Quarantine: the sources that could not be checked

Rejecting a quotation from publication and erasing it from the record are two different decisions.  This page is the record.

What this page is

The research behind the main article ran through thirteen dossiers and several hundred sources.  Most were downloaded and read.  Some were not.  This page lists every one that was not, together with the claim it was supposed to support, the exact routes that failed, and what someone should try next.

Nothing here has been deleted from the project.  A quotation that cannot be checked is unusable, but it is not thereby false, and a source that one machine could not reach on one evening is not a source that does not exist.  That second point is not theoretical.  Four sources on this page were written down as unobtainable and then obtained anyway, by a second route, inside the same research pass — they are in section G.

Three entries below touch a sentence the article actually prints, and each says which sentence.  They are Penny, Foulds & Hendy 1982, which the article flags in an aside; Reddy et al. 2017, cited from its abstract, which does support the sentence; and Strait & Grine 2004, where the article repeats one side's account of the other side's numbers.  A fourth, Barker & Lutzoni 2002 with Hipp et al. 2004, is kept only as a record:  the sentence it criticised has been rewritten and the article no longer rests on either paper.  Everything else on this page was considered and left out.

A note on one word.  "NOT OBTAINED" is the research dossiers' own flag, written at the moment the fetch failed, and it is reproduced here with the failure log intact.  It means "nobody on this side has read this," not "this is disputed."

The eight that would change the article if obtained

Ranked by how much the article would move.

1. Penny, Foulds & Hendy 1982 — the founding congruence experiment. Everything the article says about it comes from the same authors' 1986 follow-up. A1
2. Han et al. 2011, "Are transposable element insertions homoplasy free?" — the article's final verdict leans on rare genomic changes, and this is the paper that tests whether they are as clean as claimed. A2
3. Tëmkin & Eldredge 2007 — the strongest published case against artifact phylogenetics. If it contains a real test of hierarchical structure, that number belongs in the article's central section. A3
4. Strait & Grine 2004 — the article prints one sentence of an argument between two research groups while having read only one side's papers. B3
5. Klassen, Mooi & Locke 1991 — the only quantitative underpinning anyone has offered for "life's hierarchy is objective and a car's is not," which is exactly this article's question. The citation is now resolved; the body is still unread. A4
6. Siddall 1998 and Swofford et al. 2001 — the Farris zone, where parsimony beats likelihood. The article's "Where the methods fail" section is one-sided without it. A5
7. Springer & Gatesy 2016, "The gene tree delusion" — the hardest mainstream attack on the methods the article treats as settled. Abstract only. B2
8. Beck & Baillie — the strongest published result against the pseudoextinction argument the article uses to say morphology fails. Never resolved to a citation. C7

A. Paywalled — the source exists and nobody here has read it

A1. Penny, Foulds & Hendy 1982

The source.  David Penny, L. R. Foulds and M. D. Hendy, "Testing the theory of evolution by comparing phylogenetic trees constructed from five different protein sequences," Nature 297(5863):197–200, 20 May 1982.  doi:10.1038/297197a0.  PMID 7078635.

Why it matters.  This is the experiment the whole congruence argument descends from.  The article's section "The founding experiment" describes its five proteins, its eleven mammals, its metric and its null.  Every one of those details comes from Penny & Hendy's own 1986 follow-up in Molecular Biology and Evolution, which was obtained in full.  The article flags this in an aside rather than hiding it.

What was tried.  Unpaywall and OpenAlex both report is_oa: false with no OA location.  Semantic Scholar returns openAccessPdf.status = "CLOSED".  A scholar.archive.org full-text search on the exact title returned zero work hits.  Europe PMC holds a metadata-only record for PMID 7078635 with abstractText: null and one full-text link flagged "Subscription required".  nature.com serves the landing page; the PDF is behind a $39.95 paywall.  No repository copy was found at Massey University, where Penny and Hendy worked, or anywhere else.

What was obtained instead.  The publisher's own abstract and the complete reference list, by downloading https://www.nature.com/articles/297197a0 with curl and stripping the tags locally.  The abstract in full:

"The theory of evolution predicts that similar phylogenetic trees should be obtained from different sets of character data.  We have tested this prediction using sequence data for 5 proteins from 11 species.  Our results are consistent with the theory of evolution."

Note how weak that is.  "Consistent with," not "confirms."  A hostile reader who checks will find the founding paper does not claim to have proved common descent.

Two related sources also unobtained, both of which carry the 1982 null distribution:  Hendy, Little & Penny 1984, Systematic Zoology 33:283–291, table 4 (JSTOR/OUP, is_oa: false), which the 1986 paper says reproduces the 1982 table; and Steel & Penny 1993, "Distributions of tree comparison metrics — some new results," Systematic Biology 42:126–141 (OUP, is_oa: false), which has the exact null distributions.  Because neither could be read, the article's random-tree baseline was derived from first principles instead and checked against the one published figure that was in hand — Penny & Hendy 1986's f(d = 6) = 4.0 × 10⁻⁵ for 11 taxa.

What to try next.  A university library will have Nature 297 on the shelf or through an institutional subscription; this is a three-page 1982 paper and an interlibrary loan settles it.  Also worth a try:  Massey University's institutional repository, and Penny's own publication list, since several of his other papers turned up on personal and course servers.

A2. Han et al. 2011 — are transposable-element insertions really homoplasy-free?

The source.  Kin-Lan Han, Edward L. Braun, Rebecca T. Kimball et al., "Are transposable element insertions homoplasy free?  An examination using the avian tree of life," Systematic Biology 60(3):375–386, 2011.  doi:10.1093/sysbio/syq100.  PMID 21303823.

Why it matters.  The article's verdict says the sequence tree is confirmed by evidence of a different kind, and the first example given is retrotransposon insertions.  This is the paper that asks directly whether that class of evidence is as clean as the argument needs.  It was not read.  The article's counterweight — Lammers et al. 2017 on bears, where 278 and 265 insertions supported two wrong topologies against 168 for the right one — was obtained and is printed, so the section is not one-sided.  But the paper written to answer the question is missing.

What was tried.  Unpaywall (is_oa: false, oa_status: closed).  Europe PMC has the record with isOpenAccess: N, inEPMC: N and no abstract deposited at all.  PubMed efetch returns the citation with no abstract.  IA Scholar offers one Wayback-mediated Semantic Scholar PDF link, which returned HTTP 503 from scholar.archive.org and 404/503 from web.archive.org on its single CDX capture (timestamp 20190221011612).  Guesses at University of Florida lab reprint pages returned 404.  Mojeek returned nothing.

The route this entry once called the obvious next step has now been tried, and it fails.  bin/cdp-fetch.py drives a headful Chrome to academic.oup.com/sysbio/article-abstract/60/3/375/1664854 and gets a "Purchase Options" panel with no abstract at all — the OUP two-step that pulled seven other paywalled PDFs for this article, including Penny & Hendy 1986, does not work on this one.  Two further routes also fail:  the ScholarCommons repository record OpenAlex points at, scholarcommons.sc.edu/beaufort_naturalsciences_facpub/18, says "This document is currently not available here" and carries no bepress_citation_pdf_url;  and all five Wayback captures of the OUP article-pdf URL are 301/302 responses of text/html, so no bytes exist there either.  Europe PMC still returns no abstract.

Also missing, and not a fetch failure.  No published, quantitative rate of retroposon homoplasy — parallel insertion at the identical site, or precise excision — per marker could be found anywhere.  Nilsson et al. 2010 call it "extremely rare… but nevertheless not completely negligible" and cite their reference 9 without giving a number.

A3. Tëmkin & Eldredge 2007 — and the characterisation that is third-hand

The source.  Ilya Tëmkin and Niles Eldredge, "Phylogenetics and Material Cultural Evolution," Current Anthropology 48(1):146–154, February 2007.  doi:10.1086/510463.  Both authors at the American Museum of Natural History.

Why it matters, and what is wrong with how it is usually reported.  This is the strongest published case against applying cladistics to artifacts, and it is the paper most often cited for the claim that Baroque cornet data turned out to have no hierarchical structure.  The published abstract does not say that.  It says conventional phylogenetic techniques "do not provide a general theoretical and operational framewok [sic] for reconstructing material cultural history."  That is a weaker and different claim.  Whether the paper reports an actual test of hierarchical structure — a permutation tail probability, a g1 skewness value, a low consistency index, a NeighborNet — is unverified, and could not be checked.

The complete published abstract, rendered as bytes through bin/cdp-fetch.py page on the /doi/abs/ URL and not through any summarising tool:

"Cultural artefacts, like genes and languages, reflect their history.  The methodology of inference of that history, however, has been a contentious question.  Recent applications of biological phylogenetic methodology to infer historical patterns of material culture are often explicitly justified on the grounds that essentially similar processes underlie evolution in both biological and material cultural realms.  Conventional phylogenetic techniques, while helpful in some cases, do not provide a general theoretical and operational framewok [sic] for reconstructing material cultural history.  Critical analyses of the diversity patterns of two musical instruments, the stringed psaltery and the brasswind cornet, reveal paths of information transfer and the origins of innovation unique to the cultural context that are unlike those in biological systems."

The article therefore uses no number from this paper, and says in its own text that the paper could not be obtained and that the usual characterisation of it comes from third parties.

What was tried.  Unpaywall (is_oa: false, no OA locations); OpenAlex (one location, the publisher's); Semantic Scholar (openAccessPdf.status: CLOSED); bin/cdp-fetch.py file on journals.uchicago.edu/doi/pdf/10.1086/510463 (403, paywall redirect); the Smithsonian's repository.si.edu search API (403, bot wall); Internet Archive item sim_current-anthropology_2007-02_48_1 through bin/iasearch.py ("Item not available" — the microfilm serial is not lendable); Mojeek; and the extended book-chapter version, Tëmkin & Eldredge, "Material Cultural Macroevolution," in Prentiss, Kuijt & Chatters (eds.), Macroevolution in Human Prehistory (Springer 2009), 297–316, doi:10.1007/978-1-4419-0682-3_12, where link.springer.com/content/pdf/… returns HTTP 204 empty even through CDP-driven Chrome.

A second unresolved question about the same paper.  Whether anyone published a comment or reply to it could not be confirmed either way.  It sits in Current Anthropology's "Reports" section, which does not carry that journal's usual open-commentary treatment, and the paywalled landing page shows no comment links.

What to try next.  Nine pages, one interlibrary loan.  Failing that, Eldredge's own AMNH page, or JSTOR, which holds Current Anthropology — the project's JSTOR reading account can establish whether the article is there before anyone pays for it.

What was obtained instead, and it is worth reading.  Eldredge's open-access 2011 essay, "Paleontology and Cornets: Thoughts on Material Cultural Evolution," Evolution: Education and Outreach 4:364–373, doi:10.1007/s12052-011-0356-z, states the same thesis at full strength and in his own words, p. 368:

"In contrast, clear-cut lineages are very hard to come by in artifact systems—especially in the histories of manufactured products (like cornets), where information is transmitted rapidly within and between different makers."

It is an essay with no statistics.  Eldredge asserts the contrast; he does not measure it there.  The measurement, if it exists, is in the 2007 paper.

A4. Klassen, Mooi & Locke 1991 — the citation resolved, the body still unread

The claim it supports.  Douglas Theobald's 29+ Evidences for Macroevolution rests its consistency-index argument on it, and reproduces its critical-value curve as his Figure 1.2.1.  In Theobald's words:  "Klassen et al. took 75 CI values from published cladograms in 1989 (combined from three papers) and noted how they fared in terms of statistical significance. … Three of the 75 cladograms fell within the 95% confidence limits for random data."

Why it matters.  That 75-cladogram sample is the only quantitative support anybody has offered for the proposition that living things show statistically significant hierarchical structure while other objects do not.  It is the article's own headline question, answered in 1991 by a paper nobody on either side of the modern debate appears to have re-read.  Note also how low the bar is:  Theobald's own text says a consistency index of 0.2 is what random data give at 20 taxa, and 0.3 is highly significant.

The citation is now resolved, and neither version on file was right.  A Crossref bibliographic search returns one match:  Klassen GJ, Mooi RD & Locke A, "Consistency Indices and Random Data," Systematic Zoology 40(4):446–457, December 1991, doi:10.2307/2992239, also carried as doi:10.1093/sysbio/40.4.446.  The OUP landing page confirms all three authors and their affiliations — Royal Ontario Museum, Smithsonian, Toronto.  One dossier had recorded it as Klassen, Mumma & Wong 1991, Systematic Zoology 40:446–457 — right journal and pages, wrong co-authors;  another as Klassen, Mort & Wright 1991, Cladistics 7:119–130, which is wrong on every field.  Theobald's own page says only "Klassen et al. 1991."

What was tried.  The body is still unread.  Unpaywall and OpenAlex both report the paper closed, and the OUP landing page carries no abstract for a 1991 article.  The next route is JSTOR stable 2992239, which this project's reader account can reach.

A5. Siddall 1998 and Swofford et al. 2001 — the Farris zone

The sources.  Mark E. Siddall, "Success of parsimony in the four-taxon case:  long-branch repulsion by likelihood in the Farris zone," Cladistics 14:209–220, 1998.  And David L. Swofford, Peter J. Waddell, John P. Huelsenbeck, Peter G. Foster, Paul O. Lewis and James S. Rogers, "Bias in phylogenetic estimation and its relevance to the choice between parsimony and likelihood methods," Systematic Biology 50(4):525–539, 2001.  doi:10.1080/106351501750435086.

Why it matters.  The article's "Where the methods fail" section prints long-branch attraction, where parsimony is provably misleading.  The Farris zone is the mirror image:  two long branches that really are neighbours, where parsimony is right and likelihood can be biased toward pushing them apart.  Leaving it out makes the failure catalogue look one-directional when it is not.  The article currently says nothing about it, which is the correct handling of an unread source, but it is a real omission.

What was tried.  Unpaywall (both closed).  Semantic Scholar's public API returns no openAccessPdf, and returns no paperId at all for the Swofford DOI.  Semantic Scholar's internal endpoint lists, for Siddall, one alternatePaperLinks entry — Wiley's own pdfdirect URL, which 403s to curl.  IA Scholar exact-title search returned no results for either title.  OUP through the cdp-fetch cookie route:  Systematic Biology 2001 is gated.  The same route succeeded on a free 1987 MBE paper in the same session, so the failure is the paywall, not the fetcher.

Standing instruction from the dossier:  a writer must not state the Farris-zone result as though this project verified it.

What to try next.  Wiley is documented in agents/memory/research-source-access.md as 403-ing every header set, so Siddall will need a library.  Swofford's is more promising:  several of these authors post reprints, and the Wayback CDX index of a dead reprint directory has repeatedly been the winning route here — http://web.archive.org/cdx/search/cdx?url=<host>/<path>/*&output=text&collapse=urlkey.

A6. The rest

Each of these was attempted and failed, or was reached only through a summary.  None is cited in the article for a number.

B. Abstract obtained, body not

These were reached at abstract level through routes that return publisher-deposited text with no summarising model in the path — NCBI E-utilities, the Europe PMC REST API with resultType=core, or a dc.description meta tag pulled with curl.  The abstracts are quotable.  Nothing beyond them is.

B1. Rokas 2003, Prum 2015, Dunn 2008, Reddy 2017

Four Nature- and OUP-class papers behind the gene-conflict literature.  All four bodies are unread.

• Rokas, Williams, King & Carroll 2003, Nature 425:798–804. Tried Europe PMC, PMC, Unpaywall, OpenAlex, Semantic Scholar, IA Scholar (a Wayback copy exists but the Internet Archive rate-limited every request across three retries with backoff), Rokas-lab URL guessing and two web searches. Consequence: the specific figures for how many of the 106 genes supported each alternative yeast topology are unverified and must not be stated. The article never states them. It refers to this dataset only through Phillips, Delsuc & Penny 2004, which was obtained in full.
• Prum et al. 2015, Nature 526:569–573, and Dunn et al. 2008, Nature 452:745–749. Paywalled, no OA copy known to Unpaywall, OpenAlex or Semantic Scholar, no PMC record. Neither is cited in the article. Prum's data were obtained separately and are on disk.
• Reddy et al. 2017, Systematic Biology 66(5):857–879. The OUP paywall defeated the CDP route: academic.oup.com served "Get access" to a headful CDP-driven Chrome, and the article-PDF URL redirected to the abstract page, even though Unpaywall and Semantic Scholar both list the publisher PDF as an OA location. Dryad has the data, not the paper. This is the one the article does cite, for the finding that the bird-tree conflict tracks data type more than taxon sampling, and for the phrase "difficult to detect." Both are inside the abstract, verbatim, which was pulled from NCBI eutils for PMID 28369655 — so the article's sentence is safe, and the 22-page body, the specific nodes affected and the statistical tests are not. The complete publisher abstract was re-obtained independently by rendering the OUP article page through bin/cdp-fetch.py page and stripping the tags locally; it is on disk as sources/web/reddy-2017-conflicting-trees-abstract.txt, so the sentence is now confirmed by two routes. The body stays closed.

B2. Springer & Gatesy 2016, "The gene tree delusion"

The sources.  Mark S. Springer and John Gatesy, "The gene tree delusion," Molecular Phylogenetics and Evolution 94:1–33, 2016.  And its predecessor, Gatesy & Springer 2014, "Phylogenetic analysis at deep timescales," MPE 80:231–266.  Both Elsevier, both closed:  ScienceDirect 403s, Unpaywall reports no OA copy, Semantic Scholar's internal endpoint lists only the DOI.  Both abstracts were retrieved verbatim through the Semantic Scholar internal paper endpoint with a browser user-agent.

Why it matters.  This is the hardest mainstream-on-mainstream attack in the whole field, and the article does not currently mention it.  From the 2016 abstract:

"Song et al. (2012) reported an average locus length of 3.1 kb for the 447 protein-coding genes in their phylogenomic dataset, but the true mean length of these loci (start codon to stop codon) is 139.6 kb. … In addition, Song et al.'s (2012) dataset of 447 genes includes 21 loci with switched taxonomic names, eight duplicated loci, 26 loci with non-homologous sequences that are grossly misaligned, and numerous loci with >50% missing data for taxa that are misplaced in their gene trees."

Song et al. 2012 is the paper the article cites for the claim that the multispecies coalescent accounts for 77% of gene-tree variation in mammals — one of the article's stronger defensive points.  A published, named attack on that dataset's basic integrity belongs in the same paragraph.  It is not there because only the abstract has been read, and an abstract is not enough to referee a dispute this technical.

The reply was obtained in full:  Edwards et al. 2016, MPE 94:447–462, from the Faircloth lab's own reprint page.  So one side of the argument is on disk and the other is not, which is the worst possible state and the reason this entry ranks high.

What to try next.  Both authors host reprints.  Gatesy's and Springer's UC Riverside pages, and the Wayback CDX index of both, are the first stop.

B3. Strait & Grine 2004

The source.  David S. Strait and Frederick E. Grine, "Inferring hominoid and early hominid phylogeny using craniodental characters:  the role of fossil taxa," Journal of Human Evolution 47(6):399–452, 2004.  doi:10.1016/j.jhevol.2004.08.008.

Why it matters.  The article's calibration section prints Collard and Wood's result that seven craniodental clades above the 70% bootstrap threshold were all false, then adds one sentence:  "Strait and Grine replied that adding fossils fixes it.  Collard and Wood replied that Strait and Grine's own support values do not clear 70%."  The second half of that sentence is Collard and Wood's characterisation of a paper this project has not read.  Strait and Grine's bootstrap tables — the exact thing in dispute — are unread here.

What was tried.  Unpaywall (is_oa: false), OpenAlex (no OA location), ScienceDirect (403 to curl).  scholar.archive.org returned a work whose access/wayback PDF is a different paper — Organ et al. 2011 in PNAS, on feeding time.  That file was deleted rather than catalogued.  The complete publisher abstract was obtained from NCBI E-utilities for PMID 15566946, and it does state the rebuttal.

A general warning that came out of this one.  IA Scholar can mis-link a work to the wrong PDF.  Always open the first page and check the title before cataloguing anything from it.

What to try next.  Strait's or Grine's institutional page; Elsevier is otherwise a wall.  Until then the article should either read both sides or drop the sentence.

B4. The rest

C. Third-hand — the claim reached this project through somebody else's summary

A claim in this section may well be true.  What is missing is that nobody here has seen it in the source it is attributed to.

C1. The placental-mammal rooting quotation, relayed through Salvo

The quotation, as it sits in this project's own note file:

"Untangling the root of the evolutionary tree of placental mammals has been nearly an impossible task…"

How it arrived.  The note file gives a title, a journal and a year but no author, and its link ends salvomag.com/...#sthash.n1KyOhsE.dpuf.  The sthash/dpuf suffix is a ShareThis copy-to-clipboard artefact.  That means the text was copied off a Salvo magazine web page, not read at the journal.

Traced.  Emma C. Teeling and S. Blair Hedges, "Making the Impossible Possible:  Rooting the Tree of Placental Mammals," Molecular Biology and Evolution 30(9):1999–2000, 2013.  doi:10.1093/molbev/mst118.  PMID 23813980.

Verdict:  accurate.  The passage the note file quotes is the complete abstract of the editorial, word for word, retrieved through the Europe PMC REST API with resultType=core.  Once the authors are named it can be quoted.

What is still quarantined.  The body of the editorial is not open access and was not obtained, so its argument beyond the abstract is unread.  And the quotation is a 2013 snapshot of a question that has moved — the 240-genome Zoonomia-era analyses came later.  Anyone using it has to say what happened next, or a reader will supply it.

A note on which way it cuts.  The two papers the editorial introduces reached opposite answers on the same question:  Morgan et al. found Atlantogenata sister to all other placentals, Romiguier et al. found Afrotheria.  Romiguier's actual finding is more interesting than the disagreement, and it cuts against the sceptical use of the quotation:  GC-rich genes produce more gene-tree conflict and recover consensus nodes less often, because of GC-biased gene conversion driven by recombination.  That is a mechanistic account of which genes give bad trees.  The article does not use the quotation at all.

C2. "Venom has evolved 18 separate times in fishes"

What is verified.  The sentence exists in a University of Kansas press release, 5 July 2016, obtained by curl on sciencedaily.com/releases/2016/07/160705160206.htm:

"a paper appearing this week in the journal Integrative and Comparative Biology catalogs instances of venomous aquatic life, for the first time showing that venom has evolved 18 separate times in fresh and saltwater fishes."

Lead author William Leo Smith of the KU Biodiversity Institute; co-authors Jennifer Stern, Matthew Girard and Matthew Davis.

What is not verified, and this is the important part.  The paper was never read.  Worse, two dossiers attached the "18" to two different papers.  One traced it to Smith & Wheeler 2006, Journal of Heredity 97(3):206–217 — whose abstract says "12 clades," not 18 origins.  The press release plainly describes a different and later paper, in Integrative and Comparative Biology in 2016.  So the number, the paper and the count are all unsettled here.

What was tried on Smith & Wheeler 2006.  Unpaywall reports it OA at an OUP article-pdf URL; OUP served a Turnstile challenge to cdp-fetch.py, which correctly refuses to solve it, and the IA Scholar Wayback copy returned 503 on eight retries.

What the number means even if correct.  Eighteen is not an observation.  It is the minimum number of gains required when venom-gland presence is mapped onto an assumed fish phylogeny.  It presupposes the tree.  Under a different tree the count changes.  It is a convergence datum, not an incongruence datum, and the two arguments should not be mixed.  The article uses neither.

What to try next.  Find the 2016 Integrative and Comparative Biology paper by title and DOI first — the press release names the journal and the week — and stop citing the 2006 paper for the 2016 number.

C3. "In point of fact, there exists no such thing as the 'traditional textbook phylogeny'"

The quotation and where it came from.  This project's note file gives a garbled title — "Evolution of Animal Body Plans:  The Role of Metazoan Phylogeny at teh Interface Between Pattern and PRocess," Evolution and Development 2 — and a link to an image posted on X.  Two typos of exactly that kind are OCR artefacts, which means the text was read off a screenshot, not off the paper.

Traced, through Crossref and Europe PMC.  Ronald A. Jenner, "Evolution of animal body plans:  the role of metazoan phylogeny at the interface between pattern and process," Evolution & Development 2(4):208–221, 2000.  doi:10.1046/j.1525-142x.2000.00060.x.  PMID 11252564.  Sole author, then at the University of Amsterdam.

The full text is not obtained.  Tried:  Unpaywall (is_oa: false), Europe PMC (abstract only, no PMCID), scholar.archive.org phrase search (HTTP 500, no matching Wayback PDF), ResearchGate (403).  Wiley's onlinelibrary.wiley.com is documented in agents/memory/research-source-access.md as 403-ing every header set.

What the abstract does establish, verbatim from the Europe PMC record:

"In particular, 18S rRNA/DNA sequence analyses have yielded a new view of animal evolution that is often contrasted with a presumed traditional or classical view.  First, I expose this traditional view to be a simplified historical abstraction that became textbook dogma."

The quoted sentence is consistent with that thesis.  Its exact wording and its page are unverified.

Read the argument before using it.  Jenner is not saying phylogeny is unknowable.  He is saying the "traditional" metazoan tree that molecular phylogeneticists set themselves against was a straw man — a textbook simplification that never represented a real consensus among morphologists.  That is a criticism of how the molecular revolution was narrated, and it cuts against the claim that molecules overturned a settled morphological consensus at least as hard as it cuts for scepticism about trees.  A writer cannot have both.

C4. Quotations and numbers carried only inside a secondary source

Each of these appears in the research dossiers with an honest label saying so.  None is used in the article for a number.

C7. Beck & Baillie — a reference that was never resolved

Listed separately because it is the only item on this page where the citation itself could not be reconstructed.  Brady & Springer 2021 cite it as their reference 32 for the result that morphology plus hypothetical reconstructed ancestors recovers every placental superorder correctly.  That is the strongest published counter to the argument the article uses in its morphology section — Springer et al.'s finding that 13 of 18 mammal orders move to the wrong superorder when their DNA is deleted.  The dossier's own words:  "the article should not go to press without it."

Resolving it takes one step:  open Brady & Springer 2021's reference list and read entry 32.

D. Checked and failed — the search ran and the words did not come back

Every failure in this section may still be a tool failure rather than a wording failure.  Say so when repeating them.

D1. The second half of Berra's Corvette caption — now verified (2026-08-22)

The quotation in circulation, and it is everywhere in creationist writing:

"The point is that the Corvette evolved through a selection process acting on variations that resulted in a series of transitional forms and an endpoint rather distinct from the starting point.  A similar process shapes the evolution of organisms."

The source.  Tim M. Berra, Evolution and the Myth of Creationism (Stanford University Press, 1990), Figure 41 caption, p. 118.

What is confirmed.  The book was borrowed under controlled digital lending — archive.org item evolutionmythofc0000unse, the only one of four scans that lends — and searched with bin/iasearch.py.  A leaf-to-page map from the item's own BookReader/BookReaderJSIA.php endpoint fixes leaf 140 as printed p. 118.  The caption is confirmed up to the words "The point is that the", and then the index truncates.

What failed.  Six separate phrase probes for the tail — "the Corvette evolved through a selection process", "acting on variations", "series of transitional forms", "endpoint rather distinct from the starting point", "a similar process shapes the evolution of organisms", "evolution of organisms" — all returned no match.  The likely cause is that archive.org's inside.php truncates the paragraph text it returns, not that the sentences are absent.

Closed on 2026-08-22.  Figure 41 is a two-page spread:  p. 118 ends "The point is that the" and p. 119 carries the rest of the caption verbatim.  Read from the open-access scan B-001-000-890 (full OCR, no loan needed) and off both page images.  The second half may be printed as a quotation, cited to pp. 118–119.  The probes above failed for the reason this entry already suspected:  the archive's full-text index truncates the paragraph it returns, so the tail of the caption was never in the text that was searched.

What was recovered is more useful than the caption anyway.  Berra's own prose on p. 117, verbatim, and the article's section on designed objects is precisely about it:

"It would be like attending an auto show.  If you look at a 1953 Corvette and compare it to the latest model, only the most general resemblances are evident, but if you compare a 1953 and a 1954 Corvette, side by side, then a 1954 and a 1955 model, and so on, the descent with modification is overwhelmingly obvious (Figure 41).  This is what paleoanthropologists do with fossils…"

Two errors in this project's own copy of that passage were found and corrected in the process:  it had "paleontologists" for Berra's "paleoanthropologists," and it dropped the opening clause of the analogy, which is the clause that says distant models look unrelated.

D2. Phillip Johnson's "Berra's blunder"

Johnson coined the phrase in Defeating Darwinism by Opening Minds (InterVarsity, 1997).  Tried:  archive.org defeatingdarwini00john_1, the only scan, "not available to borrow at this time"; and arn.org/docs/johnson/berrasblunder.htm, 404.

The article must not characterise Johnson's argument from memory or from a second-hand creationist summary.  The substance of the objection does not need him — the Corvette series is a sequence of designed objects, each manufactured from the previous one's blueprints rather than begotten from it, which is exactly the pattern Berra offered as proof of unguided descent.  That observation is the article's headline question, and it can be made without quoting anybody.

D3. Ewert 2018's null-model files, and the arithmetic they would explain

Winston Ewert's dependency-graph paper was reproduced for this article from his released supplementary archive.  Two things in it could not be checked.

The null-model graph files for his Tables 3 and 4 are not in the 32-file supplement.  So the "vs Null" columns of both tables are unverifiable from the released data.

That matters because those columns do not balance.  Each table has three log-ratio columns against the same data, so the third must equal the second minus the first.  In Table 4 five of nine rows are off — UniRef-50 by 3,364, OMA by 1,360, HomoloGene by 6, Pfam by −2, OrthoDB by −1.  In Table 3 no row balances, with discrepancies of 232 to 648.

The arithmetic was recomputed independently and every figure above holds.  Recomputing each row from the printed Tree-vs-Null and Graph-vs-Null columns reproduces the Table 4 gaps exactly — 3,364 (UniRef-50), 1,360 (OMA), 6 (HomoloGene), −2 (Pfam), −1 (OrthoDB), with the other four rows exact — and the Table 3 gaps of 538 (JavaScript) and 608, 648, 232, 365, 307 for EvolSim 1 to 5, no row balancing.

Stated fairly, these are probably not errors of substance.  Each column is a separate importance-sampling estimate, so the three are independently noisy and need not be exactly additive, and the gaps are 0.001% to 3% of the values.  Two things still follow.  Monte-Carlo noise on the biggest datasets is of order 10³ bits, so no claim about these Bayes factors to more than three significant figures is supported.  And in every unbalanced row of Table 3 the printed figure is more favourable to the tree than the column arithmetic implies — whatever the cause, it is not a thumb on the scale in the author's own favour.

Also not obtained:  Ewert's graph-inference source code.  Only the 60-line scorer evaluate.py was published; Algorithms 1–3 exist as pseudocode on pp. 22–23 and would have to be re-implemented.  Searched dependencygraph.org, links from winstonewert.com, and the BIO-Complexity supplement list.

D4. Behe's own statement accepting common descent

Ewert classifies Michael Behe as accepting common descent, and that classification is the only verified anchor for it here.  Behe's own primary statement at page level could not be reached:  three archive.org copies of Darwin's Black Box return "Item not available," the Google Books API returned total: None, and ia-fts.archive.org timed out.  The article does not attribute a position to Behe.

E. Where the research itself is missing

Not fetch failures.  These are claims where a search of the literature found no such study, and the honest statement is that the measurement has not been made.

• No published phylogenetic or cladistic analysis of programming languages, or of Linux distributions, or of package-choice matrices. Searched; appears not to exist. This is why the article's Linux measurement had to be run in-house, and why it is described as apparently the first of its kind rather than as a replication.
• No large, clean, published design-without-descent character matrix exists. Every artifact matrix that could be obtained is contaminated by copying — Linux distributions are forks, canoes were traded and imitated, manuscripts are pure copying. Crowd-sourced product feature tables are the closest thing to a clean case, and they are unaudited, 14–24% unknown cells, and partly organised by target platform rather than by design lineage. The article states this as its own biggest limitation rather than pretending one of these datasets is the clean case.
• No survey of how much of the tree of life sits inside the anomaly zone. Degnan & Rosenberg proved the zone exists for five or more species; Linkem et al. 2016 is a single 16-taxon case study. Nobody should imply a survey exists, and the article does not.
• No study measuring what proportion of the tree of life is topologically stable across independent datasets. Searched, not found. Hinchliff et al. 2015 measures phylogeny-against-taxonomy agreement, which is a different quantity.
• No single paper states the composite statistic this argument actually needs — "X% of gene trees conflict, of which Y% is below 50% support and therefore noise." Jarvis 2014 comes closest. The article gives the two halves from two papers and says plainly that no one paper states it.
• No per-marker rate of retroposon homoplasy has been published. See A2.
• No reply to Ewert 2018 exists in print after eight years. The journal reserved a DOI for a critique; no critique was ever published. Four genuine citations, none in phylogenetics, none critical. That absence cuts against the mainstream side and the article says so.
• No independent replication of Ewert 2018 by anyone, on either side, before the one run for this article.
• Salichos & Rokas 2013's 1,070 yeast alignments and gene trees have no public deposit. The paper says "available from the authors on request." Searched figshare, Dryad, DataCite, Zenodo and Europe PMC; nothing. TreeBASE could not be checked at all — treebase.org timed out after 30 seconds with zero bytes received, and it is the one repository that was never ruled out. Thirteen years on, the single most-cited datum in the gene-conflict literature has no deposit. That is not evidence of fraud, and anyone treating it as such is wrong: the Rokas group deposited larger equivalent datasets in 2016 under CC BY and the finding has been reproduced on those. But anyone citing the 1,070-gene figure as an unchecked shibboleth should stop.
• MorphoBank project P773 (O'Leary et al. 2013, 4,541 morphological characters × 86 mammals) could not be downloaded. MorphoBank V4 is a JavaScript single-page application behind a bot filter: the permalink and project-overview URLs return a 1,105-byte JS shell to a browser user-agent, DownloadMatrix and api/v1/projects/773 return 403, and service.php paths 404. It would need the chrome-devtools route.

F. Two rules that came out of this exercise

A quotation that spans a column break is invisible to a plain search.  Two- and three-column PDFs interleave under pdftotext.  In one test, 8 of 9 phrases returned MISS on the PDF and 9 of 9 matched on the same paper's JATS XML.  In another, the key sentence stayed invisible until the page was cropped with -x 390 -W 205.  Never call a quotation missing without cropping the columns first.

pdftotext corrupts glyphs in older scans.  In the 1993 Systematic Biology scan it renders "≤" as "5" and "≥" as "2," turning "≥70%" into "270%."  And PMC/OUP HTML strips MathML, so equations vanish silently from an HTML-to-text extraction.  If a sentence reads "a support of at least should be used," the number was in an equation node.

G. Closed by a second route

These were written down as unobtainable and then obtained.  They are kept on this page on purpose, because the failure logs above are only trustworthy if the successes are shown beside them.  "It could not be found" is not "it does not exist."

• Salichos & Rokas 2013, Nature 497:327–331. Two dossiers recorded it as NOT OBTAINED after exhausting Unpaywall, OpenAlex, Europe PMC, IA Scholar and the lab's old reprint paths on as.vanderbilt.edu and lab.vanderbilt.edu, which 404 live and hold only redirect captures in Wayback. A third agent found the lab's current file server: https://wp0.vanderbilt.edu/rokaslab/wp-content/uploads/sites/191/2021/10/2013_Salichos_Rokas_Nature.pdf, plain curl, browser user-agent, HTTP 200. Every Salichos & Rokas number in the article — the 0.40 / 0.52 / 0.99 triple from Figure 1b, the 15-of-1,086 vertebrate figure, the Table 1 filtering audit — comes from that PDF, not from an abstract.
• Theobald 2010, the Nature formal test of universal common ancestry. Recorded as abstract-only after nature.com refused, then pulled on a retry loop against a Wayback capture of a download.bioon.com.cn mirror. Twelve pages, letter plus methods plus supplementary front matter.
• Penny & Hendy 1986, MBE 3(5):403–417. One dossier listed it unobtained while another already had the PDF, downloaded through the OUP two-step described in A2. It is the source of everything the article says about the 1982 experiment.
• Berra 1990. Six archive.org items hold the book; B-001-000-890 needs no loan and publishes full OCR, and evolutionmythofc0000unse lends. "Not available to borrow at this time" is also what IA returns when you already hold your one browse loan, so it is not proof that an item never lends. Testing every identifier costs seconds: curl -X POST .../services/loans/loan/ --data 'action=browse_book&identifier=<id>' returns {"success":true} or an error.

H. Routes worth trying first, for whoever picks this up

Collected from thirteen dossiers.  Each of these worked at least once on a source that had already defeated something else.

1. OUP / Silverchair. bin/cdp-fetch.py page <article-pdf-url> — a top-level navigation, not a file fetch — clears the Cloudflare challenge and redirects to a signed watermark02.silverchair.com URL, which curl then downloads. Plain curl, fetch-via-chrome.py and cdp-fetch file all get a 5.7 kB "Just a moment…" page instead. This route pulled seven paywalled PDFs for this article.
2. The Wayback CDX index of an author's dead reprint directory. http://web.archive.org/cdx/search/cdx?url=<host>/<path>/*&output=text&collapse=urlkey lists every archived file. Ziheng Yang's old UCL directory is preserved essentially complete and yielded three closed papers in one pass, as publisher PDFs with true pagination.
3. Fetch web.archive.org through the CDP browser, not curl. Direct curl to web/<ts>id_/<url> returned nginx 404s all session. The recipe that worked every time: cdp-fetch.py page <non-id_ wayback url> to set the browser's origin, then cdp-fetch.py file <same url with id_>. Without the page step the in-page fetch fails CORS.
4. Author and lab reprint pages beat repositories. profmarkcollard.com/publications/ hosts about 100 of Mark Collard's PDFs to plain curl, including three otherwise-paywalled papers this article cites. cladistics.coas.missouri.edu/pubs.html hosts about 90 of Michael O'Brien's. David Hillis's lab page at zo.utexas.edu/faculty/antisense/DownloadHillisPubs.html is live and open and has Hillis 1992, Hillis & Bull 1993 and Cunningham 1998 — three of the article's known-truth sources. The Rokas lab's wp0.vanderbilt.edu path is in section G.
5. Europe PMC resultType=core returns publisher-deposited abstracts for closed-access items, faster than fighting OUP or Wiley, and with no summarising model in the path. NCBI E-utilities efetch does the same. Both are quotable; both stop at the abstract.
6. bin/pmcget.py still works for PMC supplementary files at https://pmc.ncbi.nlm.nih.gov/articles/instance/<pmcid>/bin/<filename>. That is how the Polynesian canoe matrix was recovered when the Europe PMC full-text repo returned 403 and plain curl got a "Preparing to download…" stub.
7. The Anubis proof-of-work wall on hal.science is passable with cdp-fetch.py page followed by file on the same URL. The project memory file recorded that wall as impassable; that is now out of date.
8. archive.org rate-limits hard. web.archive.org and scholar.archive.org returned 503 constantly from this machine — one PDF took 9 retries, another 11, all on 15-second sleeps. A retry loop is mandatory, not optional, and several entries above may close on a quiet connection with no new technique at all.

Where these entries came from

Every failure log on this page is quoted or condensed from the thirteen research dossiers written for the article, which live in research/ beside it:  artifact-cladistics.md, baraminology-and-orchard.md, congruence-measures.md, datasets-for-analysis.md, ewert-dependency-graph.md, gene-tree-conflict.md, hgt-and-tree-of-life.md, known-truth-tests.md, morphology-vs-molecules.md, nested-hierarchy-argument.md, quote-audit.md, support-values.md and tree-building-methods.md.  Each carries its own "NOT OBTAINED" section with the full route log.

The companion page quotes holds the quotations that were verified, with their verdicts.