How Complete Is the Fossil Record?

"I look at the natural geological record, as a history of the world imperfectly kept, and written in a changing dialect; of this history we possess the last volume alone, relating only to two or three countries." — Charles Darwin, 1859¹

"As a record of everything that has ever lived on earth, the fossil record is an imperfect and incomplete data set. We know this. However, all data sets are incomplete." — Steven Holland, 2017, the paleontologist a young-earth article cites for the claim that the record is nearly complete²

This article asks how much of past life the fossil record keeps.  A sister article, How Complete Is the Geologic Column?, asks how much rock and time is present at a place.  The two are linked, because an animal can only be preserved where sediment was laid down and kept, but they are measured differently and fail differently.  A third, Do Clock Dates Agree with the First Fossils?, takes up the gaps between molecular-clock dates and first fossils.

Summary

1. "Complete" has at least four meanings, and the debate mixes them.  What share of all species that ever lived were fossilised and described;  what share of living species or families also have fossils;  how often a known lineage is found in the stages it lived through;  and how well the order of fossils in the rocks matches the order of branching in a family tree.  The numbers for these run from under 1% to over 90%, and each is right about its own question.

2. The share of all past species that are known as fossils is small on every estimate.  Even Michael Denton, the most quoted design writer on this point, relays an estimate that only about 2% of hard-shelled marine species that ever lived are known.³  The paleontologist he relays, Wyatt Durham, called that figure "unrealistically conservative" and thought one in a hundred "closer to reality."⁴  On a young-earth timescale the number of species that ever lived is smaller, so the share would be larger, but nobody on that side has published the calculation.

3. Per-stage completeness was computed for this article from 1.9 million fossil occurrences.  A genus of brachiopods, trilobites, cephalopods, conodonts, foraminifera or bivalves that is known before and after a stage is found inside that stage 76% to 83% of the time.  For insects the figure is 35%, for spiders and their kin 29%.  Land and freshwater collections as a whole score 56%;  marine collections 76%.  The record is not worse, by this measure, in older rocks.⁵

4. Where completeness is high, absences mean something.  A brachiopod genus as findable as the average known one, living through three stages, would be missed in all three about one time in two hundred.  An insect genus would be missed about one time in four.⁶  So a gap among shelly marine animals is good evidence that a lineage was not there, in a place and habitat that was sampled.  A gap among insects or spiders is weak evidence of anything, and land vertebrates and land plants fall in between.

5. The young-earth and design writers argue that the record is complete enough for its gaps to be real.  Denton counted 79.1% of living land-vertebrate families "found as fossils."³  Kurt Wise wrote that "a very large percentage of modern species have a fossil record."⁷  Michael Oard wrote that "the fossil record is essentially complete."⁸  Duane Gish argued from sheer numbers:  billions of fossils, so some intermediates should have turned up.⁹

6. Their headline numbers overstate, and one rests on a misreading.  Re-run against today's list of 657 living land-vertebrate families, 66% have any fossil record and 52% have one older than the Pleistocene;  of bird families, 34%.¹⁰  Of 4,715 living mammal species in the threatened and lower-concern categories of the Red List, 845, or 18%, appear in any of three fossil databases.¹¹  Holland, Oard's only mainstream source, does not say the record is nearly complete;  he says it is "an imperfect and incomplete data set."²  Stephen Meyer's quotation of a key 2000 paper drops its condition and adds "[and above]."¹²

7. The young-earth argument is right where it is narrow.  For durable, abundant marine animals at genus and family rank, completeness is high enough that missing intermediates there should be counted as evidence.  It fails as a general claim about the fossil record, and it fails worst for the transitions most often argued about, which run through land vertebrates, flying animals and soft-bodied ancestors.

8. Both models make predictions about where gaps fall.  Evolution with an incomplete record predicts that gaps cluster where preservation is poor, that fossil order matches the branching order of family trees better than chance, and that new finds mostly push first appearances earlier.  The first two are observed, and creationist writers confirm the third.¹³  A model of separately created kinds predicts gaps between kinds even in the best-preserved groups.  That is also observed, at class and phylum rank, mostly at the base of the Cambrian, where the ancestors of shelly groups would have been soft-bodied.  Neither model is finished.  The evolutionary answer for the best-preserved groups is that their ancestors did not have the hard parts that make the group well preserved.  That answer is reasonable, and it is also the one place where the theory relies on the record being poor exactly where its most important evidence should be.

Verdict:  the fossil record is complete enough to make some gaps real and too incomplete to make most of them mean anything.  Which kind a gap is can be measured, group by group, and the measurement belongs in every argument that leans on a gap.

Part 1: Darwin's argument, and four meanings of "complete"

Darwin raised the problem himself, and called it "the most obvious and gravest objection which can be urged against my theory."  If natural selection worked by small steps and the losers were driven out, the number of intermediate forms "which have formerly existed on the earth, be truly enormous.  Why then is not every geological formation and every stratum full of such intermediate links?"  His answer, on the same page:  "The explanation lies, as I believe, in the extreme imperfection of the geological record."¹

He gave three reasons.  Collections were poor:  "Only a small portion of the surface of the earth has been geologically explored, and no part with sufficient care."  Soft animals were not kept:  "No organism wholly soft can be preserved."  And sediment was laid down only here and there, so that shells and bones left on a sea floor where no sediment was settling would "decay and disappear."¹  He knew his readers would balk.  "That the geological record is imperfect all will admit;  but that it is imperfect to the degree which I require, few will be inclined to admit."¹

The third reason belongs to the rock record, and the geologic column article takes it up.  This article takes up the other two:  how much of life gets into the rocks, and how much of what got in has been found.

Darwin's question has since split into four, and the debate goes wrong whenever they are mixed:

The first question is about the whole of past life.  The second and third are about life that left at least some record.  The fourth is about whether the record that exists is a fair sample.  High numbers on the second, third or fourth do not answer the first, and a low number on the first does not refute the others.  Nearly every quotation in this debate, from either side, is an answer to one of these questions offered as an answer to another.

Part 2: How much of past life is preserved?

The share of all species ever: small on every estimate, and never measured

No study has measured what fraction of all species that ever lived is known as fossils.  It cannot be measured, because the denominator is the thing the record fails to show.  Every published figure divides a count of described fossil species by an estimate of all species, and the estimate does all the work.

Curt Teichert, 1956.  Teichert began by reporting the two earlier guesses at the number of all species, of every kind, that had ever lived:  George Gaylord Simpson's, "between 50,000,000 and 4,000,000,000 giving 500,000,000 as a possible approximation," and André Cailleux's, "between 17,000,000 and 860,000,000."  He set these aside, because they include bacteria, worms and other groups "which leave few or no preservable traces."¹⁴  He counted instead the living invertebrates "whose members have hard covers or skeletons, and which are therefore 'fossilizable,'" about 170,000 described species, rounded up to 250,000 for those not yet described.  Then he asked how many such species had come and gone since the Cambrian if the fauna were replaced every 12 million years, a figure he took from the mainstream history of marine molluscs.  He called his first answer "almost certainly too large" and settled on "somewhere in the vicinity of 7 or 8 million" fossilizable invertebrate species "waiting to be collected and described," with "about 10 million" as a ceiling for all fossil species.  His own warning about the inputs:  "The numbers of fossilizable species are little more than guesses, but they should indicate the order of magnitude."¹⁴  He left out insects on purpose and gave no number for plants.

Norman Newell, 1959.  In a Darwin centennial address Newell set Teichert's ceiling against about 92,000 described fossil animal species, added a guess for plants, and got a total "around one hundred thousand or only one per cent of Teichert's estimate of the ultimate richness of the fossil record."  His conclusion:  "the fossil record still is poorly known."¹⁵  Newell meant poorly described, not poorly preserved:  his denominator is Teichert's count of species that could have been fossilised, not of all species.

Wyatt Durham, 1967, in a presidential address to the Paleontological Society, estimated that "we know only about one of every 44 (and probably fewer) taxa of those that have existed in this environment since the beginning of the Cambrian," meaning marine invertebrates with hard parts.⁴  His own judgment went further:  "I think this ratio is unrealistically conservative, probably one out of every 100 is closer to reality."⁴  This is the figure Michael Denton relays as "as many as two per cent."³  Denton gives the upper end of Durham's range and not Durham's preferred figure, and dates the paper 1962;  volume 41 of the Journal of Paleontology is 1967.

James Valentine, 1970, built a diversity curve for nine phyla of shelled sea-floor animals and estimated that "well over 100,000 fossil marine species are now described," which "represents at least a fifteenth and possibly as much as a third of all the marine fossil species we are ever likely to recover."¹⁶  His denominator is species that left recoverable remains, not all species.

Three things travel with every number of this kind.

• The denominator is a model output. Each one multiplies a living count by a rate of turnover drawn from the mainstream timescale. On a young-earth timescale there is far less turnover, so the same described species would be a much larger share of all species. No young-earth writer has published that denominator, so the share on their model is unknown rather than high.
• Every estimate restricts itself to skeletons. Teichert's, Durham's and Newell's denominators count only species with hard parts. Soft-bodied groups, which are most species in many marine settings, are outside the count.¹⁷
• The numerator has grown. Described fossil species have multiplied since 1959, so Newell's 1% was a measure of the state of collecting, which is the point he was making.

The share of living species and families with a fossil record

The second question is easier, because the denominator is real:  living species can be counted.  The answers depend on the group and the rank.

Durable shells at family rank are nearly all there.  Land vertebrates at species rank mostly are not.  The bivalve families and the Californian molluscs are the record at its best:  thick-shelled, abundant animals living where sediment collects.  The missing Californian species are "chiefly rare today, and/or minute, fragile, and/or from deeper shelf habitats," and James Valentine concluded that "at least 85% of durably skeletonized living species may have been captured in the record."¹⁸  Note his own conditions:  "durably skeletonized," and "may."

These figures are biased upward in one direction that matters here.  A family alive today has, by definition, lasted to the present.  Most have long histories, wide ranges and many species, so they had many chances to be fossilised, and their youngest fossils sit in the youngest and least eroded deposits, where they are easiest to find.  Paleontologists call the second effect the pull of the Recent.  An extinct lineage that lived briefly, in a small area, in a habitat that was being eroded rather than buried, had far fewer chances.  So the share of living families with fossils is an upper bound on the share of extinct lineages with fossils, not an estimate of it.  Michael Foote and colleagues state the general form of this restriction for their own best-measured case, Ordovician and Silurian graptolites, of which about 75% of species are sampled:  that figure means "75% of those species that had the potential to be represented in light of where sediment was accumulating in the past, has survived to the present day and has been studied."²²

Two numbers bracket the range for mammals.  North American Cenozoic mammal species come out at 88% sampled on Alroy's composite, a well-studied record in a basin-rich continent.  Global Miocene mammal species come out at about 4% when Indrė Žliobaitė and Mikael Fortelius divide named species by the number their model says must have lived.²²²³  Both are honest.  The first asks how much of a well-sampled record is known;  the second asks how much of all mammal life is known, with a modelled denominator and a mainstream timescale built into it.  They add that "only 16% of Earth's terrestrial land surface is within sedimentary basins."²³  The other 84% of land is upland that is being eroded, not buried, and most land animals live there.

Part 3: How completeness is measured, and what each measure can see

Every working measure of completeness compares the record with itself.  None can see a lineage that left nothing at all.  That is their shared limit, and the reason the first question in Part 1 has no measured answer.  Christopher Paul, the paleontologist who did much of the early work on the question, put the rule in one line:  "Completeness and adequacy can only be evaluated against some predetermined aim."²⁴  He also drew the distinction this article relies on:  adequacy and completeness "are not the same," and "a 10% sample of past life should be adequate for many biological and geological purposes."²⁴

Paul's gap fraction.  The oldest and simplest method sums the gaps inside known ranges and divides by the total length of those ranges.  Paul applied it to cystoids, an extinct group of echinoderms, "and found 23% gap at family and series/stage level.  However, the precise value derived depends on both the taxonomic and stratigraphic levels chosen."²⁴

Per-interval sampling: how often a known lineage turns up

The FreqRat.  Michael Foote and David Raup's method of 1996²⁵ uses only the lengths of known ranges.  If a group's genera are found in most of the intervals they lived through, few will be known from just one interval;  if they are found rarely, many will.  Foote and Jack Sepkoski applied it to fifteen marine animal classes in 1999:  "If f1, f2 and f3 are the frequencies of genera with stratigraphic ranges of one, two and three intervals of geologic time, then the ratio f2²/(f1 f3) estimates the per-interval preservation probability."²⁶  The values they plotted run from about 0.06 for polychaete worms to about 0.9 for brachiopods.²⁶  Their conclusion, "completeness is rather high for many animal groups," is fair for the groups they measured.  Their Methods state what was left out:  groups with too few fossil genera, so that the data "favour more diverse and better-preserved groups (nemertines, priapulids and many other soft-bodied phyla were excluded, for example)."²⁶

Three-timers and part-timers.  John Alroy's method looks at three successive stages.  A genus found in the stages before and after a given stage was alive during it.  If it is also found in the middle stage it is a "three-timer";  if not, a "part-timer."  The share of three-timers among the two estimates the chance that a living genus is sampled in a stage:  in Alroy's words, "The overall sampling probability Ps is just 3t/(3t + Pt)."²⁷  Foote and colleagues report the database-wide value as 78% per stage.²²  The run for this article gives 75%.⁵

Range-through completeness.  The simplest measure counts, for each stage, the genera actually found in it and divides by the genera whose known first and last finds bracket it.  Benton and others call it the simple completeness metric.  It has a built-in flaw:  the first and last stages of every range are sampled by definition, so short ranges score near 100%.  This article reports it two ways, with the end stages and without them, and the version without them is the honest one.

All three agree on one thing:  they measure how often a lineage already known is re-found, not whether unknown lineages existed.  A lineage found only once, or never, contributes nothing to the three-timer count.  Žliobaitė and Fortelius put the consequence plainly:  "if advanced statistical estimates come only from well-sampled assemblages, they would represent the completeness of those well-sampled assemblages and would not be representative of global completeness on average."²³

Lazarus taxa: holes that are proved, not estimated

A genus that disappears from the record and then comes back is called a Lazarus taxon.  Its absence in between is a proven sampling gap, because the lineage was alive on both sides.  No model or estimate is needed.  Emmanuel Fara's review sets the rule for reading one:  "Interpreting the Lazarus effect is an asymmetrical procedure because the biological alternative is only favoured when the stratigraphic one cannot be documented."²⁸  Gaps are poor sampling unless shown to be something else.  After the end-Permian and end-Triassic extinctions, Paul Wignall and Michael Benton found that the share of Lazarus taxa did not track the number of fossil sites, and concluded that it "is probably a reflection of the extreme rarity of organisms at this time."²⁹  Rarity is still a sampling explanation, but of a different kind:  too few animals to leave a trace, not too few rocks.  Of the 20,015 genera in the database whose known range spans three or more stages, 79.7% have at least one stage inside their range where they have not been found.  Half of those holes are two stages long or less;  one in ten is seven stages or more.⁵  An earlier run on this site, counting gaps in years on the mainstream timescale, gave 75.9% of 16,804 genera.³⁰

Michael Oard gives a striking case from the creationist side:  a sponge genus known from the Middle Cambrian of Australia, then from nothing at all, then from the Upper Triassic of Vancouver Island.¹³  Long Lazarus gaps are rarer than short ones, and the longest are often database errors or trace-fossil names, so any single long gap needs checking before it is used.  The median is robust.

Lazarus gaps cut both ways.  They show that absence from the catalogued record is weak evidence of absence from the earth, which is the evolutionary point.  They also show that the order of the record survives its holes:  every one of these genera reappears in its expected relative place, which is the point made in The Order of the Fossil Record.

First and last appearances: why the ends of a range are always wrong

A lineage's first fossil is almost never its first member, and its last fossil is almost never its last.  If a genus is found in only some of the stages it lived through, the chance that the very first and very last stages are among them is less than one.  So recorded ranges are too short at both ends, and the shortfall is larger where sampling is poorer.

At the end of a range this is the Signor-Lipps effect, named for Philip Signor and Jere Lipps's paper of 1982, which called it "artificial range truncation" and showed that if last finds fall at random before a sudden extinction, "apparent extinctions will begin well before a mass extinction and will gradually increase in frequency until the mass extinction event, thus giving the appearance of a gradual extinction."³¹  Steven Stanley defines it as "the incorrect assignment of extinctions that occurred during a crisis to an interval preceding the crisis because of the incompleteness of the fossil record."³²  He estimates that for the end-Cretaceous extinction, genera whose last recorded find falls in the early Maastrichtian make up 10-12% of the total mass-extinction number.³²  At the start of a range the same logic means first appearances are too late.  Charles Marshall turned this into confidence intervals on the true ends of a range in 1990, and named his own weak point:  the method's "most important assumptions are that fossil horizons are distributed randomly and that collecting intensity has been uniform over the stratigraphic range."³³  This article owns the organism side of that effect.  The rock side, the shortening of the record itself by gaps in deposition, is in the geologic column article.

This gives the cleanest test in the subject.  If ranges are truncated by poor sampling, more collecting should mostly push first appearances earlier and last appearances later, and rarely the other way.  The creationist literature reports exactly this.  Michael Oard, in 2000:  "The more the sedimentary rocks of the earth are examined, the more the fossil ranges are expanded — especially downward."¹³  In 2017 he called the reported cases "probably the tip of the iceberg" and the problem "almost certainly more severe than reported in any literature, secular or creationist."³⁴  John Woodmorappe, in his long 1983 treatise on Flood stratigraphy, made the same observation:  "Even when taxa are accepted as valid there is a noteworthy trend for stratigraphic ranges to increase with further collecting," so that "there is thus no way of knowing whether the stratigraphic ranges of taxa shown in Figure 1 are reasonably final."³⁵  Oard offers range expansion as a problem for the geological column.  It is first a measurement of the record's incompleteness, made by young-earth writers, and it runs in the direction sampling theory predicts.

Congruence: does the order of fossils match the order of family trees?

The fourth kind of measure does not ask how much is missing.  It asks whether what is present is in the right order.  A family tree drawn from anatomy alone, with no use of ages, predicts which branches should appear first.  The rocks give an independent order.  If the record were a jumble, the two would agree no better than chance.

Michael Benton and R. Hitchin, 1997, scored 384 published trees:  63 echinoderm, 147 fish and 174 tetrapod.  On the stratigraphic consistency index, 78% of trees had at least half their nodes in the right order;  on the relative completeness index, which compares known ranges with ranges the tree implies but the rocks do not show, 75% scored above 50%.  But a direct rank correlation between branching order and age was significant in only 39% of trees, and they called that "disappointing."³⁶  Their summary:  "the fossil record, incomplete as it is, gives a reasonably faithful documentation of the sequence of occurrence of organisms through time."³⁶

Benton, Matthew Wills and Hitchin, 2000, extended the sample to 1,000 trees and found "no evidence of diminution of quality backwards in time."  Their abstract adds a sentence that is often left out:  "Ancient rocks clearly preserve less information, on average, than more recent rocks.  However, if scaled to the stratigraphic level of the stage and the taxonomic level of the family, the past 540 million years of the fossil record provide uniformly good documentation of the life of the past."³⁷

Benton, 2001, sorted the same 1,000 trees by the year they were published, to see whether better trees fit the rocks better.  They did not, on balance:  "the order of branching has improved slightly, but the disparity between estimated times of origination from phylogeny and stratigraphy has, if anything, become worse."³⁸

A. O'Connor and Matthew Wills, 2016, tested the indices themselves on 647 trees and found that they move with what is in the sample and with the shape of the trees.  Tetrapod and echinoderm trees score high, arthropod trees low, and a shift in the mix of groups can change an era's average.  Of the five indices, "the modified GER was the index least susceptible to bias."³⁹

What congruence shows, and what it does not.  It shows that the order of first appearances is not random with respect to anatomy:  forms that anatomy places low on the tree tend to appear lower in the rocks.  That is a fact both models must explain, and it is a fact about relative order, which this site treats as open to both sides.  It does not measure how much is missing.  Benton's own sentence says older rocks hold less, and his "uniformly good" applies at stage and family resolution, not to species, and not to everything above family.

Part 4: How completeness varies by group, by setting and by depth

By group: computed for this article

The table below was computed for this article from 1,902,080 Paleobiology Database occurrences, with footprint names and form taxa left out.  An occurrence is placed in a stage only if its whole age bracket lies inside that stage;  1,135,829 qualify.  Everything is counted at genus rank, in stages, the finest standard divisions of the column.  No duration in years enters.⁵

The ladder is the one taphonomy predicts.  Thick-shelled, abundant sea animals sit at the top:  brachiopods, cephalopods, conodonts, foraminifera, bivalves.  Animals that lived on land or in the air, or had thin cuticles, sit at the bottom:  insects, spiders, crabs.  Vertebrates sit in the middle.  The same order appears in the live-dead surveys:  about 95% of shelled molluscs get into a time-averaged death assemblage, about 75% of echinoids and land mammals, about 50% of marine decapods.¹⁷  In one bay off North Carolina molluscs were 35.0% of living animals and 98.2% of dead ones;  arthropods were 33.5% of living animals and 0.11% of dead ones.⁴⁰

Two cautions about the table.  First, the three-timer figure is conditional on the genus already being known from both neighbouring stages, so it measures how reliably a findable genus is re-found, not the chance that an unknown genus would be found at all.  Second, the two measures in the middle columns do not always agree:  ferns, horsetails and lycopods score 76% as three-timers but 32% on interior stages, because a plant genus found in neighbouring stages tends to be a common one.  Where the measures disagree, the lower one is the safer guide to an absence.

By setting: sea against land

A genus living on land is about six times as likely to slip through three stages unrecorded as one living in the sea.  The reason is where sediment is kept.  Seas and lowland basins accumulate sediment;  uplands erode.  Only about 16% of today's land surface lies in sedimentary basins, by the estimate Žliobaitė and Fortelius use.²³  Animals of the remaining 84%, including most mammals, birds, insects and flowering plants, enter the record only when they die in or are carried into a basin.

By depth in the column: no decline in older rocks

On these measures the record is not worse in older rocks.  That agrees with Benton and colleagues' finding from 1,000 family trees.³⁷  It does not mean old rocks keep as much as young ones.  Most Paleozoic genera in the database are durable marine animals, so the Paleozoic row is weighted toward the groups at the top of the ladder.  The column is also cut unevenly:  48 Paleozoic stages, 30 Mesozoic and 23 Cenozoic, and a finer division is harder to fill.  What the table does show is that a lineage already known from older rocks is re-found in its stages about as often as one in younger rocks.

Exceptional deposits: what the ordinary record misses

Lagerstätten, the rare beds that keep soft tissue, are the one direct look at what ordinary beds leave out.  The answer is most of the animals.  About 80% of the Burgess Shale fauna is soft-bodied, and about 16% cannot be placed in any known phylum.  Simon Conway Morris drew the lesson himself:  "The low diversity of normal Cambrian faunas… may be due partly to the scarcity of groups possessing readily fossilizable hard parts."⁴¹  Anna Behrensmeyer and colleagues give the general figure:  soft-bodied taxa "have negligible preservation potential under ordinary environmental conditions…  In marine sands and muds, for example, such taxa constitute 30–100% of species."¹⁷

And the windows themselves are unevenly spread.  A. D. Muscente and colleagues compiled 694 exceptional assemblages and found that "conditions conducive to exceptional preservation declined worldwide during the early Paleozoic."⁴²  So the soft-bodied record is not a fair sample of time either:  it is richest just where the shelly groups first appear.

So the completeness figures in the tables above are completeness figures for animals with hard parts.  They say nothing about worms, jellyfish, most arthropod larvae, or the soft-bodied ancestors that evolution proposes for the skeletonised groups.  How each Lagerstätte formed, and whether rapid burial explains it, is the subject of Fossil Preservation.

Part 5: The young-earth and design case, in its own words

Michael Denton, 1986

Denton, a biochemist and the most widely read design critic of evolution in the 1980s, set out three explanations for the gaps:  "Firstly, insufficient search, ie that not all fossil bearing strata have been examined.  Secondly, the imperfection of the record, ie that only a fraction of the species that lived in the past have left fossil remains.  Thirdly, saltational evolution, ie that the gaps are real and that evolution occurred in a series of jumps."³

Against the second he set a count.  "Of the 329 living families of terrestrial vertebrates 261 or 79.1% have been found as fossils and, when birds (which are poorly fossilized) are excluded, the percentage rises to 87.8%."³  His table is credited "from Romer," and his endnote says it was "compiled from information on pp 347-96" of Alfred Romer's Vertebrate Paleontology of 1966.³  Those pages are Romer's classification appendix, in which "all families of vertebrates are listed (with their present distribution) whether they include fossil forms or not," with Romer's own warning that the list "undoubtedly contains inaccuracies and omissions."⁴³  So Denton had the right denominator for his question;  the question is what his answer shows.  He added G. G. Simpson's estimate that "surely half and probably two-thirds or more" of Middle Oligocene mammal genera in one region are known,³ and Durham's 2% for hard-shelled marine species.³

Denton was fair to the other side.  On the same page he quoted Thomas Schopf on the multituberculates, a mammal order found in stages covering only 54% of its known span:  "46 per cent of the time the Multituberculata existed, there has not yet been discovered a record of the order anywhere in the world."³  And his decisive argument was not a completeness number.  It was the pattern:  "If the gaps really were the result of an insufficient search, or the result of the imperfection of the record, then we should expect to find more transitional forms between mouse and whale than between dog and cat."³  Whether gaps grow with rank is the subject of Trends in the Fossil Record, Part 2.  This article takes up only his completeness evidence.

Duane Gish, 1995

Gish, of the Institute for Creation Research, stated two models' predictions.  For creation:  "we would predict that all of the major types of life, that is, the basic plant and animal forms, would appear abruptly in the fossil record without evidence of transitional forms linking one basic kind to another."⁹  He did not claim the record is nearly complete.  He conceded that "only a minute fraction of these animals have been recovered as fossils" and argued from numbers instead:  so many fishes and amphibians are known that some of the "many hundreds of millions, even billions" of transitional animals should have turned up, and "the discovery of only five or six of the transitional forms scattered through time would be sufficient to document evolution."⁹

Carl Werner, 2008

Werner, a physician whose books and films on fossils are widely used in creationist teaching, makes Gish's argument from abundance with larger numbers:  "Overall, the fossil record is rich — 200 million fossils in museums — but the predicted evolutionary ancestors are missing, seemingly contradicting evolution.  For example:  Museums have collected the fossil remains of 100,000 individual dinosaurs, but have not found a single direct ancestor for any dinosaur species."⁴⁴  The book gives no source for the totals, and they are counts of specimens, not of kinds of animal, so they cannot be turned into a share.

Kurt Wise, 2002

Wise, a paleontologist trained at Harvard, made the completeness argument directly.  In 1995 he listed "high species preservability" among the first-order patterns of the fossil record that a creationist model should explain, without giving a number for it.⁴⁵  In his 2002 chapter on created kinds:  "If most species living in the present are known as fossils, and if species have persisted for millions to tens of millions of years, it is reasonable to expect that a fairly large percentage of the millions of transitional species that have existed in the past would have been preserved in the fossil record."  He pointed to the best-preserved animals:  "there seems to be a complete absence of transitional species among these animals—even though these are the best represented animals in the fossil record!"⁷  In his chapter on the Flood he turned it into a test between timescales:  on an old earth, "the approximately 250,000 fossil species documented to date would have to be an extremely low percentage of all the species that have ever lived," while a Flood "would probably result in a very large percentage of fossil species being preserved.  Our observation is that a very large percentage of modern species have a fossil record."⁷

Leonard Brand and Arthur Chadwick, 2018

Brand, of Loma Linda University, and Chadwick, of Southwestern Adventist University, set out what an "interventionist" model expects:  "convincing series of evolutionary connecting links are not expected in the fossil record."  Their reading of the trend:  "As more fossils are found, it becomes clearer that the gaps between major groups of organisms are real, and sequences of intermediates are not likely to be found."⁴⁶  They are candid about the exceptions.  The vertebrate series, they write, "are one of the strong points in favor of macroevolution," and their model still has "to find an alternate explanation for the apparent intermediates, for the sequence of fossils in the rocks, for the increasing percentage of extinct groups lower in the fossil record."⁴⁶

Michael Oard, 2021

Oard, a retired meteorologist who writes on Flood geology, published the bluntest version in the Journal of Creation.  Under the heading "The fossil record is essentially complete" he wrote that Steven Holland "points out that the fossil record is imperfect in a sense, but really nearly complete."⁸  He drew a prediction from the Flood:  "we would expect the fossil record to be complete, except for the small number of new fossils being discovered every year."⁸  And he set the dilemma:  "the higher the fidelity of the record … is, the less evolutionists can appeal to the incompleteness of the fossil record to explain away the morphological gaps between fossil taxa."⁸

The design writers: Günter Bechly and Stephen Meyer

Bechly, a paleoentomologist, told a 2018 audience that Foote and Sepkoski's 1999 paper showed "we have reached the saturation point in many group of organisms," that "80% of vertebrate families are recorded in the fossil record," and therefore that an appeal to incompleteness "to explain away gaps or discontinuities" was "not available anymore."⁴⁷

Meyer, in Darwin's Doubt, wrote that Foote "has shown, using statistical sampling analysis, that as more and more fossil discoveries fall within existing higher taxonomic groups … it grows ever more improbable that the absence of intermediate forms reflects a sampling bias."¹²  In a note he quoted Benton, Wills and Hitchin:  "if scaled to the . . . taxonomic level of the family [and above], the past 540 million years of the fossil record provide uniformly good documentation of the life of the past."¹²

What the case claims, stated fairly

Put together, the case has four claims, and it is only fair to test each on its own terms.

1. A high share of living land vertebrates, or of living species generally, have fossils, so the record is good enough to have caught intermediates (Denton, Wise, Bechly).
2. The best-preserved animals, the shelly marine invertebrates, show no transitions between major groups, though they are the best test (Wise, Denton, Gish).
3. The more we search, the more the gaps hold, so the pattern is not an artefact of collecting (Brand, Meyer, Bechly, Oard).
4. A Flood predicts a nearly complete record with gaps between kinds, while deep time predicts a poor one (Wise, Oard).

Part 6: Testing the case

Claim 1: most living land vertebrates have fossils

Denton's count, re-run.  His 79.1% came from Romer's classification of 1966, in which 329 living families of land vertebrates were recognised.  Today's living-taxonomy list for the same groups holds 657, mostly because birds and amphibians have been split into more families.  Each was looked up in the Paleobiology Database for its oldest fossil.¹⁰

Six bird families have a first record that straddles the base of the Pleistocene and are counted under "any" only.

Denton was right that the family-level record of living mammals, turtles and crocodilians is good.  Three quarters of living mammal families have a fossil older than the Pleistocene.  His figure was too high for the whole set, and it shrinks further once the age of the oldest fossil is asked.  Eighty-eight of the 433 families with any record, one in five, are known only from the Pleistocene or Holocene:  lemurs, three-toed sloths, gibbons, kingfishers, swallows.  A family known only from the last ice age tells nothing about whether its ancestors were caught in the deep record.  Andrew Snelling, the young-earth geologist whose two-volume Earth's Catastrophic Past is that side's standard text, states the same pattern:  "the groups of animals and plants that live today are common as fossils only in Pleistocene and upper Tertiary strata."⁴⁸  Name matching between two taxonomies misses some families that are present under another name, so the "any record" column is a lower bound, but not by enough to restore 79%.

And the inference does not follow even where the count is high.  A living family has, by definition, survived to the present.  Most have long histories, many species and wide ranges, and their latest fossils lie in the youngest and least eroded rocks.  The lineages the gaps are about are extinct ones, often short-lived and local on the evolutionary model.  So the share of living families with fossils is a ceiling on the share of extinct lineages with fossils, not a measure of it.  This is the same restriction Foote states for his graptolites, and the same self-selection Žliobaitė and Fortelius describe.²²²³

Claim 1, second form: most living species have fossils

Wise's version is stronger:  "most species living in the present are known as fossils."⁷  It is true for one kind of animal in one kind of place.  Of 698 shelf molluscs living off California, 538, or 77%, are known as Pleistocene fossils from the same region.¹⁸  Worldwide, 76% of living marine bivalve genera and subgenera have a fossil record.¹⁹  It is false for land vertebrates.  Of 4,715 living mammal species in the Red List categories Plotnick and colleagues analysed, 845, or 18%, appear in any of three fossil databases;  of threatened bird species, 3.3%.¹¹  And the Californian figure is a count of living species in the youngest deposits, which is the easiest case the record offers.

Wise's "something on the order of 95 percent of the fossil record is of shallow water marine invertebrates" has no source in his book.⁷  The same figure appears in an article by Andrew Snelling of 1991, footnoted only to "an informal talk" Wise gave at the Institute for Creation Research in 1988, and this site does not use it as evidence.⁴⁹  What can be said is that marine invertebrates dominate the database:  in the dump used here, molluscs, brachiopods, echinoderms, corals and bryozoans together hold 43% of all occurrences, and foraminifera another 10%.⁵

Claim 2: the best-preserved animals show no transitions

This is the strongest claim in the case, and it is partly right.  For brachiopods, bivalves, cephalopods and the other groups at the top of the Part 4 table, a genus as findable as the average known one, living through three stages, would be missed in all three less than one time in a hundred.  If a common, long-lived genus intermediate between two families of brachiopods had lived in the habitats that were sampled, it would almost certainly have been found.  So where such genera are missing between durable marine families, their absence is real evidence that no common, long-lived intermediate lived in the sampled seas.

Three things limit what follows from that, and each has to be argued case by case rather than assumed.

1. The figure applies to findable genera. The three-timer rate is computed from genera already known from neighbouring stages. A transitional population that was rare, local, or lived for less than a stage is not covered by it. Michael Foote showed in 1996 that with plausible completeness, "on the order of 1%-10% or more of the known fossil species would be directly ancestral to other known fossil species" for marine invertebrates, but added that "if speciation is rapid relative to the duration of species … then the chance of capturing transitions, especially major transitions among classes and orders, is substantially diminished."⁵⁰ He proposed working out how many major transitions should be visible, and wrote that this "would have important implications for whether the small number of documented major transitions provides strong evidence against evolution."⁵⁰ The calculation has not been published.
2. Most of the big gaps among shelly animals sit at the base of the Cambrian and Ordovician, where the groups first acquire the shells that make them well preserved. On the evolutionary model their ancestors were soft-bodied, and the completeness of a shelly group says nothing about soft-bodied animals: about 80% of the Burgess Shale fauna had no hard parts.⁴¹ This is a legitimate answer. It is also an answer that places the decisive evidence exactly where the record is poorest, and a skeptic of evolution is entitled to say so. The Cambrian article weighs it.
3. The sea floors that were sampled are not all the sea floors that existed. Most of the shelf record comes from basins that are now on land, and the geologic column article measures how much of each interval survives.

Claim 3: the more we search, the more the gaps hold

For durable marine groups the evidence supports it.  Trilobite and brachiopod discovery has nearly stopped at family rank:  one new trilobite family was named in the 2010s against 55 in the 1950s, and 7 brachiopod families against 86 in the 1960s.⁵¹  Foote's study of four well-sampled marine groups found that the history of their range of form had changed little as new species were described, and concluded that "in many respects our view of the history of biological diversity is mature."⁵²  And over the twentieth century, as more fossils were described and family trees were redrawn, the gap between the ages trees imply and the ages the rocks show did not shrink.  Benton found that it had, "if anything, become worse."³⁸  New finds have not been filling the ghost lineages that trees require.  That result runs toward the creationist reading, and it should be stated.

For land animals the evidence is against it.  Half of all accepted non-avian dinosaur genera were named after 2006.⁵¹  Fossil insect families were named faster in the 2010s than in any earlier decade, and 28.9% of dinosaur families are younger than 1998.⁵¹  The search is not finished, so it cannot have shown that the gaps hold.  Brand's "as more fossils are found" is a fair description of trilobites and an unfair one of dinosaurs or insects.

Meyer stretches both of his sources.  Foote's 1997 paper studied the range of form inside four well-sampled classes of marine invertebrates, the groups where Part 4 finds completeness highest.⁵²  Meyer applies it to the gaps between phyla, including soft-bodied Cambrian ones, which Foote did not measure.¹²  And Meyer's quotation of Benton, Wills and Hitchin drops "the stratigraphic level of the stage and the" and adds "[and above]."  Their sentence is about families, at stage resolution, and it follows the sentence "Ancient rocks clearly preserve less information, on average, than more recent rocks."³⁷¹²

Bechly's 80% has no home in the paper he names.  Foote and Sepkoski studied fifteen marine classes, with not a single tetrapod family among them, and print no 80% figure.²⁶  The nearest published number is Denton's 79.1% for living land-vertebrate families, tested above.

Claim 4: a Flood predicts a nearly complete record

Oard misreads his only mainstream source.  Holland's address says the opposite of "nearly complete":  "As a record of everything that has ever lived on earth, the fossil record is an imperfect and incomplete data set.  We know this."  Holland also grants "that only a tiny fraction of organisms that have lived are fossilized.  This is true."²  His argument is that the record's gaps have a known structure, set by time-averaging, by the architecture of sedimentary sequences and by which basins formed, and that paleontologists should model that structure rather than apologise for it.²  That is a case for using the record, not a claim that it is complete.

Oard's own range-expansion papers cut against his 2021 claim.  If ranges keep moving "especially downward," and the problem is "almost certainly more severe than reported," then the record is still changing in exactly the way an incomplete record changes.¹³³⁴  Both cannot be used.

The Flood prediction can be tested, and it does not fit as stated.  "We would expect the fossil record to be complete, except for the small number of new fossils being discovered every year."⁸  From 2014 to 2023 about 41 new dinosaur genera were named each year, against about 24 a year a decade earlier.⁵¹  A model can explain that, for example by saying the Flood buried more than has yet been dug up.  But then it no longer predicts a nearly complete known record, and the claim reduces to the one the evolutionary side also makes:  much is buried that has not been found.

What would make the young-earth version testable is a denominator.  Wise is right that a short history with one great burial should preserve a larger share of species than a long one.  To turn that into a prediction needs an estimate of how many species lived before the Flood and how many after it.  No young-earth writer has published one, so the share of all past species that a Flood model predicts is unknown, not high.

Part 7: Where should the gaps fall? Two models, six tests

Both models expect an incomplete record.  Evolution needs one to explain why intermediates are rare.  A young-earth model needs one too, and Werner says so in the creationist "counterpoint" he writes for his own book:  "both theories, the theory of evolution and the theory of creation, have missing fossils."  Horses, humans, bats and goats have not been found in dinosaur layers, though on his model they lived then;  their presence is "implied by the large number of other modern animals and plants that have been found."⁴⁴  So the useful question is not whether the record is complete but where each model expects the gaps to fall, and which pattern the measurements show.  Each test below uses relative order only, which both sides accept.⁵³

Tests 2, 3 and 4 favour evolution with an incomplete record.  Congruence between trees and rocks is the strongest.  Wise's own account of the few intermediates found in the right order is that they "may be nothing more than morphological intermediates which ended up in stratigraphic intermediate position by one accident or another," and that "a few stratomorphic intermediates might be expected in even a random burial process."⁴⁵  But a family tree drawn from anatomy alone predicts the order of first appearances better than chance in most groups,³⁶ and the young-earth explanation for that order, burial by ecological zone, is weighed and found wanting in The Order of the Fossil Record.  Range extensions running mostly downward are exactly what truncated ranges predict, and the creationist literature reports them.¹³  The discovery pattern, saturated for shelly sea animals and still rising for dinosaurs and insects, is what an incomplete record searched unevenly looks like, and it contradicts Oard's prediction of few new finds.

Test 5 favours the young-earth reading at the highest ranks, and an honest account says so.  Among the animals whose record is best, the gaps between classes and phyla are real gaps in a well-sampled record.  The evolutionary answer is that the ancestors of shelly groups had no shells, so the shelly record's completeness does not reach them.  That answer is reasonable and partly testable.  Burrows need no hard parts to be preserved, and the Cambrian article finds that they too multiply sharply at the base of the Cambrian, from 7 trace genera to 42,⁵⁴ so a change in preservation cannot be the whole explanation.  The same article finds arthropod burrows about two stages below the first trilobite bodies, so some animals were there before their skeletons were.  Either way, it is an answer about the one part of the record that completeness cannot measure.

Test 6 is a point the young-earth side could make and has not.  If old rocks were steadily ground down over hundreds of millions of years, a naive expectation is that old lineages would be harder to re-find in their stages.  They are not:  the per-stage figures are as high in the Paleozoic as in the Cenozoic, and Benton's 1,000 trees fit the rocks as well in old rocks as in young.³⁷  The mainstream accommodates this:  Benton concedes older rocks hold less, and his claim is about agreement at stage and family resolution, not about how much survives.  A young-earth writer could say that a record laid down mostly in one year should show no decline, and that none is seen.  It does not decide between the models, but it belongs on the young-earth side of the ledger.

So the gaps fall mostly where an incomplete record puts them, with one large exception.  The exception is the origin of the major shelly groups, where the young-earth prediction fits the data plainly and the evolutionary explanation depends on animals that left little or nothing.

Objections and responses

1. "Every completeness measure assumes evolution and deep time."

Some do and some do not.  Estimates of the share of all species ever fossilised multiply by a rate of turnover drawn from the mainstream timescale, and this article says so wherever it prints one.  The measures computed here do not.  A Lazarus gap is a genus found below and above a bed and not in it.  The three-timer and range-through measures count stages, not years.  Kurt Wise and Donna Richardson, testing 114 boundaries in the same database, concluded that "the global biostratigraphic column seems to be real."⁵³  A genus missing from beds that lie between two beds holding it is a sampling gap on any timescale.

2. "If shelly marine genera are re-found in four stages out of five, the record is complete and the gaps are real."

For common, long-lived genera living in the habitats that were sampled, yes, and Part 6 says so.  The figure says nothing about rare or short-lived populations, nothing about lineages living where no sediment was kept, and nothing about animals without hard parts.  The three-timer measure is computed only from genera already known on both sides of a stage.

3. "Lazarus taxa show the geological column is unreliable."

They show the record is incomplete, not that it is out of order.  Every Lazarus genus counted here reappears where its earlier and later finds would place it.  Oard's sponge returns in a younger bed, not an older one.¹³  Whether the order itself holds is tested, against the young-earth argument, in The Order of the Fossil Record.

4. "Appealing to incompleteness is special pleading: evolutionists invoke it only where it is convenient."

Denton's form of this, that the appeal is a "tautology," has force against anyone who assumes the record is poor because the gaps exist.³  It has none against the measures here, which are made on lineages that are present, without reference to the gaps in question, and which come out high for shelly marine animals and low for insects regardless of what anyone wants.  Where the measures are high, this article concedes that absence counts.  Where the evolutionary answer rests on ancestors without hard parts, Part 7 says that it rests on the part of the record no measure can reach.

5. "A family alive today must have had ancestors, so the high share of living families with fossils shows the record catches lineages."

It shows the record catches long-lived, widespread lineages that survived to the present, often only in their latest deposits.  One family in five with a record is known only from the Pleistocene or Holocene.  The lineages the gaps are about are extinct, and on the evolutionary model often brief and local.  The figure is a ceiling for them, not an estimate.

6. The skeptic: "Your database is a literature compilation, heavy on North America and Europe. Its numbers are worthless."

They are biased, and the bias is measured:  on earlier work on this site, the United States holds 22% of collections and ten countries 57%.⁵⁵  But the bias falls on all groups at once, and the comparisons here are between groups, settings and eras, which is what live-dead surveys find survives a biased record:  "the fossil record is likely to be reliable with respect to relative comparisons of composition and diversity."⁴⁰  The absolute numbers should be quoted with the group, the rank and the measure attached.

7. The skeptic: "You concede too much. Missing intermediates among brachiopods prove nothing, because transitions happen in small, fast-changing populations."

That is a possible answer, and Foote states its effect:  if speciation is rapid relative to species durations, "the chance of capturing transitions, especially major transitions among classes and orders, is substantially diminished."⁵⁰  But it is a claim about each case, not a fact that can be assumed for all of them.  Foote proposed calculating how many major transitions an evolutionary history should leave visible.  Until that is done, "transitions are fast and small" is a hypothesis about the very places where the record is silent.

8. The defender of Oard: "Holland's point is that the record is good. 'Nearly complete' is a fair summary."

Holland's own sentences are "an imperfect and incomplete data set" and "only a tiny fraction of organisms that have lived are fossilized.  This is true."²²  "Not unusually poor" and "nearly complete" are different claims.  Holland makes the first.

9. The defender of Meyer: "Families sit inside higher taxa, so '[and above]' is fair."

Benton, Wills and Hitchin state their conclusion at "the taxonomic level of the family" and "the stratigraphic level of the stage."  A good fit at that level does not show a good fit for the origins of classes or phyla, which is where Meyer applies it.  Meyer's ellipsis also removes the stage condition, and his note does not give the sentence before it:  "Ancient rocks clearly preserve less information, on average, than more recent rocks."³⁷

10. The defender: "The Flood laid down most of the column in a year, so a per-stage sampling rate is meaningless on a young-earth model."

The measure does not depend on how long a stage lasted.  It asks whether a genus found in the beds below and above a set of beds is also found in them.  On a Flood model those beds were laid down weeks apart and the animals were living in the waters that laid them down.  A genus missing from the middle beds is a gap in what was buried or found, on either timescale.  If anything, a Flood model has more to explain:  on it the missing genus was alive and nearby while the middle beds formed.

Issues and unanswered questions

1. No young-earth denominator. No young-earth writer has published an estimate of the number of species before and after the Flood, so the share of all past species a Flood model predicts to be preserved is unknown. Wise's argument from 250,000 described species needs that number to become a test.
2. Foote's calculation has not been done. Nobody has estimated, from completeness, species durations and the number of class- and order-level transitions, how many major transitions an evolutionary history should have left visible. Both sides cite the question; neither has answered it.
3. The three-timer rate is conditional. No measure sees a lineage that left nothing, and the high per-stage rates apply to genera already known. How far they apply to rare or short-lived populations is not measured.
4. A direct test of where family-level gaps fall was not run. If gaps between families reflect sampling, the stratigraphic gap between sister families should be smaller in well-preserved groups than in poorly preserved ones. If they reflect separately created kinds, preservation should not matter. The data to run it are in the dump used here, but a "sister" in the database is a classification relationship, not a phylogenetic one, and the result would need care.
5. Denton's denominator was not rebuilt from Romer. This article re-ran his question against a modern list of living families; it did not reconstruct Romer's 1966 count of 329. Name-matching between the two taxonomies makes the modern "any record" figures lower bounds.
6. Sources not fully read. Benton, Wills and Hitchin 2000, Signor and Lipps 1982, Marshall 1990, Fara 2001 and Wignall and Benton 1999 were read from their abstracts only; Benton and Storrs 1994 and Paul 1982 were not obtained. Who coined "Lazarus taxa" is unsettled in the sources read.
7. Old rocks and the young-earth case. The lack of any decline in per-stage sampling or in tree-rock fit with depth is a point a Flood model could use. Nobody on that side has used it or tested it.

Verdict

The fossil record is complete enough to make some gaps real, and too incomplete to make most of them mean anything.

The share of all species that ever lived is unknown and small on every published estimate.  The useful measures are relative, and they put a number on each group.  For brachiopods, bivalves, cephalopods, conodonts and foraminifera, a known genus is re-found in about four of every five stages it spans, and a common, long-lived genus living in the sampled seas would almost never be missed.  For insects and spiders it is one stage in three;  for land collections as a whole, about one in two.

The young-earth and design case is right where it is narrow.  Gaps among durable marine animals are evidence, and the class- and phylum-level gaps at the origin of shelly groups are the one place where the young-earth prediction fits plainly and the evolutionary answer relies on soft-bodied ancestors that left little.  It is wrong as a general claim.  Denton's 79% falls to 66% against today's list of living families and to 52% before the Pleistocene;  Wise's "most living species" holds for Californian shells and fails for mammals;  Bechly's 80% is not in the paper he names;  Oard's "nearly complete" is not what Holland wrote;  and Meyer's Benton quotation is altered.

On the tests that discriminate, the gaps fall mostly where an incomplete record puts them:  fossil order fits family trees better than chance, ranges extend downward as search continues, and well-preserved groups have stopped yielding new families while land and soft-bodied groups have not.  Any argument that leans on a gap should name the group, the rank and the measured completeness, and should be weighed by that number.

Sources