Fidelity review of `01-late-lessons-analysis.md`#

Strand A critical review, 26 September 2026. Scope: fidelity of claims, numbers, quotations and page references to the two EEA reports and the working files. Line numbers refer to 01-late-lessons-analysis.md as of 03:11, 26 Sept 2026.

How the check was done#

Against the source text (working/text/chunks/). Wrote a phrase finder that normalises quotes, dashes and line-break hyphens and reports the report page on which each phrase occurs. Ran it on about 290 quoted phrases, numbers and page references across every section of the document, including all twelve lessons as reprinted in LL2 (p. 11). I also read the surrounding passages for about 40 of these where the paraphrase, not only the wording, needed checking. I then extracted every quoted string in the document (about 600) and searched for each one across all 47 chunks, to catch misquotations.
Against notes, hindsight, themes and external files. Checked about 65 post-publication and contextual claims (for example Hammitt 2005; BSE active testing; Stockholm listings; Pfizer; DES “possibly effective”; the French alert commission; NNI shares; the northern cod reference point; Chimed-Ochir 14.1-fold; the Manville Trust at 5.6%; Minamata’s 55,000 people and 32-person pilot; hormone sanctions; Fukushima’s 2,351 deaths; imposex 81% → 21%; CFC-12 at 44%; BPA 20,000-fold and 250,000-fold; lead deaths; sugar beet; the chlor-alkali plants; Cohen and Tubb; the vinyl chloride fourfold figure; Lieberman and Kwon’s 28 cases; EPRS; the innovation principle; Bartrip; Michaels; Gee).
Overall result. The document is very faithful. Almost every page reference lands on the cited page or the next one. The twelve lessons are verbatim. Table A2.1, Table 26.4 (ten features against three), Box 27.4 (twelve criteria), Table 27.2’s bands, Stern’s discount-rate figures, the TCE and BPA divergence figures, and nearly all hindsight numbers match their sources. The problems below are therefore mostly about framing, several misquotations, and a few hindsight numbers that were mis-transcribed or conflict between working files. A compact log of checks that passed is at the end.

Issues, ranked by importance#

1. Lead-death revision attributed to the wrong cause (it is used as key evidence for a “Strong” mechanism)#

Location. §4.5 table, row “Appraisal boundaries decide the answer” (line 398): “lead-attributable deaths rose from ~0.9m to 3.5m as cardiovascular effects were modelled”. Repeated in §5.4 row LL2-23 (line 582) and Appendix A, LL2-23 (line 1066).
Problem. The working file does not say the rise came from adding cardiovascular effects. It says Global Burden of Disease (GBD) estimates rose “as its exposure models changed”: about 0.9m (GBD 2019), 1.48m (GBD 2021), and 3.5m (GBD 2023, using cumulative bone-lead exposure). Cardiovascular deaths were already the bulk of the GBD 2019 figure. The document’s wording turns a change in exposure modelling into proof that the omitted pathway was later added. That overstates the direct vindication of LL2-23’s claim (p. 568).
Evidence. hindsight/LL2-23.md, line 30 (“rose as its exposure models changed … GBD 2023, using cumulative bone-lead exposure”). Line 169: Larsen and Sánchez-Triana (2023) estimate 5.5m lead-attributable cardiovascular deaths, “six times higher than the GBD 2019 estimate”, and 77% of the USD 6.0 trillion cost is cardiovascular mortality.
Fix. “Global lead-attributable death estimates rose from about 0.9m (GBD 2019) to 3.5m (GBD 2023) as exposure models changed to cumulative (bone-lead) exposure. A separate 2023 estimate put lead-attributable cardiovascular deaths at 5.5m, six times GBD 2019 (hindsight LL2-23).” In §5.4 and Appendix A, replace “revised … as cardiovascular effects were modelled” or “Undercounting confirmed (lead deaths ~0.9m → 3.5m)” with “estimates rose several-fold as exposure models changed”.

2. The account of how the twelve lessons were derived leaves out half of what the reports say#

Location. §3.2 intro (line 198): “built from a pre-existing appraisal framework (ESTO) that the fourteen cases were used to ‘test or elaborate’”. §4.10 (line 483): “a checklist of failure modes built from a pre-existing framework”. §5.6 (line 603): “The framework preceded the cases”. Appendix A, LL1-16 (line 1016): “The framework came from Stirling’s ESTO project, and cases illustrate rather than test it.”
Problem. LL1-16 says ESTO “provided the initial framing” and helped in “organising the consideration of the lessons”, and that it gave “an opportunity to test or elaborate” ESTO’s points against the cases (p. 168). The Preface says the case authors’ own lessons “were then distilled into twelve ‘late lessons’ by the editorial team, under the guidance of the EEA Scientific Committee” (LL1-00, p. 3). The document never reports the Preface’s account. In Appendix A it states as fact (“cases illustrate rather than test it”) what is really an analytical inference (no coding or counter-case method is reported), and that statement contradicts the report’s own words, “test or elaborate”.
Evidence. chunks LL1-16, p. 168; chunks LL1-00, p. 3. The document itself quotes “test or elaborate” on line 198.
Fix. In §3.2: “ESTO ‘provided the initial framing’ (LL1-16, p. 168). The Preface says the lessons were ‘distilled’ from the case authors’ own lessons (LL1-00, p. 3). No coding method or counter-case search is reported, so (analysis) the cases work more as illustration than as a test.” Label the Appendix A sentence and §5.6 “The framework preceded the cases” as Analysis, and give both sources.

3. “Chapter 28’s three most-cited statistics do not reproduce” overstates, and conflicts with §5.2#

Location. §3.5 Analysis (line 268). Related: §5.5 item 2 (line 594), “the unsourced 1% research-funding figure”, listed among figures that were wrong.
Problem. Only one of the three statistics failed on checking. “Half of all articles” misreads Grandjean et al. (2011). The 1% figure “cannot be verified” (verdict unclear), and the hindsight reconstruction puts the plausible share at “between about 1% and 2%”. “4 of 88” has never been independently re-analysed (verdict partly held up). The document’s own §5.2 says “No independent re-analysis of ‘4 of 88’ exists” (line 514). “Do not reproduce” implies a failed replication that never took place.
Evidence. hindsight/LL2-28.md, lines 33–35, 69, 74, 96, 120 and the Claim 6 section (“A plausible total therefore lies between about 1% and 2% … Neither the 1% nor the 3% figure can be confirmed, and the report gives no source for either”).
Fix. “Of Chapter 28’s three most-cited statistics, one misreads its source (‘half of all articles’), one is unsourced and unverifiable, though a 1–2% share is plausible (the 1% research-funding figure, which conflicts with Chapter 27’s 3%), and one has never been independently re-analysed (‘4 of 88’).” In §5.5 item 2, add “(unsourced; plausible range 1–2%)”.

4. The northern cod reference-point cut is given as “about 40%”, but the working files disagree#

Location. §6.2 K5 evidence (line 682): “northern cod reference point lowered about 40%”. Appendix A, LL2-17 (line 1054): “a reference point lowered about 40%”.
Problem. hindsight/LL1-02 works it out from DFO’s own figures: under the old limit, 2021 biomass of 411 kt was 0.52 times the limit, and under the new one, 368 kt was 1.16 times it. The new limit is therefore “roughly 40% of the old one”, a cut of about 60%. The same file notes that DFO’s 2024 wording (“the previous LRP was roughly 40% higher than the current LRP”, which implies a cut of about 29%) “does not match” and “looks like a drafting slip”. hindsight/LL2-17 (line 41) and T07 (line 219) repeat “lowered … by about 40%”. The document picks one figure from a disputed set without flagging the dispute, in a lens entry rated Strong.
Evidence. hindsight/LL1-02.md, lines 102–106; hindsight/LL2-17.md, line 41; themes/T07, line 219.
Fix. Drop the percentage and quote DFO: “partly through a downward revision of the limit reference point, ‘not an increase in the quantity of cod’ (DFO; the size of the cut is inconsistent across DFO documents, roughly 30–60%; hindsight LL1-02, LL2-17)”. Correct T07 and hindsight LL2-17 to match.

5. “Moving target” is presented as the report’s own term; it is not in the text#

Location. §3.4, LL2-28 bullet (line 252): Three drivers of delay: the "moving target" (by the time harm is confirmed, "today's technology is now safe"). §4.7 Counter-evidence (line 457): "moving target" is asserted without one.
Problem. The phrase “moving target” appears nowhere in either report (a search of all 47 chunks finds nothing). LL2-28, p. 672 says: “by the time evidence of harm is confirmed, the technology has often changed, leading to assumptions that, unlike yesterday’s technology, today’s technology is now safe”. Putting a label in quotation marks next to a real quotation makes it look like the report’s term.
Evidence. chunks LL2-28, p. 672.
Fix. Remove the quotation marks, or write “what this document calls the ‘moving target’ problem (LL2-28, p. 672: by the time harm is confirmed ‘the technology has often changed’)”. In §4.7, write “the moving-target claim (LL2-28, p. 672) is asserted without a worked case”.

6. The mobile-phone verdict applies the symmetry rule to one side only#

Location. §5.4 row LL2-21 (line 580) and Appendix A, LL2-21 (line 1062), which rely on “a WHO-commissioned review (2024) [that] found no increased risk at moderate certainty”. Compare §4.3 Counter-evidence (line 369), §5.6 (line 607) and §6.11 item 2 (line 972), which single out the Hardell group’s funding footnote.
Problem. The document’s own symmetry check asks whether “critics’ and advocates’ funding, legal roles and stakes [are] disclosed to the same standard”. hindsight/LL2-21 records, as relevant “because the chapter’s own argument is about independence”, that the WHO-commissioned review’s authors include ICNIRP’s vice-chair, the lone dissenter at the 2011 IARC vote, and co-authors of studies the chapter attacks. It also records that the review’s full text was blocked (HTTP 403), so the check rests on the abstract, and it lists the signals that persist (CERENAT OR 2.89; a bias-adjusted Canadian Interphone OR of about 2; IARC 2024 calling the human evidence “mixed”). The document states the reassuring verdict without any of this, while it scrutinises the chapter authors’ funding.
Evidence. hindsight/LL2-21.md, lines 20–21 (access limits), 28–38 (conflicts of role), 44–54 (remaining signals).
Fix. In Appendix A, LL2-21, add one clause: “several review authors held roles in ICNIRP or in studies the chapter disputes, and the review was assessed from its abstract; isolated signals persist (CERENAT; the Canadian Interphone reanalysis; IARC 2024 ‘mixed’)”. Keep the verdict “substantially weakened” but tie it to the independent cohorts and incidence data, not the review alone.

7. The “exceptions” to “no Part A company or regulator voice” are not exceptions#

Location. §5.1 item 2 (line 492): “No Part A contribution comes from a company whose conduct is at issue or a regulator defending itself; the main exceptions are Bayer’s neonicotinoid panel with the authors’ reply (LL2-16, pp. 401–406) and Guidotti’s and Castaño’s commentaries (LL2-06, p. 145; LL2-05, p. 130)”. Carried over from T09, line 176.
Problem. LL2-16 is in Part B, not Part A (the document’s own §2.3 and §2.4 place it there). Guidotti’s panel is “Not an industry response. No company voice appears in its own terms” (notes LL2-06), and Castaño writes as a biomonitoring scientist, not a regulator defending itself. None of the three is an exception to the sentence as written.
Evidence. notes/LL2-06.md, line 58; notes/LL2-05.md, lines 49, 515, 742; the document’s §2.3 (line 96).
Fix. “No Part A contribution comes from a company whose conduct is at issue or a regulator defending itself. The only company voice in LL2 is Bayer’s dissent on neonicotinoids (Part B; Panel 16.1, pp. 401–402, with the authors’ reply, pp. 403–406). Within Part A, the nearest counterweights are Guidotti’s reading of beryllium conduct as ‘denial rather than cupidity’ (LL2-06, p. 145) and Castaño’s caution against exaggerating risk (LL2-05, p. 130).” Correct T09 to match.

8. A quotation that does not exist in the source#

Location. §4.1 table, row “Distinctive (‘signature’) outcomes” (line 314): angiosarcoma's rarity "made causation undeniable" (LL2-08, p. 189).
Problem. No such phrase appears. The source says “the rarity of the disease also meant that the causal connection with vinyl chloride was undisputed”. “Undisputed” (a social fact) is not “undeniable” (an epistemic claim).
Evidence. chunks LL2-08, p. 189.
Fix. angiosarcoma's rarity meant the causal link was "undisputed" (LL2-08, p. 189).

9. Other misquotations and quoted paraphrases (each minor; together they breach the document’s own quotation convention)#

The document’s convention (§1.4) is to quote LL1 only in short phrases. These phrases should also be exact. - Line 307: BSE, evidence "not actually being sought". The source reads “when no evidence was actually being sought” (LL1-16, p. 172), and the document quotes this correctly on line 218. Fix: use the exact phrase. - Line 228: whether a demand for more evidence is "prudence" or "paralysis" depends on who bears the costs. The source reads “‘paralysis by analysis’ or is merely part of a ‘prudent and careful evaluation’”, and says the judgement “will be influenced by the individual, social or interest group’s assessment of the likely pros and cons as they impinge upon them” (LL1-16, p. 182). Fix: quote “prudent and careful evaluation”, and paraphrase the second part as “on how the pros and cons fall on each group”. - Line 225: "cost of failure". The source has “costs of failure” (LL1-16, p. 188). - Line 399: "no longer a viable industry". The source has “would no longer be a viable industry” (Brush Wellman 1977, quoted at LL2-06, p. 135). - Line 457: "not inconvenient". The source has “has not been inconvenient” (LL1-12, p. 129). - Line 249: "jury still out". The source’s category is “the jury is still out” (LL2-02, pp. 19–21). - Lines 340 and 663: "sole cause, everywhere" is a paraphrase in quotation marks. The quotation is “solely responsible, at national level, for all” (LL2-16, p. 379). Fix: remove the quotation marks, or use the real phrase. - Line 331: "not enough is known". The source has “not enough was known” (LL2-06, p. 138). - Fix for all. Correct the wording, or drop the quotation marks where the text is a paraphrase.

10. The Minamata mother did not see congenital poisoning “before the experts”#

Location. §4.2 (line 326): “a Minamata mother who recognised congenital poisoning before the experts, Harada among them (LL2-05, pp. 105–106)”. Appendix A, LL2-05 (line 1030): “A mother recognised congenital disease before the experts”. Also digest LL2-05, line 26.
Problem. The same page records that “Shoji Kitamura mentioned in 1959, ‘it is possible that the substance causing the poisoning was transferred to the infants through the placenta’”. The mother’s remark convinced Harada during his 1961 survey. She was ahead of Harada and of the prevailing “placenta protects” consensus, not ahead of every expert.
Evidence. chunks LL2-05, p. 105; notes/LL2-05.md, line 314 (“Kitamura raised placental or breast-milk transfer in 1959”).
Fix. “a mother whose observation convinced Harada in 1961, when medical consensus held the placenta protective (Kitamura had raised placental transfer in 1959; LL2-05, p. 105)”.

11. A regulator’s definitional clause is used as evidence of an interested party’s intent to “change the rules”#

Location. §4.3 table, row “Changing the rules” (line 359), rated “Strong on intent”: a "competitiveness" ground for moving the nano definition threshold (LL2-22, p. 540). Also §6.5 I4 evidence (line 784) and T03, line 248.
Problem. At p. 540, “competitiveness” appears in the text of the Commission Recommendation’s definition: the 50% threshold “may be replaced by a threshold between 1 and 50%” “where warranted by concerns for the environment, health, safety or competitiveness”. That is a clause in an official definition. It is not documentation of any interested party lobbying to change a standard, so it cannot support “Strong on intent” alongside the tobacco and beryllium documents.
Evidence. chunks LL2-22, p. 540.
Fix. Remove it from the “Strong on intent” evidence. If kept, move it to §4.6 as an example of a cost or competitiveness qualifier written into a definition, rated suggestive.

12. Sugar-beet yield loss attributed to the ban without its stated cause#

Location. §4.4 table, row “Enthusiasm and conspicuous benefit” (line 378): “French sugar-beet yields fell from 851 to 649 q/ha after the ban”.
Problem. The ban took effect in 2018. The 851 q/ha figure is for 2019, already after the ban, and the fall to 649 came in 2020 “during an outbreak of virus yellows”. The working file treats it as a genuine counter-case, but the document’s wording implies the ban alone caused an immediate fall.
Evidence. hindsight/LL2-16.md, lines 88 and 352.
Fix. “French beet yields fell from 851 q/ha (2019) to 649 q/ha (2020) in a virus-yellows outbreak two years after the 2018 ban (hindsight LL2-16)”.

13. Cyclamate is counted as a false positive, contrary to the chapter’s classification#

Location. §5.2, “False positives were not short-lived” (line 513): “…cyclamate is still banned in the US after 55 years”.
Problem. LL2-02 classes cyclamate as “the jury is still out”, not as one of the four false positives, and hindsight/LL2-02 labels it that way (line 242). Using it as evidence about false positives’ duration applies the document’s own reading without saying so.
Evidence. hindsight/LL2-02.md, line 242 (“Cyclamate (classed ‘jury still out’)”).
Fix. “…and cyclamate, which the chapter classes ‘jury still out’, is still banned in the US after 55 years”, or drop it from this bullet and keep it in §4.2’s “bans persist after concern fades”.

Location. §3.6 Analysis (line 280): “safety-research share fell from about 10% (2016) to a requested 1% (FY2026)”. §4.4 table (line 382): “Hazard research share … fell from about 10% to a requested 1%”.
Problem. The 10% figure is the environmental, health and safety (EHS) share of NNI funding in 2016, which fell to about 4% by 2020. The 1.1% FY2026 figure is for Program Component Area 5, “responsible development” as a whole, after the categories were restructured (it was 2.3–2.4% in 2023–25). The trend is real and probably understated, since PCA 5 includes EHS, but the two numbers are not one series.
Evidence. hindsight/LL2-22.md, lines 69–70, 334 and 492.
Fix. “the EHS share of the US nanotechnology programme fell from about 10% (2016) to about 4% (2020); the whole ‘responsible development’ line was 1.1% of the FY2026 request (hindsight LL2-22)”.

15. Table 27.2 is presented as the EEA’s rule for when action may rest on weak evidence#

Location. §5.2 design choice 1 (line 502): “The EEA’s own scale lets action rest on ‘weak’ (10–33%) or ‘moderate’ evidence (LL2-27, Table 27.2, p. 658)”. Similar wording in §4.1 Counter-evidence (line 318): “LL2 allows action on ‘weak’ evidence (Table 27.2)”.
Problem. Table 27.2 is titled “Different strengths of evidence for different purposes: some examples and illustrations”. Its “weak” row cites existing regimes (Sweden’s 1973 chemicals law; WTO SPS Art. 5.7 provisional measures). It does not set an EEA threshold. The EEA working definition speaks only of “an appropriate strength of scientific evidence” (p. 649). The asymmetry argument still stands, because LL2 endorses using such bands while demanding “high confidence” for a false positive, but the wording overstates what the table does.
Evidence. chunks LL2-27, pp. 657–658 and p. 649.
Fix. “LL2’s evidence scale illustrates, with approval, regimes that act on ‘weak’ (10–33%) or ‘moderate’ evidence (Table 27.2, p. 658), while a false positive requires ‘high confidence’ (67–95%) of no harm (LL2-02, p. 18).”

16. The knowledge-to-ignorance ratio is judged as a harm predictor, which it was not offered as#

Location. §4.1 “The typology” (line 303): “The knowledge-to-ignorance ratio proved a poor predictor: it grouped nanotechnology, GM food and mobile phones as low-ratio fields, and they later diverged”.
Problem. LL2-27 offers the ratio to decide where “precautionary measures following credible early warnings and … novel research” are needed (p. 655). It does not forecast harm. That low-ratio fields later diverged in outcome shows that novelty is a weak signal of eventual harm, as §5.7 item 10 correctly puts it. It does not show that the ratio was mis-estimated.
Evidence. chunks LL2-27, p. 655; Box 27.4, p. 653 (novelty as a criterion for action).
Fix. “Used as a trigger (Box 27.4), a low knowledge-to-ignorance ratio proved a weak signal of eventual harm: nanotechnology, GM food and mobile phones, all grouped as low-ratio (p. 655), later diverged (hindsight LL2-27).”

17. The US–EU comparison is called “confirmed”; the source is more mixed#

Location. §3.1 Evidence (line 190): “The transatlantic pattern is confirmed”.
Problem. Hammitt et al. (2005) found “no significant difference in relative precaution over the period”, but also “a modest shift toward greater relative precaution of European regulation since about 1990” and “a diversity of trends across risks”. That supports the reports’ “institutions, not continents” claim. It does not confirm it without qualification.
Evidence. hindsight/LL1-00.md, line 285; hindsight/LL1-17.md, line 433.
Fix. “…is supported: no significant average US–EU difference over 1970–2004, with a modest European shift after about 1990 and diverse trends by risk (Hammitt et al. 2005).”

18. The hospital dose variation is cited to LL1-16, which gives a different figure#

Location. §3.2 note 5 (line 221): “doses varied more than tenfold between hospitals (LL1-16, pp. 174–175)”.
Problem. LL1-16, p. 175 says doses “can still vary by a factor of 100”. “More than an order of magnitude” is LL1-03’s wording (p. 35), and LL1-03’s chronology says “100x” (p. 36).
Evidence. chunks LL1-16, p. 175; LL1-03, pp. 35–36.
Fix. “doses for the same examination varied up to a hundredfold between hospitals (LL1-16, p. 175; ‘more than an order of magnitude’ in LL1-03, p. 35)”.

19. CFC-12 figures are labelled as all CFC releases#

Location. §4.7 table, row “Deployment outruns knowledge” (line 442): “CFC releases from 25 kt cumulative by 1948 to 300 kt a year by 1970 (LL1-07, p. 82)”.
Problem. These are the CFC-12 figures only. CFC-11 adds 5 kt cumulative and 207 kt a year.
Evidence. chunks LL1-07, p. 82: “cumulative releases of CFC-12 and CFC-11 were only 25 and 5 kilotonnes respectively … in 1970 the annual releases were 300 and 207 kilotonnes”.
Fix. “CFC-12 releases rose from 25 kt cumulative (1930–48) to 300 kt a year in 1970 (CFC-11: 5 kt to 207 kt)”.

20. The Minamata precedent is described as “similar evidence”, which is inference#

Location. §4.2 Standard of proof (line 340) and Appendix A, LL2-05 (line 1030): “though another prefecture had acted on similar evidence in 1950”.
Problem. The source says only that Kumamoto considered the Act “because Shizuoka prefecture had used the act to address an episode of shellfish food poisoning in 1950” (p. 98). Whether the evidence was comparable is the digest’s inference.
Evidence. chunks LL2-05, p. 98; digests/LL2-05.md, line 15.
Fix. “although Shizuoka Prefecture had used the same Act for a shellfish-poisoning episode in 1950 (LL2-05, p. 98)”.

21. PCB lag rounded up to “over 100”#

Location. §4.2 The lags (line 336): “to over 100 (asbestos, PCBs)”.
Problem. Table A2.1 gives asbestos 101 and PCBs “c. 100”. §3.4 reports this correctly (line 253).
Evidence. chunks LL2-A2, p. 702.
Fix. “to about 100 or more (asbestos 101, PCBs c. 100)”.

22. Proportionality is attributed to Chapter 1’s own view of precaution#

Location. §3.7 item 5 (line 288): “Chapter 1 treats proportionality as part of precaution (LL1-00, p. 13)”.
Problem. On p. 13, proportionality appears in a list of elements of the German Vorsorgeprinzip as elaborated for the Clean Air Act. Appendix A, LL1-00 describes this correctly (line 986). The tension with Chapter 17 is still arguable, but it should be stated with the right source.
Evidence. chunks LL1-00, p. 13.
Fix. “Chapter 1’s account of the German Vorsorgeprinzip includes proportionality (LL1-00, p. 13); Chapter 17 sets ‘proportionate and precautionary’ policies against each other (p. 194).”

23. Chapter 17 does not drop the word “illustrate”; it upgrades it#

Location. §3.3 Analysis (line 244) and §5.1 item 4 (line 494): “Chapter 17 drops Chapter 16’s caveats: ‘illustrative’…”.
Problem. Chapter 17 says “The case studies in this book both support and illustrate the need for the twelve late lessons” (p. 193). The caveat that goes is “rather than definitive”, replaced by “support”. The document’s point survives and is in fact stronger, but as worded it can be falsified by a word search.
Evidence. chunks LL1-17, p. 193.
Fix. “Chapter 17 turns ‘illustrative, rather than definitive’ (LL1-16, p. 169) into cases that ‘both support and illustrate’ the lessons (LL1-17, p. 193)”.

24. Michaels’s role is ambiguous#

Location. §2.1 (line 74): “David Michaels (then head of OSHA) on beryllium”.
Problem. He headed OSHA when the report was published, but he wrote the chapter on the George Washington University faculty, and footnote 1 says it gives personal views only. “Then head of OSHA” can be read as saying he wrote in that capacity.
Evidence. notes/LL2-06.md, line 16 (fn 1, p. 131).
Fix. “David Michaels (who headed OSHA when the report was published, writing in a personal capacity) on beryllium”.

25. Citation for the EPRS use of “4 of 88” points to the wrong file#

Location. §2.6 Afterlife (line 171): “reproduced the ‘4 of 88’ false-positive result … (context §1.5, §3)”.
Problem. context.md describes EPRS’s use of the reports and the working definition (line 214), but not the “4 of 88” reproduction. That is in critiques §5.3 (line 176) and hindsight LL2-00 (line 54).
Fix. Add “critiques §5.3; hindsight LL2-00” to the citation.

Log of checks that passed (sample)#

Against source chunks (page confirmed). - Author selection and scope. “would not have been approached” (LL2-00, p. 9; LL1-00, p. 12); Farman and Infante (LL1-00, p. 12); “distilled” (LL1-00, p. 3); “where sufficient is now known”, “spirit of the times”, “luxury of hindsight” (p. 11); “no suitable examples emerged” (p. 12); “seven structuring questions” (LL2-00, p. 10); “highly pertinent” (pp. 9, 11); reviewers named in LL1 (p. 6) but not in LL2 (p. 5). - LL1-16 and LL1-17. “relatively scarce”, “over a century”, “short-term economic and political interactions” and its four named cases (LL1-16, p. 168); “illustrative, rather than definitive” (p. 169); “virtual certainty” (p. 169); institutional against societal ignorance (p. 171); “compound uncertainty” (p. 172); “captive” (p. 174); “even if markedly inferior” (p. 177); “produced and owned” (p. 179); Swann quote (p. 181); Table 16.1’s lowest rung (p. 184); Table 17.1 and the 1965 asbestos date (LL1-17, p. 192), re-dated to 1930 in Table 27.1 (LL2-27, p. 656); “a key political decision with profound ethical implications” (p. 193); “ultimately a matter of political discourse” (p. 194). All twelve lessons are verbatim (LL2-00, p. 11). - LL2 framing and synthesis. “profoundly wrong” (LL2-00, p. 6); “knowledge and power” (p. 7); “irresponsible corporations” and “for the most part” (p. 11); “34 case studies” (LL2-27, p. 644); the working definition (p. 649); Box 27.4’s twelve criteria (p. 653); the knowledge-to-ignorance ratio (pp. 654–655); Table 27.2’s bands (p. 658); “risk takers or risk makers” (p. 658); “consolidated technological monopolies” (p. 659); “made by a few people on behalf of many” (LL2-28, p. 671); “today’s technology is now safe” and “harm expansion” (p. 672); “even at the cost of more false alarms” (p. 673); “homo-illogical” (p. 680) and “hydro-illogical” (LL2-15, pp. 360–361); Table 26.4’s ten against three (LL2-26, p. 635); the political-will footnote (p. 624); “part of the inertia” (p. 628); “does not have a good track record” (p. 640). - Annexes and costs. Table A2.1: asbestos 101, PCBs c. 100, benzene 81, radiation 65, BSE 10–17, TBT 5–30, median about 40 (LL2-A2, p. 702); “expands over time” (LL2-A3, p. 717); Stern’s 14.7 → 4.2% at 1.3 → 2.8% (LL2-23, p. 574); EUR 5–9 against below EUR 1 per kg (p. 573). - False alarms (LL2-02). 88 cases and four false positives; “high confidence” 67–95% (p. 18); about a third real risks and a third jury still out (pp. 20–21); seven lessons (pp. 34–35); swine flu’s 107 Guillain-Barré cases, six deaths and 40 million inoculated (pp. 26–28); nitrite, “nearly all bacon … within one year” (p. 25); “subjectivity” (p. 33); swine flu and saccharin reasonable ex ante (pp. 31–32). - Case chapters. BSE £1.5m, 1,200 cases, £4.2bn, 48%, “window dressing”, 17 months, MBM exports 12,553 t to 25,005 t (LL1-15, pp. 158–164); Harris’s 190,000 t against about 125,000 t and “drastic” (LL1-02, pp. 21–22); “decisively settled” (p. 18); 47% above advice since 2003 (LL2-17, p. 421); OR 88.4 (16.4–476.5) (LL2-07, p. 161); Box 7.1’s eight strategies (p. 155); vinyl chloride USD 278m against 90bn, “burden of medical uncertainty” and “frontiers” (LL2-08, p. 187); Minamata “all fish and all shellfish” and “Never stop it!” (LL2-05, p. 99); DBCP 5 ppm, 1.7 ppm odour accepted by USDA, and “impractical” (LL2-09, p. 205); DBCP emergency standard of 9 Sept 1977 (p. 206); the Bayer suits against three beekeeper leaders (LL2-16, p. 380); Gaucho “solely responsible…” (p. 379); Irgarol 1,700 ng/L (LL2-12, p. 267); the booster five-step cycle (p. 273); asbestos with smoking over 50-fold and the 25-year lung-cancer excess (LL1-05, p. 55); 66% asbestosis (p. 54); US and Canadian insurers in 1918 (p. 54); DES 2–10 million in utero (LL1-08, p. 87); the radiation limit of about 700 against 20 mSv (LL1-03, p. 33); Infante’s 5–10-fold risk and “significant risk” (LL1-04, p. 40); API 1948 and 50 ppm (p. 39); 54 per 1,000 (p. 41); the 1987 limit of 1 ppm (p. 41); Swedish 17-step chain and “not a neutral position” (LL1-09, pp. 95–96); “60 % gap closure” (LL1-10, p. 106); Santa Monica 71% (LL1-11, p. 114); France 1982 and “best information available” (LL1-13, p. 136); “none … precautionary” (p. 142); the Lamming committee disbanded and unpublished (LL1-14, p. 150); implementers as beneficiaries (LL2-14, about p. 337); “competitiveness” clause (LL2-22, p. 540; see issue 11); EEA withdrawal from IARC (LL2-21, p. 520); the Hardell funding footnote including Telia (fn 11, p. 517).

Against notes, hindsight, themes and external files (confirmed). - Context. Gee’s Friends of the Earth role (context, line 36); 22 January 2013 and doi:10.2800/73322; the innovation principle (9 Oct 2013 ERF letter, 2016, Horizon Europe recital 6); no court citation verified. - Hindsight, LL1 cases. Hammitt et al. 2005 (see issue 17 on wording); Chimed-Ochir 14.1-fold; about 7,000 BSE cases among about 50 million cattle tested; Stockholm listings 12 → 37; Pfizer paras 200–201; DES “possibly effective” in 1971; PFAS “forty years before the public health community”; the Manville Trust at 5.6%; mesothelioma peak of 2,205 male deaths in 2016; hormone sanctions of US$116.8m plus C$11.3m; Italy 8,000–22,000 deaths avoided; the OTM rule at about £2bn per death prevented; SO2 control preventing about 80,000 deaths a year; imposex 81% → about 21%; CFC-12 at about 44%. - Hindsight, LL2 cases. The French cnDAspe abolished by Loi 2026-403; the Dec 2025 pesticide-approval proposal; EU environmental-tax share down 16.8% on 2014; 29–35% of German flood residents unwarned; the Materion and USW joint draft; the Exxon 1985 MTBE memo and the rival refiner; 62 and 34 glyphosate-resistant species; five countries with about 89% of GM area; the 19–25% organic yield gap; eight chlor-alkali plants aged 42–83 years; 206 emergency authorisations; Cohen and Tubb; the vinyl chloride cost overestimate at about fourfold; the 80% producer share under the wastewater recast; Minamata’s 55,000 people and 32-person pilot; Fukushima’s 2,351 deaths; BPA 20,000-fold and 250,000-fold; saccharin’s 23 years; the jury-still-out tally (about 12, 3 and 3 of about 18 checked); leaded aviation fuel and TEL authorised to 2032; CDC’s 2012 change; EFSA 2018 uncertainty guidance and its carve-out; ecosystem accounts in 2024; the 2026 Omnibus directive; Aarhus review rights in 2021. - Notes and themes. Lieberman and Kwon supplying 28 of the 88 cases; Y2K and sludge dumping not among the 88; Bartrip; Bridges’s WTO role; the “six hormones”; Hosokawa’s cat 400; SO2 emissions of about 57 Mt read from Fig. 10.1.

Strand guard. No mention of contemporary technologies outside the reports. The only occurrence of “software” (lines 310 and 998) is LL1-07’s satellite-retrieval software, as in the source.

Fidelity review of 01-late-lessons-analysis.md#