Limits of Evolutionary Change

"Hybridization works well as additive evidence. ... however, failure to hybridize is not subtractive evidence." — Todd Charles Wood, 2002¹

"In practice, this pattern rarely obtains." — Chad Thompson and Todd Charles Wood, on the clean boundary statistical baraminology looks for, 2018²

Summary

Young-earth creationists hold that God created separate groups of living things, called kinds, and that each kind has since split into many species.  The modern study of kinds is called baraminology, from the Hebrew bara, "create", and min, "kind".  It does not teach that species are fixed.  Its own founders say that species, and often genera, arise within a kind.^{3 4}  The claim is narrower:  that the tree of life is really a set of separate trees, and that change does not cross the gaps between them.

This article asks how that claim is tested, and what the tests have found.  Rates of change, new genes and the Ark's passengers belong to other articles and are linked, not repeated.

1. The Hebrew word does not define a unit of biology. Min occurs 31 times in the Hebrew Bible, always in the phrase "according to its kind".⁵ The leading creationist researchers on kinds agree that it is not a technical term and does not fix a rank.^{6 7} Todd Wood calls the reading of "after their kind" as a law of reproduction "a mistaken interpretation".⁸ So the edges of a kind must be found from biology, not read off the text.
2. Four different claims hide under the word "limit". A cluster in today's data, a barrier to interbreeding, an inference about shared ancestry, and a limit on what change could ever do are separate claims. Evidence for one does not establish the others. Much confusion in this debate comes from sliding between them.
3. Hybrids are good evidence, in one direction. If two species can produce a living hybrid embryo, their developmental programs are close enough to cooperate. Creationists and mainstream biologists read such a hybrid the same way: the parents are related.^{9 1} Hybrid chains join large groups: 72% of grass species, eight of thirteen duck tribes, four of five families of landfowl, and canids of different genera.^{1 1 10 11}
4. A failed cross proves nothing about ancestry, and creationists say so. Interbreeding fails within genera and even within species, and sterile hybrids arise inside groups both sides agree are related.^{12 13} Direct compatibility is not transitive: in the Ensatina salamanders, neighbours interbreed all the way around California, but the two ends live side by side without hybridizing.¹⁴ So no failed cross can mark the edge of a kind.
5. Hybrid evidence reaches different depths in different animals. Birds and frogs keep the ability to hybridize far longer than mammals do.¹⁵ Nearly half of known bird hybrids cross genera, against 11% of mammal hybrids.¹⁵ So hybrids can support family-sized kinds in birds, but most proposed mammal kinds rest on appearance, which creationists call the cognitum, not on hybrids.¹³
6. Statistical baraminology finds clusters, but its boundaries move. The method compares species across published trait tables. Its output depends on which traits are kept, how much data is missing, and which species are included. In Wood's own grass study, the full data set, mostly DNA, showed no boundary between grasses and other plants; a boundary appeared only when he kept 32 anatomical traits.^{1 1} A creationist statistician says its central test "really should be abandoned".¹⁶ Wood's re-analysis of 82 mammal studies kept two-thirds of the earlier verdicts, and two standard tests of whether the data form clusters at all "did not strongly support" it.^{17 17}
7. A gap in a trait table does not establish separate ancestry. A simulation of a single family tree, run for this article, produces a holobaramin-like pattern in 25.5% to 46.2% of trials under a stated rule when the group's stem is four to nine times its crown age.¹⁸ A long stretch with no sampled side branches can therefore produce a statistical gap despite known common ancestry. Fossils and other data are needed to test which history a real gap represents.
8. Several groups hold together; their outer boundaries are less secure. Hybrids, anatomy or statistics give reasons to join members of groups such as cats, dogs, bears, horses and grasses. That agreement is incomplete: the horse study could not test an outer boundary, and the grass results differed between DNA and anatomy. Elsewhere the proposed edges shift: in the landfowl, which statistics split and hybrids join; in pigs and peccaries; in the ruminants; in fossil humans; and in feathered dinosaurs. Jean Lightner, who wrote the Answers in Genesis surveys of kinds, puts the kind "anywhere from the level of the family to the level of the order".¹⁹

Verdict.  The creationist picture of kinds is coherent and testable in parts.  Its positive evidence is real: hybrids and clusters join many species into groups much larger than a genus.  Its negative evidence is weak.  The methods reviewed here do not distinguish a separately created group from a branch whose intermediates are absent, and their proposed boundaries sometimes disagree.  No kind boundary has been shown to be an absolute limit on change.  Equally, observed change within a proposed kind does not itself show that its boundary was crossed.  Common ancestry across such boundaries must be tested against fossils and genetic evidence, rather than inferred from an observed adaptation alone.

Part 1: What a "kind" is

The Hebrew word

Genesis 1 says that God made plants and animals "according to their kinds".  The Hebrew word is min.  A count of the tagged Hebrew text finds it 31 times:  ten in Genesis 1, seven in the Flood story, nine in Leviticus 11, four in Deuteronomy 14, and once in Ezekiel 47:10.⁵  Pete Williams reported the same count in 1997, and Jean Lightner repeated it in 2021.^{7 6}  Every use comes with a preposition and a pronoun:  "according to its kind" or "according to their kinds".

The word is not used like a modern rank.  In the food laws, Israel may not eat "every raven after its kind" (Leviticus 11:15).  Lightner reads the construction as naming "two or more subgroups within the larger, named category", so the verse speaks of several kinds of ravens, not one raven baramin.⁶  Williams concluded that a fixed taxonomic level for min "cannot yet be substantiated".⁷  Lightner's summary of the Hebrew studies is blunt:  the word "is unlikely to carry the taxonomic significance that the word ‘baramin’ implies."⁶

Creationists also disagree about whether "after its kind" limits who can mate with whom.  Lightner and her Answers in Genesis colleagues wrote in 2011 that "The biblical text provides strong evidence for each kind being a reproductive unit."¹³  In the same paper they conceded that "Scripture does not emphatically state that life reproduces only after its own kind", and called the reproductive reading "a very strong inference".¹³  Todd Wood, author of the field's tutorial and its main reviews, rejects it outright.  He calls Frank Marsh's reading of the phrase "as a reproductive command or law" "a mistaken interpretation", and writes that "the Bible does not teach that the min (Hebrew for ‘kind’) are reproductive units".^{8 8}  Wood's conclusion follows:  the discontinuities creationists expect "are not clearly and irrefutably taught in the Bible and are therefore open to empirical testing".¹⁷

That is the right frame for a skeptic, too.  The question is empirical.

Not fixity of species

A common picture of creationism is that every species was made as it is.  That is not what baraminology teaches, and has not been for eighty years.  Marsh, who coined the word baramin in 1941,¹ wrote in 1947 that the test of interbreeding, used on small groups, "merely demarks physiological races which have developed and are developing within the kind".²⁰  Kurt Wise, who set out the modern method in 1990, wrote:  "Microevolutionary processes, including speciation occur within the baramin. Extra- and interbaraminic evolution is considered impossible."³  Wood and three colleagues put it this way in 2003:  "Rather than fixity of species, we advocate the persistence of baramins."⁴  The sibling articles on observed speciation and the creation model of rapid speciation show that young-earth models now need fast speciation, not less of it.

So the debate is not about whether species change.  It is about whether change has edges.

The working terms

Baraminology uses a few terms, set out by Walter ReMine and Kurt Wise in 1990 and redefined by Wood and colleagues in 2003.^{3 4}

1. Baramin: the created kind and all its descendants. Wood and colleagues call it "a purely theoretical construct", since no one can know every member.⁴
2. Monobaramin: "a group of organisms that share continuity, either genetic or phenetic." It is a group known to belong together, without regard to where the edges are.¹
3. Apobaramin: "a group of organisms that is discontinuous with everything else." Bats are the usual example: creationists think bats are cut off from other mammals, but do not know how many kinds of bats there are.¹
4. Holobaramin: both at once, "a complete set of organisms that share continuity among themselves but are discontinuous with all other organisms." Wood calls it "roughly what we call the ‘Genesis kind’".¹
5. Polybaramin: a group made of parts of different kinds, the error to avoid.⁴
6. Cognitum: "a group of organisms that are naturally grouped together through human cognitive senses", a term from Roger Sanders and Kurt Wise.^{13 21} It is the kind as people see it.
7. Basic type (German Grundtyp): the European creationists' term, defined only by hybridization. Wood counts a basic type as a monobaramin, since hybrids can join species but cannot show an edge.¹

"Continuity" and "discontinuity" carry the weight.  Wood and colleagues define them as "significant, holistic" similarity and difference, and add:  "a lack of continuity does not, by default, constitute evidence of discontinuity. Significant, holistic difference must be demonstrated."⁴  This is the burden the young-earth side sets itself, and it is the right one.

Four claims, not one

The word "limit" covers four different claims.  They need different evidence, and the debate is clearer once they are kept apart.

A cluster can exist because of separate creation, or because the side branches that once filled the gap went extinct.  A reproductive barrier can arise between species that both sides agree are related, such as horses and donkeys.  Separate ancestry is a claim about history, and so needs historical evidence.  An absolute limit is a claim about what could happen, and needs a mechanism.  Baraminology mostly measures the first claim, sometimes the second, and draws the third as an inference.  It rarely addresses the fourth, and Wood says he does not try to:  "I personally do not use baraminology to argue that an organism or group of organisms could not have evolved."⁸

Part 2: What hybrids show, and what they cannot

Positive evidence

Marsh's original test was simple:  "Two organisms are members of a kind if their germ cells will join in true fertilization."²⁰  Modern creationists keep the idea but tighten it.  Lightner's team at Answers in Genesis says that "fertilization is clearly insufficient evidence of hybridization", and asks for an embryo that develops far enough to show that both parents' genes are working together, such as an advanced blastocyst.¹³  Wood explains why a hybrid counts:  "The ability of members of two different species to produce offspring strongly indicates that they share basic genetic machinery and a common developmental path".¹  Siegfried Scherer adds that the hybrid need not be fertile, "since sterility can be caused by rather minor genetic changes."¹²

Mainstream biologists read a hybrid the same way.  Richard Van Gelder, curator of mammals at the American Museum of Natural History, argued in 1977 that if species are defined by reproductive isolation, "then species in one genus should not be capable of breeding with species in other genera."⁹  He quoted George Gaylord Simpson:  "successful hybridization is practically conclusive proof of chromosomal equivalence".⁹  Van Gelder then proposed merging many mammal genera, including foxes into Canis, Panthera into Felis, and Bison into Bos.  The creationist use of hybrids is therefore not special pleading.  It is the standard reading of what a hybrid means.

How far hybrid chains reach

Creationists also count an indirect link.  Scherer's rule is that "Two individuals belong to the same basic type if they have hybridized with the same third organism", and more generally, two species belong together "if they are connected indirectly by hybridization".^{1 12}  This is a rule about ancestry, and it is sound as such:  if A is related to B and B to C, then A is related to C.

Chains of hybrids join large groups.

1. Grasses. Wood's 2002 survey found hybrids linking 12 of 46 grass tribes, about 7,220 species. "Consequently, I can assign 72% of the Poaceae to one hybridization-defined monobaramin."¹
2. Ducks, geese and swans. "Of the 13 tribes of the duck family Anatidae, hybridization connects eight. The remaining five represent tribes of 1–3 species each."¹ Allan Wilson's laboratory noted in 1975 that many bird hybrids cross subfamilies or families, citing "the duck-goose cross and the turkey-chicken cross".¹⁵
3. Landfowl. Lightner reports that four of the five families of chickens, pheasants, turkeys and their relatives "are united by hybrid data".¹⁰
4. Dogs and foxes. Van Gelder recorded a litter from a male coyote and a female red fox at the Cohanzick Zoo in New Jersey in 1973; both pups died within three days.⁹ It was never checked genetically. In 2023 a Brazilian team used chromosomes, mitochondrial DNA and nuclear genes to identify a wild female as "a first-generation hybrid between a dog (Canis lupus familiaris) and a pampas fox (Lycalopex gymnocercus)", parents of different genera.¹¹ Its fertility is unknown.¹¹
5. Bears. Tom Hennigan reports hybrid offspring among "seven of the eight species" of living bears.²²
6. Sturgeon and paddlefish. In 2020 Hungarian fisheries researchers crossed Russian sturgeon with American paddlefish, the first hybrid "between members of the family Acipenseridae and Polyodontidae".²³ Between 62% and 74% of the hybrids were still alive a month after hatching.²³ Whether they can breed was not tested.

The young-earth reading of these chains is straightforward:  each is one created kind that has split into many species, and the chains show it.  The mainstream reading is also straightforward:  each chain is a clade whose members have not yet lost compatibility.  The sturgeon case shows how far that can stretch.  The authors give the two lines a separation of 184 million years and suggest that sturgeons' slow evolution kept their genomes compatible.²³  A creationist would put sturgeons and paddlefish in one kind.  Either way, the cross shows that the two families share a developmental program that still works together.

Negative evidence: why a failed cross proves nothing

The young-earth literature is clear that a failed cross does not show separate kinds.

1. Wood: "failure to hybridize is not subtractive evidence. There are too many factors that can cause reproductive isolation that have nothing to do with baraminic status."^{1 1}
2. Wood and colleagues, 2003: "failure to hybridize does not constitute evidence of different baramins".⁴
3. Scherer: "in some cases hybridization fails when members of the same genus or even races of the same species are tested. Successful hybridization can be prevented by very small genetic changes."¹²
4. Lightner and colleagues: "lack of hybridization data is inconclusive." They put the loss of interfertility within a kind down to "genetic changes (mutations) that have accumulated in one or both species since the Fall".^{13 13}

Their own examples show why.  Cattle and bison form live hybrids, but "the males are sterile."¹³  Sheep and goats sit in one baramin, yet in one study the hybrid embryos "died within 5 to 10 weeks", and in the reverse cross "fertilization did not occur".¹³

There is a second reason, and it is logical.  Direct compatibility is not transitive.  A can breed with B and B with C while A cannot breed with C.  The best-known case is the Ensatina salamanders of California, a chain of forms around the Central Valley.  David Wake found gene flow between neighbouring forms all the way around, with hybrids that are "healthy and fertile" where forms meet.¹⁴  Yet where the two ends of the chain meet in southern California, "the two forms are sympatric with no evidence of past or present hybridization".¹⁴  Ensatina is a single species complex, well inside any kind a creationist has proposed.  So within one kind, two living forms can be reproductively isolated.  A genome study of the greenish warbler, the other textbook ring, found the picture messier:  gene flow around the ring has been broken in places, and the two Siberian ends "have occasionally interbred".²⁴  Either way, interbreeding is a poor yardstick for ancestry.

Scherer's indirect rule handles this correctly.  It joins A and C through B.  But it works in one direction only.  It can make a kind bigger.  It can never draw its edge.

Why hybrid evidence runs out sooner in mammals

The power of the hybrid test differs from group to group.  In 1975 Ellen Prager and Allan Wilson measured protein differences between species pairs known to hybridize.  On their molecular clock, hybridizable birds had separated "about 22 million years ago", frogs about 21 million years ago, "while for hybridizable placental mammals it is only 2 to 3 million years."¹⁵  The dates rest on a mainstream molecular clock, but the raw measurements point the same way.  Hybridizing bird pairs differ in the blood protein albumin by 12 units on average, and hybridizing mammal pairs by 3.2 units, although the authors judge bird albumin to change about three times more slowly.¹⁵  Benjamin Fitzpatrick repeated the test with DNA in 2004 and confirmed "that diverging mammals lose the ability to form viable hybrids faster than birds."²⁵  Suzanne Edmands's review warns that the decay is uneven, and that "divergence time is clearly an unreliable predictor of the consequences of hybridization."²⁶

The counts follow the same pattern.  "Nearly half (44%) of the known cases (about 1000) of hybridization in birds are between genera", while for placental mammals "only 11 % of the cases are intergeneric."^{15 15}

There is a further limit.  Records exist for only a small share of species.  Klaus Schwenk and colleagues estimated in 2008 that about 1% of animal species, with a range of 0.1% to 3%, are recorded hybridizing, well below the older estimate of 10%.²⁷  Scherer notes that basic types can only be drawn for groups "that not only are very well known but also were kept in captivity or cultivation".¹²

The consequence for kinds is direct.  Hybrid evidence can support family-sized kinds in birds and in some plants.  In mammals, hybrids mostly link species within a genus.  When Answers in Genesis estimated 137 living mammal kinds for the Ark, the step from genus to family rested mostly on the cognitum, the judgement of the eye.  Its method paper says so:  "Where hybrid data is lacking, we have chosen to use a cognitum method", and "the family level may be used as the default level".¹³  The reason given for trusting the eye is theological:  "Adam would have recognized created kinds by sight."¹³  That may be true, but it is not a measurement.  The Ark logistics article shows how much the count of kinds swings with this choice.

Part 3: Statistical baraminology, its inputs and its sensitivity

Hybrids cannot test fossils, and most living species have no hybrid record.  So since 1998 creationists have also used a statistical method, built on the trait tables that systematists publish to build family trees.

How the method works

1. The input is a published character matrix: a table of species against traits, such as tooth shape or skull bones, each coded as a state. Wood's tutorial says the matrices "may exclude taxa deemed baraminologically significant, or they may simply have too few taxa or characters", and urges creationists to collect their own data.¹ In practice, nearly every study reuses matrices built by evolutionary systematists.
2. Missing data are filtered. Each trait's "relevance" is "the percentage of taxa for which a character state is known". Traits below a cutoff are dropped.¹ Robinson and Cavanaugh set the cutoff at 95%, and said they chose it "rather arbitrarily".²⁸ Later studies used 90%, 75%, or a cutoff chosen matrix by matrix.^{16 29 30}
3. The baraminic distance between two species is "the percentage of characters in which the two species differ".¹
4. Distance correlation (BDC) compares two species' distances to every other species. If their profiles rise and fall together, the correlation is positive. "Significant positive correlation indicates that the two species are members of the same monobaramin and significant negative correlation indicates that the two species are discontinuous".¹
5. A picture of the distances in two or three dimensions, by multidimensional scaling, helps the eye see clusters. Since 2021, cluster methods such as medoid partitioning and fuzzy analysis have been added.¹⁷

A holobaramin is called when a group is continuous inside and discontinuous with everything outside.  Thompson and Wood describe the ideal and add, frankly:  "In practice, this pattern rarely obtains."²

What the method has found

The largest survey is Chad Thompson and Todd Wood's 2018 study of 82 analyses of fossil and living mammal groups.  "We discovered evidence of discontinuity around a holobaramin in 57 (69.5%) of the 82 analyses."  The rest were monobaramins (3) or inconclusive (22).²  Of 59 putative holobaramins, "49 of which corresponded to families, seven to subfamilies (or portions of families), two to superfamilies (or multiple families), and one (Sirenia) to an infraorder".²  Wood's 2006 review of 66 earlier studies reached a similar view:  "it appears at this time that Price’s suggestion that the family is an approximation of the “created kind” may be correct."⁸

That is a real pattern.  Groups of living and fossil species that taxonomists call families often show up as separate clusters.  Creationists predicted that.  The next question is what drives the clusters.

Sensitivity to the traits chosen

Robinson and Cavanaugh, who founded the method, warned:  "Character selection, not the method of analysis, is expected to be the primary factor affecting baraminic hypotheses."¹⁷  Three cases show it.

1. Wood's grasses. The Grass Phylogeny Working Group matrix had 7,025 characters for 62 grass genera and 4 outgroups. After the relevance filter, 2,119 remained.¹ With those, 98% of grass pairs correlated positively, but so did 53% of the pairs between grasses and the outgroups, and "only 6% had significant negative correlation".¹ That made the outgroups look like grasses. Wood then dropped the DNA characters. "With the morphology-only dataset, 21 characters were eliminated due to low relevance, and 32 characters were used". On those 32, the grasses separated from the outgroups, apart from two small tribes that sided with the outgroups.¹ Wood concluded that DNA "biases baraminic analysis towards too much similarity".¹ A skeptic will note that the boundary appeared only after most of the data were set aside.
2. Duplicated traits. Colin Reeves re-examined Wood's turtle matrix and found that "the set of 115 characters reduces to just 51 distinct characters in the sense of supplying unique information".³¹ Repeated traits count twice in every distance.
3. Evolutionary coding. Reeves also found that the turtle traits were coded 0 or 1 by whether they were thought primitive or derived: "there is a rationale—an evolutionary one!"³¹ That criticism needs a qualification. Reversing 0 and 1 for a trait leaves every agreement and disagreement unchanged, so it leaves baraminic distances unchanged. The choice of traits and the judgement that two structures represent the same trait matter; the labels for primitive and derived states do not by themselves force a boundary.

Sensitivity to missing data

Fossil matrices are full of gaps.  In the 2016 hominin matrix Wood used, "only 4,774 (50.9%)" of the possible entries are scored.¹⁷  The dinosaur matrices analysed by Neal Doran and colleagues in 2018 were missing between 24.1% and 62.4% of their entries.²⁹  Those authors found that removing the most incomplete species changed the picture:  "removal of problematic taxa results in different spatial geometries."²⁹  Reeves notes that the relevance cutoff is usually 95%, "occasionally 90% or even less;  whether the results are sensitive to this choice is rarely mentioned."¹⁶

Sensitivity to which species are sampled

The result also depends on how many species are in the table.  Wood showed this himself when another creationist, Jean O'Micks, excluded Homo naledi from the human kind using six species.  "Here, with the same 87 characters, reducing the sample size to just six taxa results in only one correlation within genus Homo."³²  Wood's conclusion was that "reducing the taxon sample size can result in spurious baraminic distance correlations not seen in larger taxon samples."³²

How the table was built matters as much.  Wood notes that the mammal matrices in the 2018 survey were chosen for "a good sample of ingroup taxa with a smaller sample of outgroup taxa."¹⁷  That is how systematists design a table to resolve one family.  It is also the design most likely to show a gap between that family and everything else.  Wood himself cautions that "one should not conclude that the presence of different clusters alone constitutes evidence of discontinuity. Multiple clusters could be observed as a result of the biased selection of characters".¹⁷

The statistics, attacked and defended

In 2021 Colin Reeves, a statistician at Coventry University and a creationist, published a two-part critique in Answers in Genesis's own journal.  He found that the correlations are not independent, that the significance tests assume a distribution the data do not have, that many tests are run without correction, and that the character bootstrap breaks its own assumptions.  His conclusion:  "The methods used by statistical baraminology have multifarious flaws, despite the surface appearance of rigor and sophistication."  And:  "BDC really should be abandoned."^{16 16}

Reeves did not dismiss every result.  Re-running Wood's turtle data with standard cluster analysis, he found that the original two clusters "can be supported, as well as the detailed cluster composition."³¹

Wood answered by re-running all 82 mammal analyses with the methods Reeves recommended.  The findings cut both ways.

1. Pearson BDC performed about as well as the newer cluster methods.¹⁷
2. Of 219 clusters with more than one species, "151 (68.9%) were monophyletic", that is, they match branches of the published evolutionary trees built from the same traits.¹⁷
3. Holobaramin or probable holobaramin verdicts fell from 69.5% to 57.3% of the matrices, and two-thirds of the verdicts (54 of 82) were unchanged.¹⁷
4. "Most interesting of all, the Hopkins statistic and dip test did not strongly support the presence of clusterability in the distance matrices."¹⁷ Wood reads the average Hopkins value of 0.79 as showing that clusters are present, and explains the weak dip test by the small number of species and the ingroup-heavy design of the tables.

So the method's leading user, answering its leading critic, finds that 12 of the 57 earlier holobaramin verdicts, about a fifth, became inconclusive, that most clusters are branches of the evolutionary tree, and that the two tests of clustering give mixed results.¹⁷

A test for this article: does one family tree produce "kinds"?

A skeptic's natural objection is that a tree with missing branches can show gaps.  The question is whether BDC would call those gaps discontinuities.  For this article, characters were evolved on simulated family trees of 400 species, all descended from one ancestor, and BDC was run on 23-species tables with 70 traits, the size of the tables in the mammal survey.  Each table had 18 species from one branch and 5 from elsewhere.  A holobaramin-like pattern was counted only if at least 80% of within-branch pairs were significantly positive, at least half of the branch-to-outsider pairs were significantly negative, and none was positive.  Those thresholds are this article's stated test, not a reproduction of every published baraminologist's decision rule.¹⁸

The results support each side on one point.

1. Short stems rarely give a clean pattern in this model. Among randomly selected branches whose stems were shorter than their crown ages, 0.3% to 1.3% of sampled tables met this article's holobaramin-like rule. This particular simulation does not turn every branch of a family tree into a kind.
2. A long unbranched stem can produce the pattern. In the controlled runs, a stem four times the branch's crown age produced a holobaramin-like pattern in 25.5% of tables, with 95% of within-branch pairs continuous on average. At nine times the crown age, 46.2% of tables met the rule.

In a common-descent history, extinction or incomplete sampling can leave a long stem with no sampled side branches.  In a separate-creation history, the group has no ancestral connection to other kinds.  The simulation shows that known common ancestry can produce the pattern claimed as evidence of separate ancestry.  It does not show that any particular real group has an extinct stem.  A BDC discontinuity alone therefore cannot choose between these histories.  Fossils and other evidence are needed, and the fossil record and transitional sequences articles take up that question.  The simulation uses independent, unweighted and complete traits, unlike real tables.  It shows a possibility under one stated model, not how often the published method gives a wrong verdict on real organisms.

Part 4: Where the edges are tested

The real test of a proposed boundary is whether independent kinds of evidence put it in the same place.  Some groups pass.  Others do not.

Groups that hold together

1. Cats. Hybrids cross the two subfamilies of cats, and cats look alike to anyone. Lightner: "it is natural to place the level of the kind at the family."¹⁹ Thompson and Wood's survey also lists the cat family as a holobaramin.²
2. Dogs. "There is considerable hybrid data, including a cross between a coyote and a red fox."¹⁹ The 2023 dog and pampas fox hybrid now gives genetic support for one cross between genera.¹¹
3. Bears. Hybrids join seven of eight living species. The odd one out is the giant panda: "There is no evidence that the giant panda has hybridized with any other bear species." Hennigan keeps it in the bear kind on other traits, and admits that the panda's differences "make an ursid apobaramin and holobaramin hard to demonstrate."^{22 22}
4. Horses. David Cavanaugh, Wood and Wise found "that all nineteen species belong to a single monobaramin" in a matrix of fossil horses.³³ They also said what it did not show: "we cannot conclude that any discontinuity has been discovered or that a holobaramin has been identified", since the table had no non-horse outgroup and 64% of its traits were teeth.^{33 33} The horse series is one kind in creationist eyes; the horse series article covers it.
5. Grasses. Hybrids and morphology together gave Wood a holobaramin of about 10,000 species, "the largest holobaramin identified to date."¹

These cases give positive evidence for grouping species together.  They do not establish every outer boundary.  The horse study lacked an outgroup, the panda remains difficult, and the grass results differed between anatomical and molecular data.  Mainstream biology also recognizes these families as natural groups; the disagreement concerns their relationship to the groups outside them.

Groups where the edge moves

1. Grass, again. Before Wood's study, a creationist report on one grass had suggested "that the boundaries of the ‘created kind’ lie within the genus Muhlenbergia."¹ Wood's hybrids moved the edge from one genus to almost the whole family.
2. Landfowl. Hybrid records join four families of chickens, pheasants, guineafowl, quail and curassows. A statistical study of the same order, mostly from bones, split them into four groups. Lightner wrote: "This discrepancy between hybrid data and the results of statistical baraminology techniques is an important reminder that the latter, though valuable in helping to estimate the level of the kind, should be considered tentative."¹⁰ Here the two creationist tests contradict each other. If the hybrid records are right, three of the four statistical groups are one kind.
3. Pigs and peccaries. Lightner joins them by appearance: "So in this case the baramin appears to be at the level of the superfamily (Suoidea)."¹³ Thompson and Wood's statistics found "putative discontinuity observed between the suids and the peccaries (Tayassuidae)".²
4. Ruminants. Lightner wrote that the evidence "hints that one created kind could have given rise to all in Ruminantia", then split them "so as to avoid underestimating the number of kinds on the Ark."¹⁹ That is a boundary drawn for bookkeeping, as she says.
5. Fossil humans. Wood's analyses place Homo habilis, Homo naledi and Australopithecus sediba in the human kind.³⁴ Another creationist, Jean O'Micks, first wrote that "Placement of H. naledi in the human baramin is supported", then concluded months later that naledi was "Probably Not Part of the Human Holobaramin".^{35 36} A 2021 re-analysis by Sinclair and Wood could not place Homo floresiensis: "Given the lack of consistency, we cannot make a firm conclusion regarding the status of H. floresiensis." It also found Australopithecus africanus moving into the human cluster in one setting, "yet another example of inconsistency".^{34 34} In the same analyses, "Undisputed nonhuman apes, such as chimpanzees, gorillas, Ardipithecus ramidus, and Australopithecus afarensis, never cluster with undisputed humans."³⁴ The creationist sorting of hominin fossils is treated in its own article.
6. Birds and dinosaurs. Phil Senter, a paleontologist, used the creationists' own method to argue for "morphological continuity between Archaeopteryx, other early birds, and a wide range of nonavian coelurosaurs".³⁷ Wood answered in the same journal that "Distance correlation using Senter's set of taxa and characters supports his conclusion of morphological continuity, but other sets of taxa with more characters do not."³⁸ Senter's second study found that "Morphological gaps in the known fossil record separate only seven groups from the rest of Dinosauria."³⁹ Matthew McLain and colleagues, creationists, then proposed at least eight kinds of feathered dinosaurs, but conceded that "we could not find a way to separate theropod dinosaurs and birds overall into two groups" and that "creationist arguments from discontinuity are not yet conclusive."^{30 30} The Archaeopteryx article follows this dispute in detail.

The rank of the kind

The rank at which kinds are drawn has drifted with the data.  Wood's 2006 review leaned to the family.⁸  Wood now looks "somewhere between the genus and the order", because "Interspecific hybridization is extremely common, even between members of different genera".¹⁷  Lightner's mammal survey ended with "a better rule of thumb is that the level of the kind is anywhere from the level of the family to the level of the order", and called its own judgements "quite subjective".^{19 19}  Her bird survey said that equating each family with a kind "is overly simplistic."¹⁰  Older counts for the Ark used the genus.  The choice matters:  the Mammal Diversity Database lists 1,338 living mammal genera in 162 families, against Lightner's 137 living mammal kinds.⁴⁰  The Ark logistics article follows the consequences.

None of this is a scandal.  A research programme should revise its estimates.  But it means that "the kind" is not yet a measured unit.  It is a working hypothesis whose rank depends on the group and on the method.

Part 5: What would break a proposed boundary, and what would not

A boundary that nothing could break is not a scientific claim.  The young-earth literature is better on this point than its critics usually allow.  Wood wrote in 2006:  "If future work showed discontinuity at the level of genus or even species, I would accept that result also. On the other hand, if I found continuity between different mammalian orders, I would accept that result".⁸  Scherer wrote that further work "could also demonstrate that the basic type criterion submitted here will not hold up".¹²  Lightner's team named a test that would make them rethink their own hybrid rule:  "If a blastocyst could be formed between domestic cattle and a skunk, this criterion would no doubt be reconsidered."¹³

Evidence that would break a boundary

1. A well-documented hybrid across it. On the creationists' own rule, a hybrid embryo that develops past the early stages joins the two parent groups into one kind. This has already happened to proposed boundaries: the grass hybrids moved the kind from a genus to most of a family, and the landfowl hybrids joined three of the four groups that statistics had separated. The test cuts one way, though. A hybrid can only join groups. It cannot show that a boundary is real.
2. Continuity that survives better sampling. If a statistical gap closes when more species, more traits or less missing data are added, it was a sampling gap. Senter proposed using the method to trace "the progressive filling of morphological gaps in the fossil record through successive years of discovery."³⁷ McLain and colleagues accepted that "Senter did demonstrate that morphological gaps in the fossil record tend to be filled in over time".³⁰ The reverse would favour the creationist: a gap that stays put as new fossils are described, across many matrices and methods.
3. Fossils from the stem. The simulation in Part 3 shows that a long unsampled stem is enough to produce a holobaramin verdict. So the decisive evidence for a boundary is the stem itself. Fossils that fill it, with a mix of traits from both sides, break the boundary. A stem that stays empty while nearby lineages fill in supports it, though it can never prove it.
4. Agreement or disagreement between independent tests. A boundary that hybrids, anatomy, genomes and fossils all put in the same place is strong. A boundary that moves with the test, as in the landfowl, the pigs and the feathered dinosaurs, is not yet a finding.
5. Shared genetic markers across the boundary. If the same rare insertion or broken gene sits at the same spot in two proposed kinds, and in a nested pattern, that is evidence of shared ancestry. Creationists answer that a designer may reuse parts. That debate belongs to the articles on human and chimpanzee retroviruses and pseudogenes, convergence and design reuse and phylogenetic congruence, and is not repeated here.

Evidence that would not break a boundary

1. Change within a kind. Darwin's finches, dog breeds, sticklebacks and new plant species are change within groups that creationists already call one kind. Each is evidence that species change. None is evidence that a kind boundary was crossed, because no creationist model put a boundary there. The Observed, or only inferred? article sorts what has been watched from what has been reconstructed.
2. A failed cross. Creationists themselves say it proves nothing.
3. One statistical result. A single matrix can show continuity or a gap, depending on its traits and species.

What has not been shown

On the young-earth side, three things are missing.

1. No method yet tells separate creation from extinction. A gap in today's data is the expected result of both. Baraminology's own founders define discontinuity as something that "must be demonstrated", and the methods that demonstrate it also find it in single-tree data with long stems.^{4 18}
2. The reviewed baraminology sources do not demonstrate a mechanism that fixes an absolute limit. The claim that "Extra- and interbaraminic evolution is considered impossible" is a statement of the model, not a measurement.³ Wood says plainly that a "macroevolutionary theory could be devised to account for the discontinuity between holobaramins".⁸ The young-earth case against such a theory rests on other arguments, about new genes and the rates of change, and those are made in the new genes and rapid speciation articles.
3. No stable rank. The kind has been placed at the genus, the family, the superfamily and the order, depending on the group and the method.

On the mainstream side, two things are missing too.

1. The observed examples do not follow a lineage across an established kind boundary. The finches, dogs, sticklebacks and plant species discussed above belong to groups creationists already join. Common ancestry across proposed family or order boundaries is chiefly a historical inference from comparisons and fossils. That does not make it an inference of unlimited change, or put every historical explanation on equal evidential footing. Its strength depends on the independent evidence for each proposed connection.
2. The clustering still needs explanation. Groups such as cats, bears, grasses and ducks stand apart from their neighbours today, and Wood's survey found that most statistical clusters are also branches of the evolutionary tree. Extinction and incomplete sampling can leave such gaps, as the simulation illustrates. Where a proposed connecting lineage has not been found, that specific connection remains an inference to test rather than a recovered fossil sequence.

Objections and responses

1. "Creationists say species never change, and speciation has been observed, so creationism is refuted." This attacks a view baraminology does not hold. Its founders accept speciation within kinds, and the young-earth models now need faster speciation than mainstream biology does.^{3 4} Observed speciation is evidence against fixity of species, which almost no one defends.
2. "Kinds are defined so that nothing can count against them." Partly fair. The cognitum and the family default are judgements, and Lightner calls them "quite subjective".¹⁹ But creationists have also stated tests that could go against them, and some have: hybrids moved the grass and landfowl kinds, and Wood said he would accept continuity between mammal orders if he found it.⁸ The weakness is that the tests can only join groups, never prove an edge.
3. "Hybrids between genera and families show the kinds are just arbitrary lines." Hybrids between genera and families are what baraminology predicts, since its kinds are larger than genera. A documented hybrid between two groups that creationists had already assigned to separate kinds would be a problem. The nearest case is the landfowl, where hybrid records crossed boundaries that a statistical study had drawn, and creationists dropped the boundaries.¹⁰
4. "The statistical clusters are just the evolutionary tree, so baraminology finds nothing new." Wood's own count agrees that most clusters match branches of the tree.¹⁷ That is consistent with both views: creationists expect each kind to be a natural group, and a natural group is a branch. The real question is whether the branches below the cluster join the branches outside it through an ancestor. The cluster cannot answer that.
5. "Reeves showed baraminology is junk." Reeves showed that BDC lacks a sound statistical basis. He also found that standard cluster analysis supported Wood's turtle result.³¹ The clusters are mostly real features of the data. What the methods cannot do is tell a created gap from an extinct stem.
6. "Gaps in the data are gaps in the record, so they mean nothing." Not quite. A gap that stays as fossils accumulate is evidence of something, and mainstream biology must explain it by extinction. The simulation in Part 3 shows that extinction can produce the pattern, not that it did in any given case. Each claimed boundary is a separate empirical question.
7. "Genesis says animals reproduce after their kind, so kinds cannot interbreed." Genesis says that God made animals "according to their kinds". It does not state a law of reproduction, as Lightner concedes and Wood insists.^{13 8} Even if it did, it would not settle where the edges are, since the word min names subgroups of whatever category it follows.⁶
8. "Mammals lose the ability to hybridize within a few million years, so hybrid data cannot test mammal kinds at all." Largely correct for the edge of a kind. Mammal hybrids mostly join species within a genus.¹⁵ That leaves the family-sized mammal kinds of Answers in Genesis resting on appearance. It does not refute those kinds; it means hybrids cannot confirm them.
9. "Observed adaptation shows there is no limit." It shows that change happens within groups. It does not test a limit at the level of the family or order, because no observed case reaches that level. Equally, the creationist cannot cite the absence of observed crossings as proof of a limit, since no one expects such a crossing to be observable in a human lifetime on the mainstream timescale.
10. "The kind keeps moving up the ranks to fit the evidence." The rank has moved, from the genus in older Ark counts to family and sometimes order now.^{19 17} Each move was driven by data, mostly hybrids. That is how estimates should change. But each move upward makes the kind larger, the speciation required after the Flood faster, and the claim of a limit harder to test. The rate problem is priced in the rapid speciation article.

Issues and unanswered questions

1. Hybrid records are second-hand. The key compilations, McCarthy's handbook of bird hybrids and Gray's lists of mammal and bird hybrids, were not obtained for this article. The landfowl claim rests on Lightner's report of McCarthy. Wood himself warned that some hybrids in Gray's books "are not accepted as valid".¹
2. The coyote and red fox litter was never genotyped. It rests on the zoo's identification of the parents from photographs.⁹ The dog and pampas fox hybrid is one animal.¹¹
3. The sampling design needs a direct test. The studies reviewed here rely heavily on matrices built by evolutionary systematists to resolve one group. A matrix designed from the start to sample neighbouring groups and the proposed stem, with fossils on both sides of a boundary, would strengthen the test.
4. The simulation is simple. It uses independent, unweighted traits, no missing data and one model of change. Real tables have correlated traits and gaps. A run on real matrices with fossils added in the order they were discovered would test Senter's claim that gaps fill over time.
5. The landfowl conflict is unresolved. Either the hybrid records are wrong or three statistical boundaries are. The McConnachie and Brophy study is a short abstract and was not obtained.
6. The giant panda and the bats are kept in or out of kinds by judgement, not by hybrid or statistical tests. So is much of the mammal list.
7. A demonstrated mechanism for a fixed kind boundary is missing from the reviewed sources. Without one, a historical claim of separate origin does not establish what future biological change could or could not do.

Verdict

The creationist kind is not the fixed species of popular caricature.  It is a group within which speciation is expected, defined in the literature by continuity inside and discontinuity outside, and tested by hybrids, appearance and statistics.  The Hebrew text does not define it, and its leading researchers say so.

The positive evidence is real.  Hybrids show that many species, across genera and sometimes families, share working developmental programs, and mainstream biologists draw the same inference from them.  Several groups have support from more than one method, but agreement about membership must not be mistaken for proof of an outer boundary.

The negative evidence is weak.  A failed cross does not establish separate kinds, and creationists agree.  Statistical baraminology depends on the traits kept, the data missing and the species sampled, its statistics have been rejected by a creationist statistician, and its own best defence turned about a fifth of its earlier holobaramin verdicts into inconclusive ones.  Where hybrids and statistics are both available, they have sometimes disagreed.  The simulation shows that common ancestry with a long unsampled stem can produce a holobaramin-like pattern under this article's stated rule.

The best case for creationists is that many groups form natural clusters larger than genera, as their model predicts.  The strongest objection is that positive evidence for membership does not establish separate origins, and the proposed outer boundaries depend on the data and method.  The reviewed studies do not measure an absolute limit to evolutionary change.  Those boundaries remain working hypotheses to test with fossils from the gaps, genetic markers shared across them, and explicit models of the histories being compared.

Sources