1-16, p. 177). LL2 adds sunk investment, “moving target” assumptions (LL2-28, p. 672) and “institutional and ideological lock-ins” (p. 673). The cases show at least seven distinct forms.

(a) Physical capital and installed stock. - Reports: PCBs stayed in long-lived electrical equipment for decades, leaking and being illegally dumped (LL1-06, pp. 66, 69–72). Asbestos in buildings will cost “further billions” to remove (LL1-05, p. 58). Equipment and foam banks kept releasing CFCs (LL1-07, p. 77). Lined water mains prolonged PCE exposure (LL2-04, p. 76). River engineering “remained influential” for a century (LL2-15, pp. 349–350, 361–362). For energy and chemical plants, “yesterday’s investments will be redeemed before any serious risk reduction is implemented” (LL2-28, p. 672). - Hindsight: the US EPA’s 2024 asbestos rule found eight chlor-alkali plants, about one-third of national capacity, still using asbestos diaphragms. They were 42–83 years old, conversion would cost about USD 2.8–3.4 billion, the alternatives are more energy-efficient, and EPA allowed a staggered 5-, 8- and 12-year phase-out (LL2-27 hindsight). The halon bank, meant for destruction, became a supply to conserve (LL1-07 hindsight).

(b) Price lock-in through externalised costs and market power. - Reports: asbestos prices excluded health costs, and cartels slowed substitutes (LL1-05, p. 58). Unpriced harms gave asbestos, halocarbons and PCBs “an unjustifiable advantage” (LL1-16, pp. 176–177; LL2-27, p. 659; LL2-28, p. 673). Ethyl held 70% or more of US petrol by the late 1930s, refused supply to dealers selling alcohol blends, and lost an antitrust case (LL2-03, p. 55). DDT persists as a cheap incumbent (LL2-11, pp. 243–244, 251–252). - Hindsight: DDT alternatives cost two to three times more (LL2-11 hindsight).

(c) Knowledge and skill lock-in. - Reports: once capital was committed to TEL, its developers “categorically denied the existence of alternatives” they had championed in 1921, and by the 1960s those alternatives were “forgotten” (LL2-03, pp. 54–55). Herbicide-tolerant systems are “deskilling”, and lost seed networks “may be difficult to re-establish” (LL2-19, pp. 462, 472). Husbandry without routine antimicrobials existed but was “not actively promoted” (LL1-16, p. 177).

(d) Institutional and regulatory lock-in. - Reports: mandates (the MTBE oxygenate requirement, LL1-11, pp. 110–111); harmonisation constraining national action (LL1-11, p. 111; LL1-16, p. 180); an open-ended DDT exemption (LL2-11, p. 244); test endpoints unchanged for decades (LL2-10, p. 223); spray-era tools applied to seed dressings (LL2-16, p. 377); loyalty to a fisheries model (LL2-17, pp. 411, 415). - Hindsight: removing the MTBE mandate took about seven years and legislation, and the replacement mandate locked in ethanol (LL1-11 hindsight). The US aviation-fuel lead exemption, “temporary” in 1996, still stands, and EU authorisation of TEL runs to 2032 (LL2-03 hindsight). Ozone-regime feedstock exemptions became major residual leaks (LL1-07 hindsight).

(e) Dependence and masking. - Reports: growth promoters enabled intensive systems whose problems they hid (LL1-09, p. 93 and Table 9.1; asserted). Prophylactic, bundled seed treatment created lock-in (LL2-16, pp. 383–385). Buyers threatened contract claims when a supplier tried to stop shipping DBCP, and growers called it “essential” (LL2-09, pp. 209, 211). - Hindsight: EU Member States granted 206 emergency authorisations for the restricted neonicotinoids in 2013–2019, and France cycled through derogations, court rulings and a 2026 derogation law for near-equivalents (LL2-16 hindsight). After purpose-based bans, the same antimicrobials continued under “prevention” labels (LL1-09 hindsight).

(f) Adaptive treadmills. - Reports: single-tactic control of an adaptive target breeds resistance, met by more or stacked product (LL2-19, p. 462; LL2-11, pp. 241, 243, 251). - Hindsight: 62 glyphosate-resistant weed species, 34 first recorded from 2013; resistance to each stacked replacement within a few years; resistance already emerging to the newest vector-control insecticides (LL2-19, LL2-11 hindsight).

(g) Defensive lock-in and compelled adoption. - Reports: firms lock into product defence because of liability; “there must be room for them to turn around” (Guidotti, LL2-06, pp. 149–150; LL2-27, p. 646). - Hindsight, beyond the chapters: a US appeals court vacating dicamba registrations (2020) cited farmers planting tolerant seed defensively against neighbours’ drift, an externality that compelled adoption (LL2-19, LL2-28 hindsight).

Cases supporting: fifteen or more. Complications: - Several “locked-in” options had real advantages. Beryllium kept broad uses and became a designated critical material, with substitutes that “can result in substantially reduced performance” (LL2-06 hindsight). Alcohol fuel cost about USD 2 a gallon in 1921, a problem still unsolved for cellulosic alcohol a century later (LL2-03, p. 54; hindsight). The halon bank is needed because some uses lack alternatives (LL1-07 hindsight). - The claim that winners prevail for “arbitrary” reasons (LL1-16, p. 187) comes from imported theory with no case analysis. Its best in-report illustration, TEL against alcohol, is weakened by alcohol’s cost and scale problems (LL2-03 hindsight).

Rating. Lock-in as a mechanism: strong (many cases; a clear post-2013 illustration in the US asbestos rule). “Late action consolidates monopolies that keep smarter substitutes out” (LL2-27, p. 659; LL2-28, p. 673): moderate. The lock-in half is strong (LL2-27 hindsight suggests upgrading it); the assumption that the excluded substitutes were “smarter” does not hold in general. “Winners are arbitrary”: asserted.

Analysis. The forms matter for diagnosis because they unlock differently. Capital lock-in yields to phase-out schedules and finance. Price lock-in yields to internalisation. Knowledge lock-in needs deliberate maintenance of alternatives (extension services, research). Regulatory lock-in needs sunset clauses. Treadmills need reduced selection pressure, not a new product.

7. P5. Diversity of options and technological monopoly#

Reports. - Surprises “will be smaller if there are several competing technologies … rather than just one, global, near monopoly, as was the case with asbestos, halocarbons and PCBs” (LL1-16, p. 187; see also p. 171 on “societal ignorance”). - Technological diversity is one of LL1’s strategies for acting under ignorance (LL1-17, p. 192). - Precaution helps by “promoting a diversity of technologies”, “keeping options open and following multiple paths” (LL2-28, p. 673). - “Diversity in innovation buffers against lock-ins” (LL2-19, p. 477, via the STEPS “3D” agenda). - The DDT authors prefer integrated vector management, combining methods, to single-tool reliance (LL2-11, pp. 251–252). - LL2 warns of “signs of technological monopolies” in GM crops and nanotechnology, driven by R&D costs and patents (LL2-28, p. 673).

Evidence / hindsight. - Diversity as insurance: untested. No empirical test was found of whether portfolio diversity reduced the size of surprises. The main later development is conceptual (Stirling’s 2007 diversity framework, by an LL1 editor) (LL1-16 hindsight). - Diversity of tactics against adaptive targets: supported. DDT’s exit came through a mix of resistance, new insecticides, bed nets and falling malaria, not through the UNEP road map (LL2-11 hindsight). Stacking tactics bought time, not escape (LL2-19 hindsight). - Monopoly forecast: split. For GM seeds and agrochemicals it was prescient: three mega-mergers in 2017–18, and the European Commission found reduced innovation competition among “only five” integrated R&D players. For nanotechnology it was not borne out: the technology diffused as a fragmented enabling toolkit (LL2-28 hindsight). A USDA analysis also found consolidation accompanied by more private R&D and higher farm productivity (LL2-28 hindsight).

Cases supporting: three near-monopoly illustrations (asbestos, halocarbons, PCBs), two treadmill cases, one concentration forecast borne out. Cutting against: nanotechnology; the USDA finding; no case where diversity was shown to shrink a surprise.

Rating: suggestive for diversity as general insurance (a reasoned design principle, not a finding); moderate for diversity of tactics against adaptive targets. Analysis: LL2-28 hindsight draws a useful distinction. Concentration and lock-in follow when a technology is sold as an integrated proprietary system (seed, trait and herbicide), not when it is a diffuse enabling capability. The reports treat both alike. Diversity also has costs (duplication, lost scale economies) that no chapter discusses.

8. P6. Who steers the direction of innovation#

Analysis. Steering is rarely centralised: it is the net effect of what can be owned, who owns the capital, what the state mandates or funds, and unrelated technical change. Precautionary exits often succeed by riding a co-driver, which makes them effective but contingent (LL2-03 insight 9).

9. P7. Does precaution stifle or stimulate innovation?#

Reports: the claims. Precaution “can help stimulate” innovation (LL1-00, p. 4); curtailment “may actually serve to foster and intensify innovation” and bring “a competitive edge” (LL1-16, p. 182); precaution stimulates “more innovation, via technological diversity and flexibility” (LL1-17, p. 194); even false positives “sparked innovation” (LL2-02, p. 32); “clear and challenging regulations can stimulate innovation” (LL2-08, p. 192); a ten-year phase-out of leaded petrol “at any time since 1925 would have encouraged innovators” (LL2-03, p. 69); “increasing evidence” that precaution does not stifle innovation (LL2-28, p. 670); early action stimulates innovation and late action entrenches monopolies (LL2-27, p. 659).

Reports: the evidence offered, case by case.

Evidence / hindsight: the general test. - Environmental regulation generally (not action under uncertainty): the “weak” Porter hypothesis (regulation induces innovation) is better supported than the “strong” one (Ambec et al. 2013). Induced clean-technology innovation is real but its benefits are “not … large enough to outweigh the costs” for regulated firms (Dechezleprêtre and Sato 2017). A meta-analysis of 103 studies finds “the most likely scenario is statistical insignificance” (Cohen and Tubb 2018). (LL1-16, LL1-17, LL2-28 hindsight.) - Precaution-specific evidence is mixed. The 2018 REACH review found new-substance rules “stimulated” development and authorisation fostered substitution, but that authorisation “could be harming” competitiveness. The EU adopted a lighter regime for certain new genomic techniques in 2026, explicitly for innovation. GM approval took 1,763 days on average in the EU and 2,467 in the US (1998–2015). An advocates’ model put the cost of delaying vitamin-A rice in India at about 1.4 million life-years (indicative only). (LL2-28, LL2-02, LL1-17 hindsight.) - Policy: an industry-drafted “innovation principle” entered Council conclusions (2016) and the Horizon Europe Regulation (2021), which does not mention precaution (critiques.md §7). That is a position, not evidence, but the reports’ framing lost the institutional argument.

Cases supporting the weak form: five to six (VC, growth promoters, asbestos, lead phase-down, halocarbons, hormones). Cutting against the strong form: meta-analyses; GMO approval delays; REACH’s competitiveness finding; transition costs (growth promoters, sugar beet); induced substitutes with new hazards (halocarbons, BPA, neonicotinoids).

Rating. - “Restriction redirects innovation toward substitutes and cleaner processes”: moderate (several cases, consistent with the weak-Porter literature). - “Regulated parties’ forecasts of compliance cost are often overstated”: moderate, with conditions: when a substitute or engineering fix exists, when rules leave room for innovation or trading, and when the forecast is compared like for like. - “Precaution stimulates more innovation, does not stifle it, or pays for itself”: asserted / contested. The reports rely on illustrations and on the Porter literature, which concerns environmental regulation, not precaution (a conflation LL2-28’s digest flags). The claim also comes from the same network that wrote the cases.

Analysis. The reports are right that “stifling” is the wrong frame when a restriction channels effort toward alternatives, and wrong to imply that channelled effort reliably produces better outcomes. What the record supports is conditional. Induced innovation is likeliest when a binding, dated requirement meets an available engineering or substitute pathway in a concentrated industry. It is weakest when targets are soft, incumbents are much cheaper, or the only available substitutes share the old principle.

10. P8. GM crops and agroecology: pathways, not products#

Reports. LL2-19 is the reports’ fullest innovation-pathway case. It contrasts “top-down” technology transfer (GM crops) with “bottom-up”, participatory agroecology (pp. 458, 461). - Diagnosis: incentive systems “largely determine who is innovative and what innovative products will look like” (p. 459). Seed-based bias is “an unavoidable outcome” of IP-led innovation (pp. 465–466). Herbicide-tolerant systems create a resistance “treadmill” and “deskilling” (p. 462). Concentration and public research’s drift to private logics narrow the pipeline (pp. 466–467). Developer-controlled, confidential, underpowered risk studies tilt toward “no evidence of harm” (Box 19.2, pp. 468–470). - The late lesson “may be” institutional rather than harm-based, although “early indications of harm are just emerging” (p. 470). - Claims for the alternative: “The results speak for themselves” (p. 474), citing Badgley 2007 and before-and-after project syntheses (+79%, +116%, 2.13-fold). - Hedges: “Either pathway” could cut food costs and offer a way out of poverty (p. 461); “the right mix of innovation is essential” (p. 470); GM need not “hypothetically” be high-input (fn 1, p. 460). - Elsewhere in the reports: Gee cites the French GMO debate as precaution acting as “a trigger” for debate on the future of agriculture and innovation (LL2-27, p. 644). LL2-28 asserts, uncited, that some GM crops “present a threat to human health” (p. 674). LL1 calls critiques of EU GMO assessment as too narrow “cogent” (LL1-16, p. 183).

Evidence / hindsight. - Strengthened: the treadmill (62 glyphosate-resistant species; resistance to stacked replacements; dicamba drift and court vacatur); narrow delivery (same traits, same five countries); consolidation (three mega-mergers; Commission findings of reduced innovation competition); and the institutional critique (mandatory EU 90-day studies, 2013; transparency law, 2019; the Blaise judgment; a funding effect in 40% of Bt-efficacy papers; a 2025 ghostwriting retraction of a glyphosate review). EU law now excludes herbicide-tolerant plants from the lighter new-genomic-techniques regime because they breed resistance “regardless of the breeding technique” (Reg. 2026/1388). - Weakened: the health “indications” (Séralini 2012 retracted; an EU-funded two-year study found no adverse effects; NASEM 2016); the agroecology yield claims (average organic yield gap about 19–25%, about 33% in US commercial data; Badgley 2007 overstated); the categorical smallholder prediction (a publicly developed Bt eggplant trial showed large smallholder gains). - Partly held up: recommendations adopted in rhetoric (FAO 2019; HLPE 2019; an EU partnership 2024), then eroded (EU pesticide regulation withdrawn 2024); funding about 1–1.5% of relevant budgets. (LL2-19 and LL2-28 hindsight.)

Rating. - Appropriability steering direction, treadmills, consolidation narrowing the pipeline, developer-controlled evidence: strong–moderate. - “Who develops and delivers a technology shapes its distribution” (a hindsight lesson from public-sector Bt crops): moderate. - The health warnings and agroecology yield claims: weak; the chapter applied asymmetric standards. - The rival explanation for patchy adoption, that restrictive regulation and fear of losing export markets contributed (LL2-19 notes, Paarlberg), is not engaged. If it is right, precautionary regulation itself shaped the adoption pattern the chapter reads as rejection on the merits.

Analysis. The chapter is most useful as an account of how innovation systems select problems, not as an evaluation of a technique. Its own concessions (fn 1; “right mix”) point there, and hindsight vindicates that layer while undercutting the harm and yield claims.

11. P9. Narratives of progress, essentiality and “problem solved”#

Reports. Framing recurs as a steering device. - Progress and competitiveness. Industry argued at the 1925 conference that leaded petrol was essential to progress and that “all innovation entails risks” (LL2-03, p. 52). Ethyl’s president made critics “appear to be reactionaries who were retarding human progress” (authors’ reading, p. 53). A consultant wrote privately that “human progress cannot go on under such restrictions … if we are to survive among the nations”, adding that the argument could apply to “the thousand and one other poisons” (p. 53). That is a national-competitiveness frame plus a slippery slope. - Essentiality and “no alternative”. TEL was “the only material available” (1925, LL2-03, p. 54). Asbestos was “irreplaceable” (Lancet 1967, LL1-05, p. 58). DBCP was “essential” to growers (LL2-09, p. 211). Imperfect alternatives were used to resist substitution, which LL1-13 calls “a somewhat negative argument” (p. 141). - “Problem solved” and the moving target. Asbestos disease was repeatedly said to reflect superseded conditions (LL1-16, p. 173). Partial TBT controls were “hailed by some as a ‘solution’” (LL2-12, p. 267; LL1-13, p. 138). “Today’s technology is now safe” (LL2-28, p. 672). - Innovation as product flow. EU policy defines innovation as ideas “turned into new marketable products” (LL2-19, p. 461). “How innovation is conceived shapes how it is promoted, and who benefits” (p. 461). Innovation is “assumed to be steerless but inherently good” (von Schomberg, LL2-27, p. 661). Framing problems as technological directs innovation “toward technological products” (LL2-19, p. 477). A “tragedy of the commons” framing narrowed fisheries solutions to fleet cuts (LL2-17, p. 419). - Precaution framed as anti-innovation. Bayer’s panel warns that suspensions risk “stopping innovation” (LL2-16, pp. 401–402). The editors reply that precaution is not “blanket opposition to innovation” (LL1-16, p. 169) and that discouraging one pathway differs enormously from “channelling” innovation elsewhere (p. 182).

Evidence / hindsight. “No alternative” claims recur and sometimes win. The TEL producer still states there is “no currently available alternative” for aviation fuel despite approved alternatives (LL2-03 hindsight). US exemptions in 2025–26 accepted that compliance technology “is not available” (LL2-08 hindsight). “Essentiality” hardened into strategic designation for beryllium (LL2-06 hindsight). The critics’ counter-frame, the “innovation principle”, became EU procedure; the reports’ own is Stirling’s “steering, not stopping” (critiques.md §7).

Cases supporting: eight to ten instances across six cases. Cutting against: some essentiality claims were substantially right (beryllium performance; halons for some uses; DDT where no equivalent existed, LL2-11, p. 252).

Rating: moderate. The instances are well documented, often from contemporaneous records, but how often such narratives decided outcomes is not measured. Analysis: the useful test is not whether a claim of essentiality is made but whether it is checked against ex post outcomes, as the vinyl chloride chapter did for 1974 cost forecasts (LL2-08 hindsight).

12. P10. Scale and speed of deployment outran appraisal#

Reports. - MTBE became the third most produced organic chemical in the US by 1995, before its fate in water was assessed. It escaped pre-market rules as a pre-1981 “existing” substance (LL1-11, pp. 110, 116). - Halocarbon releases grew to hundreds of kilotonnes a year (LL1-07, p. 82); BPA production reached 3.8 Mt a year (LL2-10, p. 216); PVC grew from 450 t (1933) to about 34 Mt (LL2-08, pp. 180–181). - Emerging technologies are “largely unknown, yet already widespread” (LL2-00, p. 10); scale “puts very difficult demands” on monitoring (LL2-28, p. 672). - The lag between nano products and regulators’ use of safety data “will not be less than two decades” (LL2-22, p. 549), a figure overstated from its sources (LL2-22 notes). - Lesson: deployment scale should itself trigger scrutiny (LL1-11 insight 3; LL2-02 lesson 4, p. 35).

Evidence / hindsight. REACH closed the grandfathering gap, and tonnage plus dispersive use triggered MTBE’s re-evaluation (LL1-11 hindsight: strengthened). But the US scaled ethanol by mandate before its multi-media assessment, even after the lesson had been written up (LL1-11 hindsight). The two-decade lag was about right for horizontal law and too pessimistic for targeted instruments (LL2-22 hindsight).

Rating: strong for the case-level lesson that high-volume, dispersive use of a persistent or poorly characterised option should trigger appraisal; moderate as a general claim about emerging technologies.

13. Counter-evidence, complications and critiques#

1. Selection on the outcome and hindsight. Every LL1 case was chosen because harm materialised (LL1-00, pp. 11–13). So “celebrated technology later caused harm” is built into the sample. It cannot tell us how often celebrated technologies do so. Critics’ strongest point is that the reports support mechanism claims but not frequency or expected-value claims (critiques.md §4, §9.1; Marchant 2003). The same applies to lock-in: we see lock-in to harmful options, not lock-in to beneficial ones that later proved hard to improve.

2. The emerging-issue warnings often did not come true. Part C applied the “wonder then harm” template prospectively. GM food health concerns weakened sharply (LL2-19, LL2-28 hindsight). The mobile-phone brain-tumour warning is “the report’s clearest candidate for an early warning that has so far largely not been borne out” (LL2-21 hindsight). Nanotechnology has “no realised harm to test outcomes against”, and the feared monopoly did not form (LL2-22, LL2-28 hindsight). A lens built on these reports must allow for false alarms, which the reports’ own false-positive analysis (4 of 88, LL2-02) addressed only partly and from within the same network (LL1-17, LL2-02 hindsight; critiques.md §5).

3. Substitution is not always regrettable, and restriction has transition costs. Most asbestos substitutes, hydrocarbon refrigerants and biocide-free coatings improved outcomes (§5). Meanwhile, withdrawing an embedded input produced acute adjustment burdens concentrated on the most vulnerable part of the system: weaner pigs after the growth-promoter ban, sugar beet after the neonicotinoid ban (LL1-09, LL2-16 hindsight). The reports rarely plan for these costs. The DDT chapter’s own evidence, that withdrawing an incumbent before a substitute was proven “probably increased the threshold of expectation of proof” for alternatives (LL2-11, p. 250), is a warning that premature exits can set substitution back.

4. Risk-risk trade-offs are under-weighted. Graham and Wiener, Goldstein (who uses MTBE, an LL1 case) and Sunstein argue that precautionary measures create countervailing risks (critiques.md §3.1–3.2). The reports document such trade-offs (MTBE, tall stacks, halocarbon substitutes), but frame them as failures of narrow appraisal. LL2-02 classifies them outside the false-positive count. The removal of sulphate unmasking warming (LL1-10 hindsight) is a large example the reports did not anticipate.

5. The innovation thesis borrows evidence from elsewhere. The reports’ “increasing evidence” (LL2-28, p. 670) is the Porter literature on environmental regulation, which later meta-analyses place near null (§9). Precaution-specific evidence is thin on both sides: the reports offer illustrations, and the innovation-principle campaign offers macro comparisons and selected cases (critiques.md §7). Documented innovation costs of prolonged precautionary gatekeeping exist (GM approval times) (LL2-02 hindsight). LL2-24 concedes that assurance bonds “may overly burden new technologies” (p. 601).

6. Asymmetric scrutiny inside the reports. LL2-19 subjects GM evidence to forensic standards while accepting uncontrolled agroecology syntheses (LL2-19 notes; hindsight lesson 11). LL2-03 calls alcohol “equally effective” and cleaner (p. 46) while its own Panel 3.1 reports that it “would cost two dollars per gallon” (p. 54). The reports’ rule that benefit claims deserve scrutiny (Lesson 6) was not applied to the preferred alternatives.

7. Lock-in is not always irrational. Some incumbents persisted because alternatives were genuinely worse or costlier (beryllium, halons, DDT in some settings, alcohol fuel in 1925). “Arbitrary” winners (LL1-16, p. 187) is asserted, not shown.

8. Firms differ; the business account selects on failure. LL2-25’s claim that continuing was perceived as profitable “in virtually all reviewed cases” (p. 607) is close to built into a failure sample. Hindsight found within-sector variation: a refiner that declined MTBE, and a beryllium producer that co-drafted a stricter limit (LL2-25, LL2-06 hindsight). Monsanto restricted open PCB uses as sole supplier in 1970 (LL1-12, p. 128). The one voluntary “success” the reports cite, Manville’s fiberglass labelling, was a false positive (LL2-25, p. 615).

9. Exits are fragile; lock-in is sticky. (Analysis, from hindsight.) Precautionary exits and commitments to alternatives proved politically reversible: neonicotinoid derogations, the EU’s GMO reform, the withdrawn pesticide regulation, post-2011 nuclear phase-outs reversed outside Germany, and US retrenchment in 2025–26 (LL2-16, LL2-19, LL2-18, LL2-08 hindsight). Incumbent capital, by contrast, persisted for decades (chlor-alkali asbestos, PCB equipment, aviation lead). The reports assume that once lessons are learned they stay learned. LL2-28 itself describes a “homo-illogical cycle” of fading vigilance (p. 680), but it does not apply this to innovation governance.

10. Standpoint. The synthesis chapters carrying the innovation claims are advocacy by the project’s editors (LL1-16, LL2-27, LL2-28 notes). The GM chapter’s lead author sat on the editorial team (LL2-19 notes). The EEA’s 2013 restatement that precaution “can encourage” innovation is the same organisation repeating its thesis, not independent confirmation (LL1-17 hindsight).

11. What survives. Critics concede the mechanism claims, including lock-in of incumbent products (critiques.md §4, §9.1). The mechanisms (P1, P3, P4) therefore outweigh the predictions (P5, P7) and preferred pathways (P8).

14. Technology-neutral diagnostic questions#

Each question is tied to the evidence above and can be asked of any emerging technology and of those developing, funding and governing it.

1. Is the property that makes this technology valuable (durability, stability, potency, reach, self-propagation, inertness) also a property that could make harm persistent, mobile or hard to reverse? If so, has that property itself triggered scrutiny proportionate to scale? (P1: PCBs, CFCs, DDT, asbestos, TBT, EE2, MTBE, systemic seed dressings; persistence and mobility later became hazard classes.)

2. What benefit is claimed, who has tested it independently, and is the benefit specific to this option or to the wider system it rides on? (P2: DES never worked; MTBE’s specific air benefit was small; growth promoters and many seed treatments added little; sugar beet shows benefits can be real and concentrated.)

3. Against what is the option judged: only the incumbent it replaces, or also options that meet the underlying need by a different principle? Does it share the incumbent’s hazard-conferring property or operating principle? (P3: HCFCs “unduly tolerated”; TBT to boosters; same-form fibres; function-level redesign worked for coatings and refrigerants.)

4. If this option were restricted or failed tomorrow, what would fill the gap, and has that replacement been assessed with equal rigour? (P3, hindsight: BPS/BPF, hydrocarbon solvents, pyrethroid sprays, zinc oxide and ionophores; group restrictions as the institutional response.)

5. What is scaling the technology (mandates, subsidies, bundled contracts, market pressure), and will it reach high volume and dispersive use before appraisal is complete? (P10, P6c: the MTBE oxygenate mandate; ethanol scaled again before assessment; grandfathering.)

6. What long-lived capital, installed stock, standards, contracts, skills and exemptions does deployment create, and is there a planned exit route with sunset dates? (P4a, P4d: chlor-alkali asbestos plants of 42–83 years; PCB equipment; the “temporary” aviation lead exemption; ozone-regime feedstock exemptions.)

7. Is the technology cheap partly because others bear its costs (health, environment, clean-up, future generations), and who would pay if those costs were internalised? (P4b: asbestos, PCBs and halocarbons’ “unjustifiable advantage”; cartels; DDT as cheap incumbent.)

8. Does control rely on a single tactic against an adaptive system, and what happens when the target adapts? (P4f, P5: herbicide and insecticide treadmills; stacked traits bought years, not escape.)

9. Is it sold as an integrated proprietary system whose use by some compels adoption by others, and how concentrated are supply, intellectual property and the R&D pipeline? (P5, P6a: seed–trait–herbicide bundles; dicamba defensive adoption; consolidation found to reduce innovation competition; the contrast with diffuse enabling technologies.)

10. Who is choosing the direction, and are promotion and oversight separated? What share of public research goes to understanding harms compared with developing products? (P6d–e: industry-funded TEL research for 40 years; hazard-research shares of about 1–3% in the reports; promoter–overseer programmes; MAFF.)

11. Which alternatives cannot attract investment because their benefits flow to users rather than providers, and who is responsible for keeping them alive, including the skills and networks they depend on? (P4c, P6a, P8: management-based agriculture; forgotten fuel alternatives; deskilling; husbandry “not actively promoted”.)

12. What claims of “essential”, “no alternative”, “progress” or “competitiveness” are being made, by whom, and can they be checked against ex post outcomes? (P9, P7: TEL “only material available”; asbestos “irreplaceable”; 1974 VC cost forecasts about four times too high; “technology not available” arguments recurring and succeeding in 2025–26.)

13. If restriction is proposed, what are the transition costs, who bears them, and is there a plan for the most vulnerable part of the system? (P7, §13 item 3: weaner pigs after the growth-promoter ban; sugar beet after the neonicotinoid ban; substitutes carried their own risks.)

14. Are the same evidential standards applied to the incumbent, the challenger and the preferred alternative? (P8, §13 item 6: forensic scrutiny of one pathway, face-value acceptance of another; “equally effective” alternatives that cost more.)

15. What unrelated developments (co-technologies, obsolescence, market shifts) are likely to change the trajectory anyway, and how durable would a precautionary exit be if those drivers reversed? (P6g, §13 item 9: catalytic converters and “pure chance”; production ended as “no longer necessary”; DDT’s exit through resistance and new tools; derogation cycles and reversed phase-outs.)

Sources used#

• Digests: all 47 sections.
• Notes: LL1-16 and LL2-19 in full; LL1-11, LL2-12, LL2-03 and LL2-27 in their relevant sections; the lock-in, substitution and innovation sections of all other notes, extracted by heading.
• Hindsight: the innovation, substitution and lock-in claims in LL1-05, LL1-07, LL1-09, LL1-11, LL1-13, LL1-14, LL1-16, LL1-17, LL2-02, LL2-03, LL2-04, LL2-06, LL2-08, LL2-10, LL2-11, LL2-12, LL2-13, LL2-16, LL2-19, LL2-22, LL2-27 and LL2-28; the supplied summaries for the rest.
• External: critiques.md (§3, §4, §7, §9); context.md (§2–4).
• Quotations spot-checked against the source text: LL1-00 (pp. 4, 11), LL1-03 (p. 31), LL1-05 (pp. 53, 58), LL1-06 (p. 64), LL1-13 (p. 141), LL1-16 (pp. 169, 182, 187), LL1-17 (p. 194), LL2-02 (p. 32), LL2-03 (pp. 46, 52–55, 60, 64, 69), LL2-08 (p. 192), LL2-11 (pp. 241, 250), LL2-12 (p. 275), LL2-19 (pp. 460, 461, 466, 474), LL2-22 (pp. 546, 549), LL2-23 (p. 567), LL2-25 (p. 607), LL2-27 (pp. 646, 659, 661), LL2-28 (pp. 670, 672–673, 680).
• Limits: several hindsight files were compiled without general web search, and their caveats apply; the innovation meta-analyses are cited as reported there.