Late Lessons, Jensen Huang and AI
Markdown version

LL2-16 — Ch16 Seed-dressing systemic insecticides and honeybees#

Late lessons from early warnings: science, precaution, innovation (EEA Report No 1/2013), Part B “Some emerging issues”, Chapter 16. Report pages 369–406 (PDF pages 371–408). Chapter text pp. 369–393. Table 16.1 p. 394. References pp. 394–400. Panel 16.1 (Bayer CropScience) pp. 401–402. Panel 16.2 (authors’ reply) pp. 403–406.

Reading record. I read the full text extract in order and reached the final marker (PDF 408 / report p. 406). I checked these pages as images in the PDF: p. 369 (summary box), p. 376 (toxicity values), p. 379 (CTP quotations with the authors’ added emphasis), pp. 384–385 (Figures 16.1–16.3), p. 394 (Table 16.1) and p. 401 (Panel 16.1). The extract was accurate for all of them. Figure contents are described below from the images.

Currency of the text. The chapter dates itself “at present, February 2012” (p. 374) and reports the Paris criminal investigation as of March 2011 (p. 380). The detailed analysis runs to the 2004 maize ban. The authors say sections 16.4.4 and 16.4.5 (2004–2011, EU level) “are less exhaustive and result essentially from reactions to the different comments made during the reviewing process” (p. 372). The summary box (p. 369) refers to recent confirmation by EFSA that the chapter body does not document, so the box was probably written later than the body.

Authors and standpoint#

Authors. Laura Maxim and Jeroen van der Sluijs (p. 369). The chapter gives no affiliations or disciplinary background. [General knowledge, to verify: Maxim is a CNRS researcher in risk governance and knowledge quality whose doctoral work covered this case. Van der Sluijs (Utrecht University, later also University of Bergen) is known for uncertainty and knowledge-quality assessment (the NUSAP tradition). After 2013 he and J.-M. Bonmatin, whose studies feature heavily here, were among the authors of the Task Force on Systemic Pesticides’ Worldwide Integrated Assessment (2015). Critics have described that task force as advocacy-oriented.] The chapter condenses the authors’ earlier peer-reviewed work on the case. Footnote 2 on p. 372 points to Maxim and Van der Sluijs (2007), “Uncertainty: cause or effect of stakeholders’ debates?”, and p. 391 cites their 2010 paper. One of the chapter’s data series, Figure 16.1, lists “Source: Laura Maxim” (p. 384): she compiled it from France Miel cooperative data.

Analytical frame. This is a science-and-technology-studies and knowledge-quality analysis, not an ecotoxicology review. It asks how evidence was produced, judged and used. It sorts the relevance of knowledge for action into three “qualities” (pp. 371–372): 1. Substantive quality: technical (“is the measurement accurate?”), methodological (is the method appropriate?) and epistemological (is enough known?). 2. Procedural quality: the quality of the research and expert processes. This covers competence, field experience, institutional affiliation, “well-being at work”, financial dependencies, relationships with stakeholders and whether local knowledge is included. 3. Social quality: the value judgements shaping how experts and stakeholders communicate and use scientific information in political debate.

Sections 16.2, 16.3 and 16.4 correspond to these three qualities (p. 372). The authors’ stated aim is to describe “the social processes that lead to the application of the precautionary principle in France” and “the ways in which stakeholders have used scientific findings to influence policy” (p. 371).

Evident stance. The chapter sympathises with the beekeepers and the “public scientists”. It is critical of Bayer, of DGAL (the Ministry of Agriculture’s General Directorate for Food), of the Commission for Toxic Products (CTP) and of AFSSA’s 2009 report. It supports the French precautionary bans. Markers of stance: - It openly selects its data: “The scientific data considered in this chapter are thus not exhaustive but selected to reflect the French debate” (p. 371). Post-2004 references are included only if they come from France or from European countries that banned imidacloprid (p. 372). Footnote 3 (p. 373) says the authors picked data “that played an important role in the debate, rather than uselessly trying to inventory all the data available”. - It adds emphasis (bold in the PDF, called “underlining” in fn 23) to words in regulators’ conclusions, such as “only”, “unquestionable effect”, “to completely exclude”, “solely” and “all” (p. 379). - A footnote describes the maize growers’ association AGPM as “also a defender of industrial agriculture, for example through cultivating GMOs and intensive biofuel crops” (p. 385, fn 45). - It concludes that imidacloprid “seems to be a substance particularly ‘fit for the precautionary principle’” (p. 392). - Its closing reply to Bayer calls the French partial ban an “explicit application of the precautionary principle” (p. 405).

Limits the authors acknowledge. They have no information on the experiences of Bayer-funded scientists; they asked a Bayer researcher and got no answer (p. 378). They have no information on the economic consequences of the bans for industry (p. 383). Pollination losses to agriculture were not assessed (p. 385). The relative contribution of insecticides and other factors to the decline of French beekeeping is “unclear” (p. 383). They concede that public researchers can have conflicts of interest too (p. 390).

Panels. - Panel 16.1 (pp. 401–402): “The Bayer CropScience view on Maxim and van der Sluijs”, by Dr Richard Schmuck, Head of the Department of Environmental Safety of Bayer CropScience. Position: the chapter is unbalanced. It takes Gaucho “a priori” as “THE key cause” of losses, ignores the multifactorial literature and omits a book on the socio-political drivers (Rivière-Wekstein 2006). Large-scale monitoring found no correlation between losses and neonicotinoid seed-dressings. Gaucho met all re-evaluation requirements and is still registered elsewhere without reported losses. Losses continued after the French suspension. Suspensions carry “the risk of stopping innovation” and should be reversible. Note on the author: Schmuck is not a neutral commentator. He wrote or co-wrote several studies at the centre of the dispute: Schmuck and Schöning 1999 and Schmuck et al. 2001 (cited p. 376 and fn 32), Schmuck 2004 (criticised p. 387), and Schmuck 1999, “No causal relationship between Gaucho® seed dressing in sunflowers and the French bee syndrome” (panel references, p. 402). - Panel 16.2 (pp. 403–406): the authors’ “Response to the Bayer Cropscience (Richard Schmuck) comments”. A point-by-point rebuttal. - Longer versions of both are on the EEA website (pp. 402, 405). The structure is an industry right of reply followed by the authors’ rejoinder, so the authors have the last word.

Section-by-section notes#

Summary box (p. 369)#

16.1 Introduction (pp. 370–372)#

16.2 Development of scientific understanding (pp. 372–378)#

16.2.1 Technical and epistemological quality

Early warnings (pp. 372–373). - Gaucho seed-dressing was authorised for beet in 1991, maize in 1992 and sunflower in 1993, and first used on sunflower in 1994. - Signs: after days of foraging on flowering sunflower, many foragers did not return. Bees clustered on the ground or hovered disoriented. Queens laid extra brood to compensate. Dead bees were sometimes seen. - Magnitudes: sunflower honey yield fell 40–70% below the prior average, against a normal variation of ±10%. End-of-winter losses were 30–50% against the usual 5–10%. Sources: “personal communications from 20 beekeepers”, Coordination des Apiculteurs 2001 and Alétru 2003 (p. 372). These are beekeeper sources. - The signs varied by year and place, apparently with the local food mix (CST 2003). - Beekeepers linked the problems to the first use of Gaucho on sunflower and to the growing treated area, and asked Bayer about toxicity (p. 373). A long series of studies followed, involving Bayer-funded scientists, the Ministry, AFSSA, beekeepers and “public scientists” (French public researchers are salaried civil servants, fn 4). - Bayer’s 1992 position was no risk “provided it was applied as seed‑dressing”. Its field and tunnel studies reported no risk. At the Montpellier conference (January 1997) and the ACTA meeting (October 1997) they “were criticised”, and public scientists were brought in (p. 373).

Exposure, 1993–1999 (pp. 373–374). - Bayer’s 1993 detection limit was 10 ppb (Placke and Weber 1993). Its studies “either could not detect … or detected it but could not quantify it” (CST 2003). - A 1999 Bayer radiolabel study quantified 3.3 ppb in pollen and 1.9 ppb in nectar (Stork 1999). It was published in 2001 (Schmuck et al., fn 32). - The detection-limit episode (p. 373): - DGAL asked public researchers for “the lowest detection limit possible”, but “without going below 0.01 mg/kg” (10 ppb). - For the 1998 programme it noted that “it is not useful to try to work with the lowest detection limits”. - That limit matched Bayer’s method, and “Bayer representatives also participated in the committee charged with developing the research protocol”. - CETIOM had already estimated that about 1.4 ppb was needed for nectar (pp. 373–374). - Libération reported this, raising “doubts about the DGAL’s impartiality” (p. 374). - The first public studies found residues below 10 ppb in sunflower but could not quantify them (Pham-Delègue and Cluzeau 1998). The Ministry’s summary: the results “raised suspicions about the effects of the product, without formally proving its responsibility” (Ministère 2001b). - January 1999: Minister Jean Glavany banned sunflower seed-dressing under the precautionary principle. The ban was renewed in 2001 (two years) and 2004 (three years) and was still in force in February 2012 (p. 374). Inconsistency: p. 381 lists sunflower decisions in 1999, 2001 and 2003.

Exposure, 2000–2006 (pp. 374–375). - Signs persisted after the sunflower ban. Three hypotheses were offered: exposure through maize pollen (the crops flower together), soil persistence from treated beet or cereals taken up by untreated sunflower, and Régent TS (provisionally authorised December 1995). - In 2001 the sunflower ban was extended and a maize ban refused. The Ministry created the CST (Scientific and Technical Committee for the Multifactor Study of Honeybee Colony Decline). - 2000–2002 public measurements: 2–4 ppb in treated sunflower and maize pollen (Bonmatin et al.), 13.3 ppb in sunflower pollen (Laurent and Scalla 2001) and 1.6 ppb in nectar (Lagarde 2000). - Exposure understanding: up to a month of exposure during flowering. Stored food means effects can come immediately or weeks later. Pollen foragers carry pollen without eating it while nurses and larvae eat it (Rortais et al. 2005). Nectar foragers’ intake depends on flight distance and resources. - In 2002 Bayer publicly declared exposure of 0–5 ppb (AFSSA 2002). - The CST validated 3.3 ppb (sunflower pollen), 3.5 ppb (maize pollen) and 1.9 ppb (sunflower nectar) (p. 374). - Two metabolites (olefin, 5-hydroxy) are acutely toxic to bees, but no metabolite measurement in pollen or nectar could be validated. The CST called for better analytical limits (p. 375).

Effects: definitions (p. 375). - Four types: acute lethal (48-hour LD50), chronic lethal (about 10 days; no standard protocol, so expressed as 10-day LD50, NOEC or LOEC), acute sublethal and chronic sublethal. - Sublethal effects on behaviour, physiology or immunity can “become lethal in time” or make colonies more vulnerable. A disoriented bee may not get home. “This would not be detected in standard pesticide tests, which focus on acute mortality.” - The colony is a “superorganism” (fn 13, Moritz and Southwick 1992), so individual impairments in key roles affect the whole colony. There were no standard sublethal protocols.

Effects: the data (pp. 375–376). - Spray poisonings leave dead bees at the hive. Here beekeepers saw most foragers disappear, and suspected effects on orientation (p. 375). The harm signature did not match what the tests were built to detect. - Bayer’s LOEC fell over time. Three Bayer scientists reported 5,000 ppb in January 1997. Bayer-funded Kirchner (1998–2000) found 20 ppb, about 0.5–1.4 ng per bee (pp. 375–376). - Bayer-funded NOECs, 1999–2000: 0.25–0.7 ng/bee (10 ppb) up to 9 ng/bee. - Public LOECs: 0.075–0.21 ng/bee (3 ppb), 0.15–0.42 (6 ppb), 0.25–0.7 (10 ppb) and 0.31–0.87 (12.5 ppb) (ACTA 1998; Pham-Delègue; Colin et al.). - The “strange” fact that Bayer NOECs are larger than “most of” the public LOECs is explained partly by test type: Bayer’s were acute, the public ones chronic. Different protocols and endpoints were also used. “Of course, the results depend on what, and how, one measures” (p. 376). Minor slip: the text refers to “the two values” where four are listed. - 2001: public scientists reported a chronic LD50 of 12 pg per bee after 10 days at 0.1 ppb (Suchail 2001). My judgement: this sits one to two orders of magnitude below the other chronic values in the chapter (3–20 ppb). It was contested later (p. 387). - 2002: Bayer said its studies showed that “below 20 ppb, no negative effect can be observed on honeybee colonies” (AFSSA 2002) (p. 376).

Persistence (p. 376). - Bayer’s dossier gave soil half-lives (DT50) of 188 ± 25 and 249 ± 40 days, beyond the three-month threshold of Directive 91/414/EEC (Annex VI). The Directive bars authorisation of persistent substances “unless it is scientifically demonstrated” that there are no unacceptable residues in succeeding crops or environmental effects. - Gaucho was authorised under this Directive. Implied: the persistence data should have triggered that demonstration requirement. - Public scientists measured 10.25 ppb in soil in the treatment year and 4.4 ppb a year later (Bonmatin et al. 2000).

16.2.2 Methodological quality (pp. 377–378) - Spray assessment: laboratory mortality, then semi-field, then field. Hazard quotient HQ = application rate ÷ LD50. Bayer’s dossier used this. - Why it does not fit systemic seed-dressings: 1. “The field application rate … is … a highly inadequate measure for the true exposure of honeybees”. What matters is the concentration in pollen and nectar. 2. The LD50 covers only acute effects on adults. 3. Sublethal effects reach the whole colony through food brought into the hive. 4. Risk varies with age and caste (p. 377). - The CST therefore used PEC:PNEC (from industrial-chemicals assessment), covering lethal and sublethal, short- and long-term effects across castes and food matrices (Halm et al. 2006). - Field versus laboratory (pp. 377–378). Bayer held that field experiments decide the risk “regardless of whether they conformed to the results of laboratory studies”. Public scientists held that field trials “cannot be decisive”: colonies and forage vary, bees fly beyond plots, and control and test fields sit too close. A field trial informs “only … about the particular situation in which [it was] done”. Footnote 18 says semi-field tunnels understate exposure and hide orientation effects. Footnote 17 separates “field experiments” from “monitoring” (which measures clinical signs in real conditions). The distinction matters because Bayer’s panel relies on monitoring programmes (p. 401). - “In the end, it was not a scientific institution but the highest judicial administrative institution in France, the State Council, that decided (29 December 1999)” that both laboratory and field results are legitimate (p. 378).

16.3 Processes of generating knowledge and assessing risk (pp. 378–381)#

16.3.1 Public scientists (p. 378). - Their work was judged by “whether or not it supported the positions of certain stakeholders”. - Testimony 1: a researcher received a letter from Bayer’s lawyers in January 1998 threatening a defamation suit with financial reparations (AFP 2003). Bayer also asked the researcher’s superior to influence the researcher’s press statements. The superior refused but urged “extreme care with the press”. - Testimony 2: another researcher was asked by managers “to change topic” after three years (Elie and Garaud 2003, a documentary film). - Testimony 3: in 2000 a superior “suddenly stopped” an EU-funded imidacloprid programme despite confirmed funding (personal communication). - The authors lack information on Bayer-funded scientists’ experiences. A Bayer researcher asked during review provided none. - Evidence quality: specific but anonymous and one-sided.

16.3.2 Official evaluators (pp. 378–380).

Commission for Toxic Products (CTP). A Ministry of Agriculture body of toxicology and ecotoxicology experts, replaced by AFSSA groups in 2006 (fn 21). - 1993: it approved Gaucho on Bayer’s claim of no bee exposure, “without consulting its specialist Honeybee Working Group”, which advised on the “honeybee label” (fn 22) (pp. 378–379). - 11 December 1997 to December 2002: “ambiguous” and “ambivalent” assessments. It could not “confirm or deny a causal link” and repeatedly asked for more studies (p. 379). - 1997 quote: Bayer’s demonstration “is not made in a rigorous and complete manner”, but the beekeepers’ reports cannot show Gaucho is “the only cause”. Bold emphasis is the authors’ (fn 23, called “underlining”). - 1998 quote: no “unquestionable effect”, but not possible “to completely exclude” an effect, given low-dose toxicity. - 2002 quote (on maize): the assessment cannot show that “maize seed‑dressing with Gaucho® can be solely responsible, at national level, for all colony losses” (p. 379). The authors also put their own phrases “everywhere in France” and “in intensive sunflower and maize seed-dressed cultures” in bold (checked on the page image). - Diagnosis: the CTP answered “a question that had never been asked” (all losses, everywhere in France) instead of the real one (losses in intensive seed-dressed sunflower and maize areas) (p. 379). - Procedure: no clear operating rules; “unstructured expert judgement”; no criteria-based quality review; members handling several dossiers at once (p. 379). - Composition: one bee specialist. The Honeybee Working Group was not consulted until 2000 because, according to a former member, “two members were beekeepers and were considered to have an interest in banning Gaucho” (personal communication). Bee scientists were under-represented even in the group. - Result: divergent data, too little expertise and no time or criteria “all contributed to producing ambiguous advice” (p. 379).

State Council (Conseil d’État) (pp. 379–380). It applied legal criteria. - 1999: Bayer challenged the sunflower ban, with seed consortia (Monsanto, Novartis, Rhône-Poulenc, Pioneer, Maisadour, Limagrain) bringing a similar case. The beekeepers’ union UNAF (about 22,000 beekeepers) co-defended the Minister. The Council upheld the ban as based on “an appropriate evaluation” of the 1998 programme and the CTP’s doubts. - 2002 and 2004: it told the Minister to reconsider the refusals to ban maize use, because the Ministry had not assessed harmlessness as the law required. In 2002 the problem was maize-pollen use and its effects; in 2004 it was effects on larvae (p. 380).

Other courts (p. 380). - In 2001 Bayer sued three beekeeper-union leaders (in Châteauroux, Mende and Troyes) for “discrediting” Gaucho and lost all three cases. The courts relied on the unions’ freedom to play their role and express opinions publicly. The chapter does not date the judgments. One court “explicitly criticised the attempt of Bayer CropScience to intimidate a syndicate leader”. - UNAF’s 2001 criminal complaint in Paris “continues to stagnate”. Two judges had been replaced, and in March 2011 it was still at the investigation stage.

16.3.3 CST (pp. 380–381). - From 338 references, the CST found serious risks via larvae, nurses eating pollen, foragers eating nectar and bees eating honey in the hive. For sunflower it excepted one route: pollen ingestion by foragers while making pollen balls (scenario 3). Its conclusion is explicitly conditional on “the scenarios we developed” and “the uncertainty factors chosen”. - Its PEC:PNEC ratios “are of concern”, consistent with beekeepers’ observations in intensive maize and sunflower areas. Sunflower dressing “poses significant risks for bees of different ages”. Maize is concerning for nurses eating pollen, which may explain continued weakening “despite the ban on Gaucho® on sunflowers”. Research should continue because other factors contribute (pp. 380–381). - Handling: DGAL did not pass the 2002 interim report to the CST Management Committee. The Ministry withdrew its logo just before publication in 2003. Of the two readings (independence, or lack of support), the authors favour the second, citing a DGAL letter that found the findings “too precise” and asked for more studies (p. 381).

16.4 Societal debate and policy responses (pp. 381–388)#

16.4.1 Stakeholder strategies (pp. 381–382). - Beekeepers systematically compared all studies with their own observations and published them to mobilise civil society. They had public support and good national press coverage. - DGAL was “ambiguous”. Its “lack of transparency undermined trust”: it released the Gaucho dossier only partly, and in full only after the Minister and the Commission for Access to Administrative Documents intervened. Ministers’ decisions contrasted with “DGAL’s procrastination”, and ministries differed among themselves (p. 381). - Bayer had “an inappropriate communication strategy on scientific figures, which contributed to increasing mistrust” (p. 381). - In 2002 it cited exposure “between 0 and 5 ppb, which is the quantification limit”. The authors call this “a major step forward in Bayer’s communication of scientific figures” but “still vague”. Its own funded study had 3.3 and 1.9 ppb (Stork 1999), and those radiolabel data had been published in 2001 (Schmuck et al. 2001, fn 32). - Public scientists had published quantification limits of 1 ppb and detection limits of 0.3 and 0.8 ppb, online since February 2001 (fn 33). Bayer ignored them. - Bayer “steadfastly maintained” there was no effect. Its 2006 website omitted the CST conclusion (fn 35 logs website searches in 2006, 2009 and 2011) (pp. 381–382).

16.4.2 Policy response (pp. 382–383). - The 1999 sunflower ban was framed as “Applying the precautionary principle” after studies that “yielded suspicions … without, however, formally proving” responsibility (Ministry dossier 2001c) (p. 382). - Why maize came later: “The stakes were higher”. Sunflower was only 10% of Gaucho revenue, and maize covered about 2.5 times the area (1,764,767 ha against 728,555 ha in 2000, fn 37). The stakes were higher too for farmers, beekeepers, the public “and probably for the Minister himself as a politician”. Maize is often grown without rotation, so there is more pressure from pests “(and maize growers)” (p. 382). - The Minister’s error: he told the State Council that maize produces no nectar, so bees do not visit it “for producing honey”, “apparently unaware” that bees collect and eat maize pollen (p. 382). The nectar claim itself is correct. The error was to leave out pollen exposure. - Ministers named in the chapter: Jean Glavany (sunflower ban, January 1999, p. 374); Hervé Gaymard (maize suspension, per the 2004 press-release title, p. 399); Michel Barnier (2007 statement cited by Bayer, p. 402). - Sequence: - 21 January 2002: maize authorisation renewed for ten years with a court case pending; the inquiry was extended to cover it. - October 2002: the State Council advised reconsideration. - 2003: the Minister refused again. - September 2003: the CST found a serious risk to nurse bees. - March 2004: the State Council advised reconsideration again. - “July 2004”: maize use banned (p. 382). Date flag: the reference list cites a Ministry press release of 25 May 2004 (p. 399). The press release said the risk “seems less important than in case of sunflower … but is, however, of concern”. - Insider view (pp. 382–383): on 21 November 2003 the head of DGAL’s pesticide bureau publicly described “only three public servants to deal with 20 000 applications for authorisation per year”, risk assessments “performed jointly with the industry” (“une cogestion”, fn 39), no transparency and too little attention to residues in food. He concluded: “it is impossible for the Bureau to accomplish its mission”. Citation flag: fn 39 cites Le Point of 21 November 2003, but the reference list has Le Point 2001.

16.4.3 Costs and benefits (pp. 383–385).

Chemical industry (p. 383). - More regulation, withdrawals and pest resistance have raised R&D costs, to about USD 50 million per product. (Bayer’s panel gives EUR 250 million per compound, p. 401; the two are not reconciled.) - Systemics are “a highly profitable investment”: imidacloprid products earned DM 800 million (about EUR 409 million) in 1998 and EUR 556 million in 2007. - “A networking investment for the industry because partnerships are made with seed producers and distribution chains.” - No information on how the bans affected industry.

Beekeeping (pp. 383–384). - The sector “worsened significantly”. Hives went from 1,370,220 in 1994 to 1,360,973 in 2004, and beekeepers from 84,800 to 68,800 (GEM-ONIFLHOR 2005). The stable hive count hides higher turnover. - The relative weight of insecticides, the honey market and disease is “unclear”. Imports rose from 6,000 t (1993) to 17,000 t (2004) while consumption stayed at about 40,000 t. - Data from France Miel, Poitou-Charentes, Deux-Sèvres and a national audit show sunflower honey losses from around 1994: “not exhaustive but … significant”. - Beekeepers paid for research and litigation from EU beekeeping development funds, so “the financial burden was double”. They were not compensated, though the EU gave general support after 2003.

Figure 16.1 (p. 384, from the image). - France Miel sunflower honey index for western France, 1988–2001: roughly 94–115 up to 1994 (100.17 in 1994), then 91.61, 81.55, 64.85, 58.28, 54.37 (1999), 53.72 and 42.77 (2001). - Arrows mark “Gaucho® is authorised for sunflower and maize” (between 1993 and 1994), Régent TS authorised for maize and sunflower (1995), and Gaucho banned on sunflower but still used on maize (1999). A fourth label at the end of the arrow reads “RégentTS® is also used for sunflower and maize seed-dressing”. The note mentions Régent TS on sunflower after 1995. Source: “Laura Maxim”. - Inconsistencies: the base period is 1988–1998 on the axis but 1988–1994 in the note (the 1988–1994 values do average about 100), and the data are said to run to 1998 although the bars run to 2001. The first arrow places the maize authorisation at 1993–94, but the text dates it to 1992 (p. 372). - The 1994 bar (100.17), the first year of Gaucho on sunflower, is at baseline. The decline starts in 1995, which matches p. 383 (“started in 1995”) but not the same page’s “started around 1994”. - The index keeps falling after the 1999 sunflower ban, and the series cannot separate imidacloprid from fipronil.

Figure 16.2 (p. 384). - Poitou-Charentes cooperative sunflower honey, about 1.0 million kg (1995), 1.1 million (1996), 0.64 million (1997, 1998) and 0.5 million (1999), with a trend line. - Source: Coordination des Apiculteurs (a beekeeper organisation). There is no pre-1995 baseline.

Agriculture (pp. 383–385). - Seed-dressing spread “rapidly … even when pest control was rarely needed (for example for sunflowers)”. CETIOM calls wireworm risk on sunflower “low or zero” in most areas (on its website since 2003). - Pest control shifted from curative treatment above an economic threshold to prevention “regardless of the presence and abundance of pests” (p. 384). - Seed-dressing saves labour, but some farmers reported unchanged or lower yields and empty seeds. The source is a documentary film (p. 384). - AGPM said the bans raised pest pressure. In 2002 the pesticide authorisation committee’s rapporteur named only terbufos (due to be withdrawn in 2003) and Régent TS as “really effective” maize alternatives (p. 385). - In 2005 farmers claimed losses of 500,000 t of maize (EUR 50 million), partly blamed on the ban. Others blamed the hot, dry summer, and AGPM itself cited drier summers since 2003 (p. 385). - Figure 16.3 (AGRESTE, 1995–2007, read from the image): maize mostly about 77–91, with a low of about 72 in 2003 (heatwave) and a high of about 95 in 2007. 2004 is about 90, 2005 about 83 and 2006 about 86. Sunflower is flat at about 21–25. The chapter: the ban “is not directly correlated with productivity”. Note: the ban came in mid-2004, so 2005 was the first maize season without Gaucho. That year does show a dip (about 90 to 83), which the chapter reports others attributed to the hot, dry summer. Unit flag: the axis says “kg/ha”, but the values only make sense as quintals per hectare. Method flag: national averages cannot detect localised pest damage. - Beekeepers said maize can be grown undressed; AGPM said there is “no authorised alternative” for wireworms. Pollination losses from 1994 to 2004 were “not … assessed”. Persistent seed-dressings make resistance more likely than sprays do (p. 385).

16.4.4 Debate in France, 2004–2011 (pp. 386–387). - Mortality figures are “very diverse and heterogeneous”. - UNAF says colonies partly recovered after 2003 and high summer mortality stopped in intensive areas. The source is a party to the dispute. - AFSSA’s 120-hive study (2002–05) found normal activity and 5–10% winter losses (Aubert et al. 2008). - AFSSA itself (Faucon and Chauzat 2008) and some beekeepers reported high mortality in winter 2005–06, but only 1.2% (2006) and 0.6% (2007) of beekeepers declared mortality (EFSA 2008). Apiaries were reported in good condition in spring 2007 (Clément 2007, a UNAF editorial). - The CNDA survey of winter 2007–08 (168 professionals) found about 30% losses, “some 12 % higher” than the previous two winters (p. 386). - “The reality is that no system exists in France for the accurate and extensive monitoring of honeybees” (bold in original). Official statistics focus on contagious disease, and Ministry and AFSSA disease data diverge (p. 386). - AFSSA’s post-2004 work stresses multiple causes, disease in winter losses and varroa pressure (AFSSA 2009). - Critique of AFSSA 2009 (pp. 386–387): - The authors first concede that the two reports “differ in terms of their objectives and, therefore, in their methods”. - Only 5 of the CST’s 338 references appear in it. The gap runs both ways: 173 pre-2003 references used by AFSSA 2009 were not in the CST report. The authors state this but do not treat it as a limit of the CST. - About 43% of its sources concern disease, about 15% ecotoxicology and 3% imidacloprid. - Its claim to be “almost exhaustive” is therefore unwarranted. - It does not assess Gaucho or Régent in sunflower and maize areas. - Its conclusion that its deliberations “do not confirm the hypothesis of a predominant role attributed to pesticides by beekeeping professionals” is vague on which pesticides, which beekeepers, which period and what method. - Suchail and Schmuck (p. 387): Schmuck (2004, Bayer-funded) found no effects of two metabolites (urea and 6-chloronicotinic acid) under chronic exposure and rejected Suchail’s parent-compound results through literature comparison alone. The authors say he did not analyse the sources of the difference or the protocols. - New exposure routes: sowing dust on wild flowers (Greatti et al. 2006) and maize guttation drops (Girolami et al. 2009). - Persistence: 1–2 ppb in untreated sunflowers a year later, detectable after two years (Bonmatin et al. 2005). - Contamination and synergy: - French pollen: imidacloprid in 49.4% of samples, 6-CNA in 44.4%, at least one of the two in 69% (Chauzat et al. 2006). - Beekeepers had hypothesised pesticide-induced immunodepression (Alétru 2003). - Alaux et al. (2009): Nosema plus imidacloprid produced the highest mortality. Infection raises food intake and so exposure, and the combination suppresses an enzyme that sterilises larval food (p. 387). - Successors: Poncho (clothianidin, Bayer) was not authorised for maize. Cruiser (thiamethoxam, Syngenta) was authorised and monitored in 2008–09, left unrenewed in June 2009, then reauthorised in December 2009 for 2010 (p. 387). - The bans “significantly calmed the controversy”. Relations improved and a Technical Beekeeping Institute was created (p. 387).

16.4.5 EU level (p. 388). - Germany’s Draft Assessment Report (DAR) went to EFSA in 2005, and imidacloprid was approved at EU level in 2008 while France kept its ban (fn 47). - The DAR omitted French exposure studies (“none of Jean‑Marc Bonmatin’s studies”), gave sublethal effects “only limited importance” and used methods not adapted to seed-dressings. - NGOs (Stichting Natuur en Milieu, PAN-Europe, Inter Environnement Wallonie, Nature et Progrès, MDRGF) wrote to the EU Health Commissioner. Their criticisms: no tests per bee category, underestimated nectar intake, validation without criteria, discrediting of unfavourable studies alongside thorough validation of favourable ones, flawed consumption and colony-size tests, too little on dust, and no attention to pathogen synergy (Kindemba 2009). Source note: this is an NGO critique, reported second-hand through Kindemba 2009 (an NGO-commissioned review; general knowledge, to verify). The chapter does not assess it or give the rapporteur’s response.

16.5 Lessons (pp. 388–392)#

Covered in full under “The authors’ own lessons” below. From the framing text: - The lessons “may be relevant for governance of the controversy about systemic insecticide risks in general” (p. 388). - “The most important factor fuelling the debate in France was increasing mutual mistrust between the parties involved”, partly from failures to generate and share information (p. 388). - Local services initially confirmed the beekeepers’ observations. Beekeepers were later denied raw Ministry data, criticised “paralysis by analysis” and the steering of research towards “too ‘complex’ subjects”, and were kept out of protocol design. The Ministry did fund two postdoctoral posts that helped the CST (p. 388). - Monitoring bodies “should have the trust and acceptance of the field actors directly concerned”. “Where key actors are not properly engaged, the monitoring process becomes discredited and ineffective” (p. 388). Experts should be selected transparently for competence, and disciplines balanced, since “specialists in, for example, diseases are more prone to produce conclusions on diseases” (pp. 388–389). - Overarching principle: governance of chemical-risk controversies “must therefore be guided by a continuous focus on promoting mutual trust between the stakeholders, including scientists and policymakers” (p. 389).

16.6 Conclusions (pp. 392–393)#

Table 16.1 Early warnings and actions (p. 394)#

References (pp. 394–400): the source base#

Panel 16.1: Bayer CropScience, Dr Richard Schmuck (pp. 401–402)#

Panel 16.2: authors’ response (pp. 403–406)#

Case timeline#

Authorisation and early warning - 1990–91: Bayer’s initial imidacloprid dossier (4 April 1990) and first supplement (8 January 1991), per the Bayer 1999 reference title (p. 395). 1991: Gaucho authorised for sugar beet; Directive 91/414/EEC adopted (pp. 372, 394). - 1992: maize authorisation. Bayer says there is no risk to bees if used as a seed-dressing (pp. 372–373). - 1993: sunflower authorisation. The CTP approves without its Honeybee Working Group, relying on Bayer’s claim of no exposure. Bayer’s residue method has a 10 ppb detection limit (pp. 372–373, 378–379). - Summer 1994, first warning: Gaucho is first used on sunflower. Beekeepers observe foragers disappearing, disorientation, yield losses of 40–70% and winter losses of 30–50%. Who: practising beekeepers. Strength: clear, repeated field signs coinciding with a new product; causation undemonstrated (pp. 372–373). - December 1995: Régent TS (fipronil) provisionally authorised on the same crops. This is a confounder (p. 374). - 1995–97: Bayer’s field and semi-field studies report no harm. Sunflower honey output keeps falling and Gaucho sunflower area expands (p. 394, Figs 16.1–16.2). - January 1997: Bayer reports an LOEC of 5,000 ppb (p. 376). October 1997: at the ACTA meeting (24 October) Bayer’s studies are criticised and public scientists are brought in (p. 373). Table 16.1 says beekeepers publicly blamed Gaucho in 1997 at “an important meeting” with Bayer and the Ministry. That is probably, but not explicitly, the ACTA meeting (p. 394). - 11 December 1997: the CTP is ambiguous. Bayer’s demonstration is not rigorous, and nor are the beekeepers’ claims of a sole cause (p. 379).

Public research and first precautionary action - 1997–98: DGAL tells public researchers not to go below 10 ppb, matching Bayer’s method, with Bayer on the protocol committee. CETIOM had estimated about 1.4 ppb was needed (pp. 373–374). - January 1998: a researcher receives a defamation threat from Bayer’s lawyers (p. 378). - 1998: public research detects (but cannot quantify) imidacloprid in sunflower. Bayer-funded Kirchner reports an LOEC of 20 ppb (reports dated 1998–2000; the text says “from 1999 onwards”). Public LOECs of 3–12.5 ppb appear (1998–2002). The CTP (16 December 1998) can neither confirm nor exclude an effect. On 17 December more than 1,000 beekeepers protest in Paris (pp. 374–376, 379, 394). - 1999: Libération exposes the detection-limit instruction. January 1999: sunflower use suspended under the precautionary principle (“suspicions … without formally proving”). Bayer and seed companies challenge the ban in court and UNAF co-defends. Stork’s unpublished Bayer study quantifies 3.3 ppb in pollen and 1.9 ppb in nectar. The State Council upholds the ban. On 29 December it rules that both laboratory and field data are legitimate; the chapter’s two descriptions (pp. 378, 380) appear to refer to the same ruling but do not say so (pp. 373–374, 378, 380). - 2000: imidacloprid is detected in maize pollen and soil persistence confirmed (10.25 ppb, then 4.4 ppb); beekeepers demand a ban on all uses. The CTP Honeybee Working Group is consulted for the first time. A researcher’s EU-funded programme is stopped. DGAL withholds most of the dossier (pp. 376, 378–379, 381, 394). - 2001: the sunflower ban is extended and a maize ban refused, and the CST is created. Bayer sues three beekeeper union leaders; it loses all three cases (judgment dates not given). UNAF files a criminal complaint. Suchail reports a chronic LD50 of 12 pg per bee. In December the European Parliament resolves that seed coatings have caused “mass poisoning” of colonies (pp. 374, 376, 380, 392).

Maize: refusal, court pressure, action - 2002: on 21 January the maize authorisation is renewed for ten years, with a court challenge pending. Bayer states publicly that exposure is “0–5 ppb” and that there is “no negative effect below 20 ppb”. On 9 October the State Council tells the Minister to reconsider maize. The CST interim report is completed but not forwarded. In December the CTP concludes Gaucho cannot be shown to be “solely responsible … for all” losses (pp. 374, 376, 379–382). - 2002–03: poisonings affect several thousand hives, with fipronil residues in dead bees (p. 394). - 2003: the Minister again refuses a maize ban. The Ministry removes its logo from the CST report. September 2003: the CST finds a serious risk from maize dressing to nurse bees. DGAL calls the findings “too precise”. In November the head of DGAL’s pesticide bureau publicly describes an agency unable to do its job (pp. 381–383). - March 2004: the State Council again tells the Minister to reconsider, since effects on larvae were not assessed. May or July 2004: maize use banned and Régent TS banned for all agricultural uses (the date differs between p. 382 and the p. 399 reference). The sunflower ban is renewed (pp. 374, 380, 382, 394).

After the bans - 2005: Germany’s DAR goes to EFSA without the French exposure studies. French maize growers claim losses of 500,000 tonnes (pp. 385, 388). - 2005–07: French beekeepers report that high summer mortality has stopped, per UNAF and Table 16.1. The AFSSA 120-hive study (2002–05) finds normal mortality (pp. 386, 394). - 2006: sowing dust identified as an exposure route (Greatti). The CTP is replaced by AFSSA’s DIVE (pp. 378, 387). - 2008: imidacloprid approved at EU level while France keeps its ban. Germany suspends maize seed treatments after a poisoning incident; Bayer attributes it to poor seed-coating quality. Italy’s moratorium begins (pp. 388, 392–393, 401). - 2009: Nosema synergy (Alaux) and guttation (Girolami) reported. Italy’s reported cases fall from 185 to 2. Cruiser is suspended, then reauthorised. AFSSA’s report says its deliberations “do not confirm” a predominant role for pesticides (pp. 386–387, 392–393). - 2010: EPPO’s revised bee scheme is adopted; the authors judge it inadequate. The Commission mandates an EFSA working group (year not stated). Imidacloprid is authorised on fruit trees in France (pp. 372, 390). - 2011: Slovenia loses 2,500 colonies at sowing. Germany lifts most suspensions. The Paris criminal investigation is still open in March 2011 (pp. 380, 393, 404). - February 2012: the French sunflower ban is still in force (p. 374).

Lags - From first signs (1994) to sunflower action (January 1999): about 4.5 years; the authors say “about five years”. - From first signs to maize action (2004): about 10 years. - From the 1997 “public boom” of controversy: 2 years for sunflower and 7 for maize (p. 403). - From the CST’s maize finding (September 2003) to the maize ban (May or July 2004): 8–10 months, needing a second State Council intervention. - At EU level, imidacloprid was approved in 2008, so the French evidence did not produce EU action within the chapter’s timeframe. - Action on sunflower came about one year after the first public-scientist detection (1998) and before proof, framed explicitly as precaution.

What was known when - At authorisation (1991–93): high soil persistence (DT50 188–249 days in Bayer’s dossier, beyond the Directive’s 3-month trigger) (p. 376). The CTP accepted Bayer’s claim that bees were not exposed (p. 378); Bayer says it relied on the aphid logic (p. 401). My inference, not the chapter’s: the 10 ppb detection limit (p. 373) made that claim hard to test. Bee specialists were not consulted (p. 379). - By 1998–99: residues in pollen and nectar were quantified at a few ppb by Bayer’s own radiolabel study (Stork 1999; published 2001). Public researchers had detected residues below 10 ppb but not yet quantified them (1998). Sublethal effect levels from Bayer-funded work fell from 5,000 to 20 ppb, and public LOECs of 3–12.5 ppb appeared between 1998 and 2002. - By 2000–02: public researchers quantified 1.6–13.3 ppb in pollen and nectar (p. 374). The exposure and effect ranges now overlapped in the public record. - By 2003: a formal PEC:PNEC assessment (CST) supported the field observations. - After 2006: new exposure routes (dust, guttation, succeeding crops, wild plants), pathogen synergy and near-ubiquitous pollen contamination.

Harms and costs, as the chapter reports them - Sunflower honey yields in affected areas down 40–70% (p. 372), and the France Miel index down from about 100 to 43 by 2001 (p. 384). - Beekeeper numbers down 16,000 (84,800 to 68,800) between 1994 and 2004, with high colony turnover (p. 383). The chapter says the relative contribution of insecticides, the honey market and disease to this decline is “unclear” (p. 383). - Beekeepers bore the costs of research and litigation from development funds and were not compensated (p. 383). - Losses of pollination services are unassessed (p. 385). Agricultural costs of the bans are contested and unquantified; farmers claimed EUR 50 million in 2005, but this is confounded by drought (p. 385). - Industry costs of the bans are unknown (p. 383).

The authors’ own lessons and conclusions#

Framing lessons, reinforcing the 2001 report (pp. 388–389) - A. Mutual mistrust, fed by failures to generate and share information, was the main driver of the controversy. Hence the need for long-term monitoring and research into early warnings (p. 388). Derived from evidence: data access (p. 388), DGAL opacity (p. 381), Bayer’s communication (pp. 381–382). “Most important factor” is the authors’ judgement. - B. Use lay and local knowledge. Monitoring bodies need the trust of the field actors concerned, and engaging beekeepers and farmers improves protocols and helps explain variability (pp. 388–389). Partly derived (beekeepers’ early and consistent observations; exclusion from protocols). Partly advocacy. - C. Select specialists by competence and transparent procedures, include the relevant disciplines, and balance specialisations, since specialists tend to find causes in their own field (p. 389). Derived from the CTP composition (p. 379) and the AFSSA/CST contrast (pp. 386–387). The “more prone” claim is asserted.

The eight new lessons (pp. 389–392), each stated in bold as a lesson 1. Methods for new technologies. Preamble: governance “must focus on identifying potential properties of new chemicals and anticipating surprises that may arise from them” (p. 389). Lesson: “When dealing with new technologies, verify whether the methods already in use for risk assessment are relevant, given the specific new properties and characteristics of new risks” (p. 389). Evidence: LD50 and HQ tools for sprays were applied to systemic seed-dressings “without any assessment of their adequacy” (pp. 377, 389). Footnote 48: “This is a very old lesson, developed in detail in the analysis of DDT by Dunlap (1978).” Strongly derived from the case. 2. New tests. “Develop new tests to assess sublethal effects of pesticides, their chronic effects and their effects on the colony” (p. 389). Without standard protocols, results vary with bee subspecies, age, temperature, fasting time and colony genetics (Evans and Pettis 2005; Alaux et al. 2009). Field experiments cannot represent “average” conditions. A post-2004 French expert group (“appointed (but not funded)”) drafted tests. EPPO’s 2010 scheme, based on ICPBR proposals, “failed to significantly change the risk assessment pattern” and “still does not consider sublethal and chronic effects properly”, so the Commission mandated an EFSA working group (pp. 389–390). Derived. 3. Resources. “Do not underestimate the resources needed to implement policies.” Ensure “adequate personnel (in number and competence) and financial resources” (p. 390). Evidence: three staff for 20,000 applications a year (p. 382). Derived, but from one insider letter. 4. Independence, competence, transparency. “The independence and competence of the experts on the issue at hand must be assured, as well as complete transparency of the research process” (p. 390). This applies to public and private researchers alike. Public researchers face conflicts too: underfunded labs seek industry money, and some consult for industry, for example on varroa products for firms that also make crop insecticides. Conflicts are financial or intellectual (“prior commitment of a scientist to a particular world view”). Recommendations: - Balance funding between the “knowledge economy” (private) and the “knowledge society” (public, with researchers given a status ensuring “the ‘highest possible level’ of independence”). - Contracts guaranteeing publication “regardless of their content”. - Transparent industry dossiers open to external scrutiny, with balanced opportunities to comment that stop “the most powerful stakeholder from capturing the process with repeated comments”. - Substance Information Review Forums modelled on REACH’s Substance Information Exchange Forums. - A “cost‑free database” of the original studies behind risk assessments. - Public conflict-of-interest declarations. - Researchers not evaluating products of companies that fund them (pp. 390–391). Mixed: the need is derived from the case, while the specific instruments are advocacy and design proposals. 5. Social quality of communication. “Be aware that the social quality of the scientific information you communicate in the debate determines your public trustworthiness” (p. 391). Evidence: “major deficiencies” in communication by Bayer and some state services. Six recommendations: be reliable (use all available knowledge), robust (answer criticism), complete (do not ignore contrary information), relevant (address the specific signs and area, not general issues), logical (do not contradict yourself unless you say you have changed your view) and legitimate in your sources (competent, conflict-free researchers) (p. 391). Normative code, drawing on case evidence and Maxim and van der Sluijs 2007. 6. Quality criteria for evidence. Bodies assessing authorisation applications “should develop clear and standardised scientific quality criteria” (p. 391). All literature, including published papers, should be considered, and industry data should be quality-assessed. “An important issue is the balance between the burden of proof … and its credibility.” Selecting studies by unjustified “expert appreciation” allows “arbitrary or subjectively justified exclusion”, which “can potentially have a decisive influence on the final result”. Good Laboratory Practice (GLP) “only provides guarantees about the transparency and the traceability of the laboratory work. It does not guarantee the scientific quality of the study”, and new risks increasingly need non-standard tests (p. 391). Derived: CTP unstructured judgement (p. 379), the DAR omitting French studies (p. 388), the NGO critique of selective validation (p. 388). 7. Multicausality. “Prioritise the potential causal factors and address them separately before assessing potential correlation or synergies among them.” Many causes “should not become an excuse for not dealing with particular clinical signs and particular causes”. Causes may be “primary” or “secondary”, for example pesticide-induced immunodepression favouring disease. Countrywide multiple causes do not contradict a specific risk in treated areas. Prioritise by feasibility, potential to reduce effects and co-benefits such as less social conflict, and invest in research on synergies at low doses (pp. 391–392). A normative-analytical argument. It is well reasoned but contestable in how priorities are set. 8. Protect early-warning scientists. “Build the regulatory background needed to protect early-warning scientists” (p. 392). What does society owe scientists, and how much institutional freedom and legal protection do they get? “Generalised social conflict” around an issue “is likely to discourage scientists from working on the subject.” Democratic knowledge production needs accountability through peer review and “freedom of scientists to pursue their work independently on socially sensitive issues”. “Misuse of scientific results to support predetermined conclusions, and actions that provoke anxiety and psychological pressure are unacceptable” (citing Gleick 2007) (p. 392). Derived from the testimonies on p. 378, but those are thinly sourced.

Conclusions (pp. 392–393) - Imidacloprid is “fit for the precautionary principle” because its effects are highly variable and dependent on the environment, and some are “uncontrollable”. Heterogeneous evidence lets each stakeholder find arguments. Similar declines and precautionary bans elsewhere in Europe corroborate the concern. The honeybee is a sentinel. Scientists’ independence matters. Social concerns should shape research agendas.

Evidence-derived findings versus advocacy - Most evidence-derived: lessons 1, 2 and 6, and the monitoring gap. - Mixed: lessons 3, 4, 5 and 7. - Most advocacy-inflected: lesson 8 as a legal agenda; “fit for the precautionary principle”; and the implied (never explicitly argued) suggestion that France’s precaution was vindicated. That suggestion rests mainly on UNAF recovery reports, the calming of the controversy and Italian figures (pp. 386–387, 393, 404).

Mechanisms and dynamics#

1. The harm did not match what the testing system could see. The regulatory template was built around sprayed pesticides, where poisoning appears as dead bees at the hive and is measured by acute LD50 relative to application rate (pp. 375, 377). A systemic seed-dressing produces low-level, chronic, sublethal exposure through pollen and nectar. Its signature is disappearance and disorientation, which is invisible to acute tests and to “dead bee” monitoring (pp. 372, 375). The new technology’s mode of exposure fell outside the assessment’s concept of harm, so the absence of acute mortality was read as absence of risk. The CTP’s 1993 approval rested on Bayer’s claim that bees were not exposed (p. 378).

2. Detection limits and measurement choices produced “no evidence”. The early “not detected” results reflected a 10 ppb analytical limit (p. 373). When public research began, the regulator set that same limit as a floor (“without going below”), declared lower limits “not useful”, aligned the protocol with the manufacturer’s method and let the manufacturer help write it (p. 373). A technical parameter set the boundary of what could be known. Meanwhile the manufacturer’s own radiolabel study had quantified residues in 1999. Those data were published in a journal in 2001 (fn 32), yet the company’s 2002 public statement still gave only a vague “0–5 ppb” range, which the authors call “a major step forward” but “still vague” (pp. 373, 381). The chapter documents what happened but does not claim intent.

3. Contesting what counts as evidence. The dispute was partly over which kind of evidence decides: field trials, which Bayer said were decisive, or controlled laboratory studies, which public scientists preferred (p. 377). The chapter says field trials cannot hold conditions constant, cannot stop bees foraging beyond plots, and only inform about the particular situation tested. Semi-field tunnels understate exposure and hide orientation effects (fn 18, p. 378). My inference, not the chapter’s wording: such trials are prone to null results, and if a null field result could trump positive laboratory findings, the choice of standard in effect set the burden of proof. A court, not a scientific body, settled that both types are legitimate (p. 378). Later, Schmuck (2004) rejected Suchail’s results by literature comparison rather than by analysing protocols (p. 387), another contest over evidentiary rules.

4. The framing of the question and the standard of proof. The CTP repeatedly answered whether Gaucho was the only cause of all losses nationally (p. 379). That standard can never be met in a multicausal system, so the assessment could not reach a conclusion, and “ambiguous advice” (p. 379) left the status quo in place. The authors’ reframing asks whether there is a risk in exposed areas, with multicausality handled by prioritisation (p. 391; Panel 16.2, p. 403). The industry panel pairs “multi-factorial” with “no correlation” (p. 401). On the authors’ reading, multicausality works as a rhetorical resource for delay. Their counter-mechanism is to split the question by scale and place.

5. Institutional capacity, composition and co-management. - Three civil servants for 20,000 applications a year, risk assessment “performed jointly with the industry”, and opaque procedures (p. 382). - An expert commission with one bee specialist, working by unstructured judgement under heavy workload (p. 379). - The Honeybee Working Group was sidelined because beekeeper members were seen as interested (p. 379), while the manufacturer sat on the research protocol committee (p. 373). This is my synthesis, not the authors’ phrasing: conflicts of interest were policed asymmetrically, with the harmed party’s interest treated as disqualifying and the producer’s treated as normal participation. - Different expert bodies with different disciplinary mixes and bibliographies reached different conclusions: the CST mostly ecotoxicology, AFSSA 2009 mostly disease (pp. 386–387). - The EU-level DAR omitted the French exposure studies (p. 388). Assessment conclusions follow from which literature enters the process, and there were no transparent criteria for inclusion (p. 391).

6. Fragmented, divided government. Ministers acted (sunflower 1999; maize 2004) while DGAL “procrastinated”, withheld documents and held back the CST interim report, and the Ministry withdrew its logo from the final report (pp. 381, 388). But ministers were not simply ahead of the administration. They also refused a maize ban in 2001 and 2003, renewed the maize authorisation in 2002, and defended the refusal with the no-nectar argument (pp. 374, 382). The chapter contrasts ministerial decisions with “DGAL’s procrastination” (p. 381) but does not explain the administration’s motives. Speculation, not in the chapter: the administration may have had sunk commitments in its earlier approvals.

7. Economic stakes and sequencing of action. Action came first where stakes were lowest (sunflower, 10% of Gaucho revenue, a crop rarely needing pest control) and last where they were highest (maize, 2.5 times the area, monoculture, organised growers) (pp. 382–383). Seed companies joined Bayer’s legal challenge (p. 380), reflecting the “networking investment” in partnerships with seed producers and distributors (p. 383). The maize refusal was defended with an incomplete exposure model: maize produces no nectar, so bees do not visit it to make honey. That ignored pollen, which bees collect and eat (p. 382). The chapter presents this as ignorance (“apparently unaware”), not as motivated reasoning.

8. Prophylactic deployment and lock-in. Seed-dressing turned pest control from threshold-based, curative treatment into blanket prevention “regardless of the presence and abundance of pests” (p. 384). It was adopted “rapidly” even where pests were rare (p. 383). The input came pre-bundled with seed, supported by partnerships along the supply chain (p. 383). Once established, alternatives looked scarce: “no authorised alternative” for wireworms (p. 385), and only terbufos or fipronil as effective substitutes (p. 385). After the 2004 bans, authorisation was sought for other systemic seed-dressings (Poncho, refused for maize; Cruiser, authorised, suspended and reauthorised), both neonicotinoids, and several countries had stop-start authorisations (pp. 387, 393). Régent TS (fipronil, a different chemical class) was not a post-ban substitute: it was authorised in 1995, ran in parallel with Gaucho and was banned alongside it in 2004 (pp. 374, 385). The chapter also flags evolutionary lock-in: persistent molecules put constant selection pressure on pests (p. 385).

9. Complexity and superorganism dynamics. Stored food delays and prolongs exposure. Different castes are exposed differently. Individual sublethal impairment can reduce colony-level function. Pathogen–pesticide synergy (Nosema) can turn sublethal doses lethal (pp. 374–375, 387). Toxicity values vary by up to 1,000-fold, half-lives vary with the environment, and more than one study found no dose-effect relationship (p. 392). This is why the authors think some effects are “uncontrollable”. It also means evidence is heterogeneous, which stakeholders can mine selectively (p. 392).

10. Persistence and time lags. Soil persistence carries exposure into untreated succeeding crops and wild plants over one to two years (pp. 376, 387). Residues are found all year round (p. 405). Decision lags were 4.5 years for sunflower and 10 for maize. Some harms are hard to reverse: 16,000 beekeepers left the trade over the decade (p. 383), though the chapter calls the insecticides’ share of that decline “unclear”.

11. Pressure on knowledge producers. Legal threats, approaches to managers, redirection of research topics and cancelled programmes (p. 378). Bayer also sued beekeeper union leaders and lost (p. 380). The chilling effect is the authors’ inference: “generalised social conflict … is likely to discourage scientists” (p. 392). Separately, the authors point to structural limits on independence: underfunded public research that depends on industry money or consultancy, and few toxicology posts (pp. 390, 393). The chapter does not link these directly to the specific pressures on p. 378.

12. Law as corrective. The administrative judge (State Council) forced the executive to comply with statutory assessment duties. It required the Ministry to establish harmlessness for maize pollen and larvae (p. 380) and upheld precaution against industry challenge (p. 380). Civil courts protected beekeepers’ right to speak (p. 380). The criminal route stalled (p. 380). Access-to-documents procedures (the CADA), together with the Minister’s own intervention, forced disclosure (p. 381). In this case, legal institutions substituted for weak scientific-administrative governance.

13. Lay and local knowledge. Beekeepers were the detectors. They noticed the signs, linked them to the timing of the new product, systematically compared studies and mobilised the press and public (pp. 373, 381). Local agricultural services corroborated them (p. 388). Their reports were judged “not rigorous and stable enough” to show that Gaucho was “the only cause” (CTP 1997, p. 379; the same finding said Bayer’s own demonstration was not rigorous either), and they were shut out of protocol design (p. 388). The first 2001 report’s lesson on lay knowledge is reaffirmed, with trust added as a condition (pp. 388–389).

14. Communication, trust and escalation. Trust eroded through vague numbers, selective citation, website omissions and administrative opacity (pp. 381–382). The escalation included street demonstrations, lawsuits both ways and a criminal complaint (pp. 380, 394). The authors treat “social quality” (completeness, responsiveness, relevance, consistency) as a condition of credibility, not a matter of PR (p. 391). Once bans were in place the controversy “significantly calmed” (p. 387): resolving the regulatory question reduced the conflict.

15. Distribution of costs and benefits. - Gains: the manufacturer (global sales of several imidacloprid products, not Gaucho alone: EUR 409 million in 1998 and EUR 556 million in 2007, p. 383), seed and distribution partners, and possibly farmers through lower labour, though yield benefits are unclear (pp. 384–385). - Harms: beekeepers, through lost yields, colony replacement, exit from the trade (causes mixed, p. 383), and research and legal costs paid from development funds without compensation (p. 383); unpriced pollination services (p. 385); wild pollinators, the least visible (pp. 370, 393). - Burden of proof: in practice it fell on those harmed. They had to generate evidence and fund litigation (p. 383).

16. How key actors thought (mental models and blind spots). - The manufacturer: - “Confined to the soil” and tiny doses mean no exposure, even “reducing aerial pollution” (p. 371). - Aphids re-invading before bloom mean no residues reach bees at bloom (p. 401). That model ignores storage, castes, chronic intake and the implied lack of selectivity (p. 404). - The field trial is the final arbiter (p. 377). - Quantification limits framed as ranges (“0–5 ppb”) (p. 381). - Framing: coincidence (“market introduction coincided with a time when bee health issues had increased”), “accidents” and “incidents”, “a priori”. - Precaution treated as a “positive tension” with innovation that should be reversible (p. 401). - Regulators: spray-era tools as default (p. 389); lower detection limits “not useful” (p. 373); findings “too precise” (p. 381); no nectar means no bee exposure, ignoring pollen (p. 382); sole-cause framing (p. 379). - The authors (reflexively): treat public scientists and beekeepers as more credible than industry-funded work and ministries, and read the post-ban recovery as confirming the ban.

17. Framing and language. - “Reducing aerial pollution” as an environmental selling point (p. 371). - “Suspicions … without formally proving” as the explicit trigger for precaution (p. 374). - The CTP’s hedged double negatives, which the authors counter with bold emphasis (p. 379). - “Paralysis by analysis” (p. 388). - “Almost exhaustive” as a claim of authority in AFSSA’s report (p. 386). - The “honeybee label” (fn 22). - “Alarm bell” and sentinel (p. 393). - “Fit for the precautionary principle” (p. 392). - Bayer’s “spirit of precaution” and “bee-responsible farming” (p. 401). - The authors’ insistence that dust poisonings are “not incidents or accidents” (p. 404).

18. Multi-level governance. France banned two uses while the EU approved the substance (2008) on a DAR that omitted French studies (p. 388). Other member states banned, reapproved and re-banned (pp. 392–393). National precaution without harmonised EU assessment produced an unstable patchwork. The EU route (EPPO revision, then an EFSA mandate) moved slowly (p. 390).

Transferable insights (technology-neutral)#

  1. Legacy assessment tools applied to a new mode of action create systematic blind spots. If an innovation changes how exposure happens (route, duration, internal distribution), tests calibrated on the old mode can report “safe” by construction. Evidence: HQ based on application rate and acute LD50, designed for sprays, used for systemic seed-dressings without any check of their adequacy (pp. 377, 389). Acute tests miss disappearance (p. 375). Strength: strong. The mismatch is well documented and conceptually clear, and the process-level record (EPPO revision, EFSA mandate, p. 390) shows it was acknowledged within the regulatory system.

  2. Measurement thresholds define what can count as evidence, and whoever sets them shapes the outcome. “Not detected” often means “below the method’s limit”. Aligning protocols with the incumbent’s methods, or letting interested parties design them, can manufacture absence of evidence. Evidence: 10 ppb floor set to match Bayer’s method, Bayer on the protocol committee, CETIOM estimating 1.4 ppb was needed (pp. 373–374); vague “0–5 ppb” communication (p. 381). Strength: strong for the documented sequence. Moderate for any inference about intent, which the chapter does not claim.

  3. How the causal question is framed decides whether an assessment can reach a conclusion. Asking whether something is the sole cause of all harm everywhere guarantees inconclusive results in multicausal systems. Asking about contribution in the relevant exposure context can be answered. Evidence: CTP conclusions 1997–2002 (p. 379); authors’ reply (p. 403). Strength: strong. Documented in the assessors’ own words, and logically robust.

  4. Multicausality can be used to justify inaction. The counter is to prioritise tractable, specific causes rather than wait for a full causal account. Evidence: Lesson 7 (pp. 391–392); Panel 16.2 (p. 403); the contrast with Bayer’s panel (p. 401). Strength: moderate. The logic is sound, but deciding which cause is “primary”, and whether acting on it helps, remained contested in this case. Bayer’s point that losses continued after the bans (p. 402) cuts partly against it.

  5. When evidence is heterogeneous and methods unstandardised, each party can assemble “its own science”, and controversy hardens rather than resolving. Evidence: NOEC/LOEC ranges and protocol diversity (p. 376); 21-, 40- and 1,000-fold variability (p. 392); the “diverse ecotoxicological portfolio” (p. 392); the CST and AFSSA bibliographies barely overlapping (pp. 386–387). Strength: strong. Documented on several sides, including by the authors’ own selectivity.

  6. Disputes over which kind of evidence counts (controlled studies or real-world trials) are in effect disputes over the burden of proof, and they may end up settled by courts rather than by science. Evidence: field versus laboratory dispute and the State Council ruling (pp. 377–378). Strength: moderate. A single case. The limits of field trials are well characterised in the text. The framing of low-power real-world nulls overriding positive controlled findings is my inference from it.

  7. Under-resourced regulators that co-produce assessments with applicants and rely on unstructured expert judgement produce ambiguous, delay-favouring advice. Evidence: three staff for 20,000 applications, co-management, opacity (pp. 382–383); CTP procedure and workload (p. 379). Strength: moderate. The insider letter and CTP record are strong for this case, but the causal link to outcomes is inferred.

  8. Who sits on an expert body, and which literature it considers, shapes what it concludes. Disciplinary imbalance tilts findings towards the experts’ own field. Evidence: one bee specialist on the CTP and a sidelined Honeybee Working Group (p. 379); the CST/AFSSA bibliography contrast (pp. 386–387); the DAR omitting French studies (p. 388). Strength: moderate. The composition facts are documented. The claim that disease specialists are “more prone” to disease conclusions (p. 389) is asserted, with the bibliography contrast as suggestive support.

  9. Conflicts of interest can be policed asymmetrically: the harmed party’s stake is treated as disqualifying while the producer’s participation is treated as normal. Evidence: the Honeybee Working Group sidelined because of beekeeper members (p. 379), versus Bayer on the protocol committee (p. 373) and “co-management” of assessments with industry (p. 382). Strength: suggestive. My synthesis of separate facts in the chapter; the authors do not argue it explicitly.

  10. Practitioners in daily contact with a system are often the first to detect harm. Excluding them from monitoring design and data access degrades both the monitoring and trust. Evidence: beekeepers’ 1994 detection and correlation with the new product’s area (pp. 372–373); lack of access to raw data and exclusion from protocols (p. 388); corroboration by local services (p. 388). Strength: moderate. The early detection is well documented. Whether engagement would have improved outcomes is argued, not shown.

  11. Without baseline and post-intervention monitoring, neither harm nor the effect of interventions can be established, and each side fills the gap with its own numbers. Evidence: “no system exists in France for the accurate and extensive monitoring of honeybees” (p. 386); declaration rates of 1.2% and 0.6% (p. 386); conflicting post-ban accounts (pp. 386, 402). Strength: strong. The authors state it in bold, and official statistics are said to support neither quantification nor causal identification (p. 386). The Bayer panel does not concede the gap: it treats research monitoring programmes, including French AFSSA studies, as adequate evidence (p. 401). The gap is between those research surveys and a standing national monitoring system.

  12. Pressure on independent researchers (legal threats, approaches to employers, redirected topics, cancelled programmes) can suppress early-warning research. Structural dependence on interested funders makes this more likely. Evidence: testimonies (p. 378); intimidation criticised by a court (p. 380); funding dependence (pp. 390, 393). Strength: moderate for the occurrence, which is specific and partly corroborated by court findings in the beekeeper cases. Suggestive for the systemic chilling effect: anonymous, one-sided, with no data on how widespread it is.

  13. Prophylactic, bundled deployment of a technology (pre-applied to everyone regardless of need) expands exposure beyond demonstrated need and builds supply-chain lock-in that later makes restriction look costly. Evidence: seed-dressing adopted even where pests were rare, especially sunflower, where CETIOM rated wireworm risk “low or zero” in most areas (pp. 383–384); maize growers disputed that alternatives existed (p. 385); the “networking investment” with seed and distribution partners (p. 383); seed companies joining the legal challenge (p. 380); claims of no alternatives (p. 385). Strength: moderate. The pattern is well described. The size of the lock-in is not quantified.

  14. Persistent, mobile agents create exposure routes that only appear after wide deployment. Assessment built around intended-use scenarios will under-predict them. Evidence: carryover into untreated crops, sowing dust, guttation, all-year pollen residues (pp. 376, 387, 393, 404–405); persistence data in the original dossier that exceeded the Directive’s trigger (p. 376). Strength: strong. Multiple independent studies across several countries are cited.

  15. The costs of contesting evidence (research, monitoring, litigation) fall mainly on the parties harmed, who are usually the least resourced. This amounts to an informal reversal of the burden of proof. Evidence: beekeepers funding research and legal fees from development funds, uncompensated (p. 383); beekeepers compiling and publicising the evidence themselves (p. 381). Strength: moderate. Documented qualitatively but not quantified.

  16. Administrative courts and access-to-information law can compensate for weak technical governance by forcing agencies to meet their statutory evidence duties. Evidence: State Council rulings of 1999, 2002 and 2004 (p. 380); the CADA forcing document release (p. 381). Strength: moderate. Clear in this case. How well it transfers depends on the legal system.

  17. Action tends to come first where the commercial stakes are lowest and last where they are highest, so precaution may protect the least economically significant uses first. Evidence: sunflower (10% of revenue) banned in 1999, maize in 2004 after repeated refusals and court interventions (p. 382); the authors’ stated lags (p. 403). Strength: suggestive. One sequence, and the authors’ inference about stakes. Other explanations are possible: the maize evidence came later (maize pollen detected in 2000; CST finding in 2003).

  18. Communication that ignores or blurs available data destroys credibility faster than disagreement over interpretation does. Completeness and responsiveness condition trust. Evidence: “0–5 ppb” while precise figures existed; website omissions; DGAL opacity (pp. 381–382); lesson 5 (p. 391). Strength: moderate. Well documented for the manufacturer and administration. The link to escalating controversy is the authors’ judgement, though consistent with the calming after the bans (p. 387).

  19. Restricting one product in a class without addressing the class often leads to substitution by near-equivalents and stop-start regulation. Evidence: post-2004 applications for Cruiser and Poncho, both neonicotinoids (p. 387); banning, reapproving and re-banning in Slovenia and Germany (p. 393). Régent TS was a parallel product, not a post-ban substitute (p. 374). Strength: suggestive. Described but not analysed as a mechanism by the authors.

Limitations, contestation and bias check#

Advocacy elements alongside the analysis - The authors say openly that they select data “to reflect the French debate” (p. 371). Their post-2004 inclusion rule admits only French sources or sources from countries that banned imidacloprid (p. 372), citing focus and space. That rule structurally excludes evidence from jurisdictions that found no problem, though the chapter does not apply it strictly (it cites, for example, US and Greek studies, pp. 389, 405). The case is therefore told largely from the evidence base of the side whose position prevailed in France. - Stance markers: the characterisation of AGPM in a footnote (p. 385); bold emphasis added to regulators’ words (p. 379); “fit for the precautionary principle” (p. 392); “Can Europe afford this?” (p. 393).

Thin or partisan evidence - Key facts about harm and pressure rest on the following: - Personal communications from 20 beekeepers (p. 372). - An unnamed former CTP member (p. 379). - Anonymous researcher testimonies (p. 378). - A documentary film, used both for a researcher quote (p. 378) and for farmers’ reports of empty seeds (p. 384). - Press reports (Libération, Le Point, AFP). - Beekeeper-organisation data: Coordination des Apiculteurs (Fig. 16.2), UNAF recovery claims (p. 386). - An author-compiled index (Fig. 16.1). - A farm-union news item as the citation for the Italian APENET figures (p. 393). The underlying source is a research programme, and Panel 16.2 cites APENET material (fn 51), so this one is less partisan than its citation suggests. - An NGO letter reported second-hand (Kindemba 2009) for the critique of the EU draft assessment (p. 388). - Several of these are the natural sources for a governance study, but they cannot bear strong causal weight.

Causal attribution - The chapter’s subject is governance, but it often treats Gaucho’s harm as established. The summary says “evidence pointed to” Gaucho (p. 369). It speaks of “the poor pollination associated with Gaucho®” (p. 384) and of Gaucho having “posed a risk” (p. 391). - The CST’s PEC:PNEC finding is a risk inference with chosen uncertainty factors (“based on the uncertainty factors chosen”, p. 380). It is not a field demonstration of harm. - Major confounders: - Régent TS (fipronil), introduced on the same crops from 1995 and banned at the same time as Gaucho on maize in 2004 (pp. 374, 384, 394). The post-2004 improvement cannot be attributed to imidacloprid alone. - Varroa and disease (p. 386). - The honey market (p. 383). - Weather. - Figure 16.1 shows yields kept falling after the 1999 sunflower ban (p. 384). The chapter reports three hypotheses (maize pollen, soil persistence, fipronil; p. 374), and the CST names maize-pollen exposure as one possible explanatory element (p. 381). These are plausible but untested in the chapter.

Asymmetric use of scale - The authors fault the CTP and AFSSA for national-scale framing of bee losses (pp. 379, 386–387). Yet they use national average yields (Fig. 16.3) to counter farmers’ claims of ban-related pest damage (p. 385), while conceding that agricultural repercussions are “unclear” (p. 383). National yields cannot detect localised wireworm damage any more than national mortality figures can detect localised poisoning. This is a double standard in method (my critique, not raised in the panels). The figure also shows a maize dip in 2005, the first season after the ban (about 90 to 83), which the chapter attributes, via others, to the hot, dry summer.

Suchail et al. (2001) - The 12 pg-per-bee chronic LD50 at 0.1 ppb (p. 376) is one to two orders of magnitude below the other chronic effect levels in the chapter (3–20 ppb, p. 376). The chapter criticises Schmuck’s (2004) literature-based rebuttal (p. 387) but does not report whether independent laboratories replicated the result. [To verify in the hindsight pass: my recollection is that such extreme chronic toxicity was not widely reproduced.] The chapter also uses the absence of a dose-response relationship (p. 392) as evidence of “uncontrollable” effects, when it could equally point to experimental artefacts.

Counter-evidence engaged with only partly - Bayer’s panel cites large-scale monitoring that found no correlation (Chauzat et al. 2009, 2010; Nguyen et al. 2009; Genersch et al. 2010, p. 401). The authors’ printed reply attacks the weaker Bulgarian and Croatian surveys and turns a Chauzat 2010 residue paper against Bayer. It does not engage with Nguyen, Genersch or the Chauzat 2009 null finding (pp. 404–405). The longer online reply was not checked. - The AFSSA 120-hive study finding normal mortality in 2002–05 (p. 386) is reported but not reconciled. - The bibliography gap between CST 2003 and AFSSA 2009 runs both ways (5 of 338 CST references in AFSSA; 173 pre-2003 AFSSA references not in CST, p. 386). The authors use it only against AFSSA. - The “global pollination crisis” is cited partly to Ghazoul, who questioned it (p. 370).

Double standard on sources - Rivière-Wekstein is rejected because he directs a corporate consultancy and writes on unrelated subjects (p. 403), which is a mix of interest and competence grounds. The authors then commend books by journalists and NGO figures (p. 403) without applying either test to them.

Fair points on the chapter’s side - The authors acknowledge that public researchers have conflicts too (p. 390). - They separate risk in exposed areas from national losses. - They caution against the parallels with US Colony Collapse Disorder (“rightly or wrongly”, p. 393). - They admit to uncertainty over economic effects and causes (pp. 383, 385). - They document the regulator’s contradictions from its own texts. - They credit Bayer’s 2002 exposure statement as “a major step forward” (p. 381), and they concede that the CST and AFSSA reports had different objectives (p. 386). - Many of their documentary claims come from primary official documents (CTP minutes, State Council rulings, Ministry releases). - Bayer’s aphid argument (p. 401) is weak as stated, and the authors’ rebuttal (p. 404) is persuasive. - The core critique of methods, that spray-era tests are inappropriate for systemic exposure, became mainstream. [To verify: EFSA’s 2012 opinion and 2013 bee guidance.]

Fair points on Bayer’s side - Colony losses really are multifactorial, and varroa and disease are major drivers (AFSSA, p. 386). The chapter concedes this. - Losses did continue after the bans (p. 402; consistent with Fig. 16.1 and the 2007–08 winter figures, p. 386). - The sequence of evidence from 1994 to 1998 rests largely on temporal coincidence (p. 373). - The chapter’s selection rule (p. 372) does bias its evidence base. - Reversibility of precautionary measures is a legitimate governance principle. The chapter does not discuss what evidence should trigger lifting a ban.

Hindsight bias - Moderate. The chapter was written 2010–12 about events of 1994–2004, and it reads early ambiguity (CTP 1997–98) partly through later findings (CST 2003; dust, 2006–11; Nosema, 2009). But the early critiques of the detection limits and of the method were raised at the time (CETIOM’s estimate, p. 374; criticism at the October 1997 ACTA meeting, p. 373).

Case selection - In Late lessons terms this is a case where precaution was applied, with delay, and presented approvingly (the authors imply, but do not argue, that it was vindicated). There is no counterfactual analysis of agricultural costs, of pest outbreaks after the bans, or of effects of substitution (for example pyrethroid sprays replacing seed-dressings).

Internal inconsistencies and errors (for the record) - Dates of sunflower suspensions: 1999, 2001 and 2004 (p. 374) against 1999, 2001 and 2003 (p. 381). - Maize ban date: “July 2004” (p. 382) against the reference to a Ministry press release of 25 May 2004 (p. 399). - Figure 16.1 base period: 1988–1998 on the axis, 1988–1994 in the note. Data coverage: 1988–1998 in the note, bars to 2001 (p. 384). - Figure 16.3 unit: “kg/ha”, most likely quintals per hectare (p. 385). - Yield loss: 40–70% (p. 372) against 30–70% (Table 16.1, p. 394). - Italian figures: 185 to 2 cases (p. 393) against 185 to 3 beekeepers (p. 404). - Development cost: USD 50 million (p. 383) against Bayer’s EUR 250 million (p. 401). Different measures, not reconciled. - “The two values cited from public scientists” when four are listed (p. 376). - Le Point citation year mismatch (fn 39 against the reference list). - “CPT” for CTP (p. 379). - Table 16.1 says “proving” synergy (p. 394), stronger than the text (p. 387). - The 35% crop-output claim (p. 370) is likely overstated relative to Klein et al. - France Miel losses “started in 1995” and “started around 1994” on the same page (p. 383). Figure 16.1 shows 1994 at baseline. - Figure 16.1 places Gaucho’s maize authorisation at 1993–94; the text says 1992 (pp. 372, 384). - Table 16.1 puts Bayer’s 1999 challenge in the Paris administrative court; the text describes the challenge before the State Council (pp. 380, 394). - Panel 16.2 cites Vidau et al. 2011 for imidacloprid–Nosema synergy, but that paper tested fipronil and thiacloprid (pp. 404, 406). - Thomson 2000 (p. 376) and “Klein, 2007” (p. 370) are cited but not in the reference list.

Dissent within the chapter - The panel exchange is the only formal dissent. The authors have the last word, and the EEA provides no neutral adjudication.

Post-2013 pointers for the hindsight pass (general knowledge, not verified in this read) - EFSA’s January 2013 conclusions found high acute risk to bees for certain uses of clothianidin, imidacloprid and thiamethoxam. The EU restricted uses under Implementing Regulation (EU) No 485/2013 from December 2013, then limited outdoor uses to permanent greenhouses in 2018 (Regulations 2018/783–785). Imidacloprid’s EU approval later lapsed. - EFSA’s 2013 bee guidance document was long not endorsed by member states. A revised version appeared in 2023. - France banned neonicotinoids from September 2018 under its 2016 biodiversity law, allowed sugar-beet derogations from 2020, and was constrained by a CJEU ruling of January 2023 (C-162/21) against emergency authorisations of banned treated seeds. A 2025 French law (“loi Duplomb”) sought to reintroduce acetamiprid, and the Conseil constitutionnel struck that provision (August 2025). - Major field studies: Henry et al. 2012 (Science); Rundlöf et al. 2015 (Nature), which found effects on wild bees but not significantly on honeybee colonies; Woodcock et al. 2017 (Science), which found country-dependent effects. These bear on the honeybee “sentinel” claim (p. 393). - The EU Transparency Regulation (2019/1381) made studies submitted to EFSA public, partly implementing lesson 4’s “cost-free database” (p. 390). - The outcome of the Paris criminal investigation (p. 380) is unknown to me.

Notable quotes#

  1. “Most important, the French case highlighted the major weaknesses of regulatory risk assessment and marketing authorisation of pesticides” (p. 369).
  2. “it is not useful to try to work with the lowest detection limits” (DGAL, quoted p. 373).
  3. “This would not be detected in standard pesticide tests, which focus on acute mortality.” (p. 375)
  4. “In the end, it was not a scientific institution but the highest judicial administrative institution in France, the State Council, that decided” (p. 378).
  5. “the CTP conclusions were answering a question that had never been asked” (p. 379).
  6. “it is impossible for the Bureau to accomplish its mission” (head of DGAL’s pesticide bureau, 2003, quoted p. 383).
  7. “The reality is that no system exists in France for the accurate and extensive monitoring of honeybees.” (p. 386)
  8. “a diverse ecotoxicological portfolio allows each stakeholder to identify their own ‘scientific arguments’” (p. 392).
  9. “Suspending products may be helpful in some instances. It does however bear the risk of stopping innovation if it is not handled carefully” (Schmuck, Bayer CropScience, Panel 16.1, p. 401).
  10. “Multicausality cannot become an argument for avoiding dealing with specific causes” (authors, Panel 16.2, p. 403).

Open questions#

  1. Did the bans work, net of confounders? Can the partial French bans of 1999 and 2004 be linked to recovery in exposed areas once the simultaneous fipronil ban, varroa trends and weather are separated out? The chapter’s recovery evidence comes mainly from UNAF. The AFSSA data are mixed (normal mortality in 2002–05 but high mortality reported in winter 2005–06), and Bayer’s sources report continued losses.
  2. What did the bans cost agriculture? What were the actual effects on maize and sunflower growers (localised wireworm damage, substitution by other insecticides such as sprays), given that national yields are too coarse a measure?
  3. Was Suchail et al. (2001) replicated? Its extraordinary chronic toxicity finding matters for how much weight the “uncontrollable effects” argument can carry.
  4. What happened in the Paris criminal investigation opened in 2001?
  5. Did the regulatory learning stick? How far did EU-level assessment after 2013 adopt the chapter’s proposals: chronic, sublethal and colony tests; exposure-based PEC:PNEC; transparent quality criteria for studies; public access to studies? How long did implementation take (the long-stalled bee guidance)?
  6. How was conflict of interest handled in later expert bodies? Did the asymmetry noted in insight 9 persist, and were early-warning researchers given any protection?
  7. Is the honeybee a good sentinel for wild pollinators? Later evidence suggests some wild bees are more sensitive, and honeybee colonies can buffer losses. How does that affect the “alarm bell” argument (p. 393)?
  8. When should a ban be lifted? The Bayer panel raises reversibility (p. 401). The chapter does not say what evidence would justify lifting a precautionary ban, or how to avoid stop-start regulation of the kind seen in Slovenia and Germany (p. 393).
  9. Did new products avoid the same mistakes? After restrictions, were successor insecticides assessed with methods fit for their exposure modes, or did insight 1 repeat itself?
  10. Can the governance claims be separated from the chemistry? How robust are they if the ecotoxicological case against imidacloprid in these specific uses turns out weaker than the authors believed?

Audit log#

Independent audit on 2026-09-25. I re-read the full extract (PDF 371–408) and checked the PDF page images for report pp. 379, 384 and 385. - Box 16.1: separated the box’s contents from main-text claims on p. 370. Added a second overstatement flag on the 35% figure (it covers all animal pollinators, not wild ones only) and a Klein citation mismatch. - DGAL detection limit: “In 1998 it added” corrected to “For the 1998 programme” (p. 373). - NOEC/LOEC comparison: corrected to Bayer NOECs larger than “most of” the public LOECs (p. 376). - Suchail 12 pg value: “extreme” now labelled as my judgement and quantified against the chapter’s other values. - 16.2.2: added fn 17’s distinction between field experiments and monitoring. - CTP 2002 quote: specified that it concerns maize seed-dressing, and recorded the authors’ own bolded phrases (checked on the p. 379 image). - Beekeeper-leader lawsuits: added the courts’ stated basis (union freedom of expression) and that the judgments are undated. - CST conclusion: added the scenario-3 exception and its dependence on chosen scenarios and uncertainty factors. - Bayer’s “0–5 ppb” statement: added the authors’ concession (“a major step forward”) and the 2001 publication of Bayer’s radiolabel data (fn 32). - Minister’s maize claim: corrected. He said bees do not visit maize “for producing honey”; the nectar claim is true and the error was omitting pollen. Also added the ministers’ names from the chapter (Glavany, Gaymard, Barnier). - Figure 16.1: added the fourth arrow label, the 1994 bar at baseline, and the 1992/1993–94 maize-authorisation mismatch. - Figure 16.3: corrected the self-contradictory maize range, and added the 2005 post-ban dip (about 90 to 83) read from the image. - 16.4.4: added AFSSA’s own report of high 2005–06 winter mortality, the spring 2007 recovery report, the reciprocal 173-reference bibliography gap, and the authors’ concession that CST and AFSSA had different objectives. - Schmuck 2004: named the two metabolites tested (urea, 6-CNA). - NGO critique of the draft assessment report: flagged as a second-hand NGO source that the chapter does not assess. - 16.5 framing: added the “continuous focus on promoting mutual trust” principle and the statement that monitoring without engaged actors “becomes discredited and ineffective”. - Conclusions: attributed the no-dose-effect finding to Suchail 2001 and Schmuck 2004, and attributed the Italian figures to APENET (cited via a farm-union item) rather than to a farm-union source alone. - Table 16.1: added the 1998 Régent TS entry and the Paris administrative court versus State Council discrepancy. - References: “450,000 hives” corrected to “ruchers” (apiaries or hives). Added citations missing from the list (Thomson 2000, Le Point 2003, Klein 2007). - Panel 16.1: added Bayer’s acknowledgement that views on acceptable precaution vary, and Schmuck’s further co-authored references (Curé et al. 2001; Maus et al. 2003). - Panel 16.2: gave the authors’ actual grounds for rejecting Rivière-Wekstein and noted his book is itself a polemic. Flagged the Vidau et al. 2011 miscitation (fipronil and thiacloprid, not imidacloprid) and that Cummins 2007 is an ISIS report. Qualified “not engaged” as applying to the printed reply only. - Timeline: corrected the Bayer 1999 dossier page (p. 395, not p. 394). Marked the ACTA meeting / Table 16.1 “important meeting” link as probable. Dated Kirchner as 1998–2000, flagged the State Council 1999 rulings as probably one ruling, and noted the lawsuit outcomes are undated. Replaced “AFSSA downplays pesticides” with its actual wording, and marked the EPPO judgement as the authors’ own and the EFSA mandate year as unstated. - “What was known when”: corrected the claim that public researchers had quantified residues by 1998–99 (they quantified them in 2000–02), and labelled the detection-limit explanation for the 1993 approval as my inference. - Harms: added the chapter’s own caveat that the insecticides’ share in the fall in beekeeper numbers is “unclear” (applied in the timeline, mechanism 10, mechanism 15 and the digest). - Evidence versus advocacy and case selection: “vindicated” softened to an implied, not argued, claim. - Mechanism 2: removed the incorrect claim that Bayer’s residue data were not publicly acknowledged until 2001–02. - Mechanism 3: labelled the null-result and burden-of-proof reasoning as my inference. - Mechanism 6: removed the unsupported “sunk commitments” claim (kept as labelled speculation) and noted that ministers also refused the maize ban twice. - Mechanism 7: replaced “factual error … motivated reasoning” with the chapter’s own framing (incomplete exposure model; “apparently unaware”). - Mechanism 8 and insight 19: corrected the claim that Régent TS (fipronil) was a chemically related post-ban substitute. - Mechanism 9: “dose-response is not monotonic” corrected to “no dose-effect relationship found”. - Mechanisms 11–13: separated the authors’ structural-funding point from the specific pressures, noted the Minister’s role in the document disclosure, and softened “dismissed” for the CTP’s view of beekeeper reports. - Mechanism 15: clarified that the sales figures are global sales of several imidacloprid products, for two years only. - Insight 6: labelled the low-power framing as my inference. - Insight 11: removed the claim that the Bayer panel implicitly concedes the monitoring gap. It contests it. - Insight 13: narrowed “beyond any demonstrated benefit” to exposure beyond demonstrated need, mainly in sunflower. - Insight 15: replaced an off-point evidence item. - Limitations: softened the section heading and “side that won”, and noted the inclusion rule is not strictly applied. Recharacterised the Italian source, added the NGO second-hand source, and corrected the attribution of the post-1999 decline hypotheses. Added the post-ban 2005 maize dip and the chapter’s concession of “unclear” agricultural effects to the scale critique. Added the one-sided bibliography critique and the authors’ concessions, and qualified Panel 16.2’s non-engagement as applying to the printed text. - Internal inconsistencies: added five further items (1994/1995 start, the Figure 16.1 maize date, court venue, the Vidau miscitation, missing references). - Open question 1: corrected “AFSSA and Bayer data point the other way” to reflect mixed AFSSA data. - Digest: added beekeeper sourcing for yield losses; reworded the CTP framing, capacity attribution, lawsuits (“discrediting”, not defamation) and the anonymity of testimonies. Fixed the Italy/EU chronology and added the APENET attribution. Added causation caveats on beekeeper numbers and national yields (including the 2005 dip). Softened the stakes mechanism and corrected “substitution”. Corrected the fipronil date. Added the one-sided bibliography caveat and the printed-reply caveat, the authors’ “major step forward” concession, and a to-verify flag on “later mainstreamed”. - Strand check: no mentions of contemporary technologies or companies outside the report were found in either file. None were added.