Extended Evolutionary Synthesis

Speculative If developmental bias can be measured in a population before selection acts, and used to predict which way that population will move — which has been done retrospectively many times and prospectively never, and section 7 says what would count — then evolutionary biology acquires a forecast where it currently has an explanation issued after the fact. The condition is the whole claim, and putting it second would misrepresent the field.

Established The magnitude is best stated in the units the one funded programme reported. The John Templeton Foundation awarded Laland and Uller $7,480,634 at St Andrews from September 2016 to May 2019 to put the extended synthesis to the test; the project completed 22 studies. What they bought is a set of demonstrations that the proposed processes occur — butterfly eyespot patterns constrained by development, Caribbean anoles whose plastic traits are also the more evolvable ones, dung beetle larvae whose growth and horn size depend on a maternally transmitted microbiome. What they did not buy is a single forward prediction that the standard framework would have got wrong. That gap, stated plainly, is the honest size of the prize so far.

Frontier Where the payoff is concrete, it is concrete and small. The clearest measured case is plasticity-led evolution in spadefoot toads: populations that express an environmentally induced carnivore morph about 95% of the time produced tadpoles that beat tadpoles from populations expressing it about 20% of the time in 35 of 51 competitions for the specialist prey, and showed carnivore-like jaw musculature even without the dietary cue. That is a real, quantified result about how often a phenotype is exposed to selection determining how well it is refined, and it is the kind of thing the extended synthesis is for.

Handwave The condition, and it points straight at section 5: the upside is real only if at least one of the three components makes a prediction the standard framework forbids. Niche construction, developmental bias and transgenerational epigenetic inheritance have different evidence behind them and different prospects, and the habit of citing the strongest to support the weakest is the single biggest obstacle to finding out which of them pays.

Established The framework under dispute is the Modern Synthesis, and the dispute is about its structure rather than its facts. The synthesis, assembled between roughly 1918 and 1950, holds that evolution is change in the frequency of genetic variants in populations; that variation arises by mutation and recombination, undirected with respect to fitness; that natural selection, drift, migration and mutation are the processes that change those frequencies; and that development, while real, is a black box between genotype and phenotype that the theory need not open. Nobody in this argument disputes that this framework works.

Frontier The extended evolutionary synthesis proposes that the box must be opened, and sets out what changes if it is. The 2015 statement by Laland, Uller, Feldman, Sterelny, Müller, Moczek, Jablonka and Odling-Smee lists core assumptions — reciprocal causation between organism and environment, inheritance extending beyond genes to epigenetic, ecological, behavioural and cultural channels, non-random phenotypic variation from developmental bias, and developmental systems that both facilitate adaptive variation and modify their own selective environments — and derives ten explicit predictions, among them that “phenotypic accommodation can precede, rather than follow, genetic change” and that “heritable variation will be systematically biased towards variants that are adaptive and well-integrated with existing aspects of the phenotype”.

Established The central analytical point of this brief is that those components are not one claim and do not stand or fall together. Niche construction is a claim about ecology feeding back into selection, and it has a large comparative dataset behind it. Developmental bias is a claim about the distribution of available variation, and it has mechanism and pattern but almost no prospective test. Transgenerational epigenetic inheritance is a claim about heredity, and its evidence is strong in nematodes, real in plants, and contested to the point of denial in mammals. Frontier Treating them as a single position is what allows an argument won on worms to be banked against a claim about people.

Frontier Half of the disagreement is not about observations at all. Reviewers sympathetic to the programme and reviewers hostile to it agree on almost every fact and disagree about what a fact explains. Tim Lewens, reviewing the programme’s 2024 book-length statement, argues that the split between developmentalists and traditionalists reflects “perceptual salience” rather than factual disagreement — different training makes the same developmental facts variously visible. Speculative If that is right, no experiment settles the dispute, and the field needs a different kind of argument than the one it is having.

Frontier This brief takes the position that the dispute is empirically tractable in parts and not in others, and says which. Sections 3 and 7 separate the questions that a designed experiment could answer from the ones that are about standards of explanation. The historiography of framework change in general belongs to Scientific Revolutions, which treats the Kuhnian account and its measurement; this page is about one live structural dispute and its numbers.

Established Start with the component that has the most data: niche construction, and the one quantitative test of its central prediction. Niche construction is the claim that organisms modify their own selective environments — burrows, nests, dams, altered soil chemistry — and that this changes the selection they and their descendants experience. Clark, Deffner, Laland, Odling-Smee and Endler assembled 1,045 temporally replicated, 257 spatially replicated and 1,230 pooled selection gradients from 34 studies and 31 wild species and tested the prediction that selection arising from organism-constructed sources is weaker and less variable than selection from sources the organism does not control. Established Temporal variation in selection was substantially lower for constructed sources, mean absolute selection gradients were lower, and the result survived alternative definitions, taxa and analytical methods — though the spatial result was weaker, the sample smaller, and the two distributions partially overlap.

Frontier Two qualifications the result does not usually travel with. The analysis is a meta-analysis of gradients already in the literature rather than an experiment, so the constructed and non-constructed categories are assigned by the analysts after the fact; and four of its five authors are the principal architects of niche-construction theory. Frontier That does not make it wrong — it is the only quantitative test of the prediction that exists — and it does mean an independent replication with pre-registered category assignments is the cheapest valuable thing anyone could do in this subfield. Speculative The encyclopedia treatment of niche construction written from inside the programme is candid on the same point, giving the earthworm as the canonical case and no quantitative selective advantage for it at all.

Established Developmental bias has mechanism, pattern, and a measurement problem its own advocates state. Uller, Moczek, Watson, Brakefield and Laland define it as bias in the distribution of phenotypic variation arising from the structure of the developmental system, relative to an assumption of isotropic variation, and the evidence is comparative: tetrapod digit number and limb proportions, insect wing size–shape–pigmentation couplings, mammalian tooth morphology predicted across species and matched to the fossil record by gene-network models, Drosophila wing-shape divergence aligning with mutational bias, and mycalesine butterfly eyespot characters whose within-species bias matches their between-species diversity. Established The same paper states the obstacle: the contributions of natural selection and developmental bias to adaptation “are not easily quantified or disentangled”.

Frontier The macroevolutionary version of the claim is suggestive and its own reviewers say so. Stansfield and Parsons set out the case that adaptive radiations run along lines of least resistance: stickleback divergence aligned with the major axis of genetic variance for more than four million years, Lake Victoria cichlids producing more than 500 species in about 10,000 years, repeated benthic–limnetic ecotypes across sticklebacks, cichlids and labrids, and developmental decoupling of oral and pharyngeal jaws facilitating cichlid diversification. Frontier Their own assessment is that the evidence is suggestive rather than definitive and that the macroevolutionary consequences of bias remain poorly understood, which is an unusually honest summary and the one this brief adopts.

Established Transgenerational epigenetic inheritance is where the anchor for this brief needed correcting, and the correction reverses the headline. In Caenorhabditis elegans the phenomenon is real and mechanistically understood: heritable RNA interference silences a target across generations, and the standard duration is a three-to-five generation bottleneck. Persistence is not a smooth decay but a ratchet — silencing continued from the first to the second generation in 46.5% of lineages, from the second to the third in 80.5%, and from the third generation onward in 92.6% to 100%. Established Inheritance past twenty generations is real but is not the phenomenon: it requires disabling the termination machinery — mutants in met-2, an H3K9 methyltransferase, inherit stably — or artificial selection, under which the response can be maintained beyond 80 generations.

Established The genuinely striking result is the opposite of long memory: the worm switches inheritance off on purpose. Houri-Zeevi, Teichman, Gingold and Rechavi showed that starvation for six days, two hours at 37 °C, or 48 hours in 350 mM sodium chloride each reset ancestral small-RNA responses and produced a genome-wide reduction in heritable small-RNA levels, orchestrated by the p38 MAPK pathway and the transcription factor SKN-1/Nrf2 and executed by MET-2. Frontier Duration of epigenetic memory is therefore a regulated trait with dedicated machinery, not a passive persistence, and a theory that treats long inheritance as the interesting finding has the biology upside down.

Established In mammals the same phenomenon is contested to the point of denial, and the critique is specific rather than rhetorical. Adrian Bird’s 2024 review notes that DNA methylation is largely stripped from gametic and early embryonic genomes, that no validated mechanism exists by which the external environment communicates with the germline epigenome, and that the best-characterised mammalian case — the agouti mouse — failed to replicate on larger samples while a screen for equivalent loci was largely negative. Established He records that replication of the Swedish famine study in a much larger cohort failed to reproduce the paternal-grandfather effects, and that the widely cited Holocaust-survivor study rested on 32 people with limited controls. Established Horsthemke’s earlier review makes the definitional point that decides most of these cases: an effect in F1 or F2 after a maternal exposure is fetal programming, not germline transmission, and only an F3 effect qualifies. His conclusion, verbatim, is “For now, I remain skeptical”.

Established Plasticity-led evolution has the best-designed test in the whole programme and it passed. Levis and Pfennig competed spadefoot toad tadpoles from populations that express an environmentally induced carnivore ecomorph at very different frequencies — about 95% in sympatry with a competitor, about 20% in allopatry — and found that tadpoles from high-expression populations grew significantly more on the specialist diet, won 35 competitions to 16, consumed prey faster, and showed more carnivore-like jaw musculature even without the dietary cue, which is the signature of genetic assimilation. Frontier This is a real result and it is one system; the general survey of plasticity-led cases could not be retrieved for this brief, so no claim is made here about how many such systems satisfy the full criteria.

Established The critics’ position is not that these phenomena are unreal but that they are already accommodated. Charlesworth, Barton and Charlesworth reviewed the same advances in genetics, developmental and molecular biology and concluded that “no radical revision of our understanding of the mechanism of adaptive evolution is needed”, with the findings supporting neo-Darwinism rather than directed mutation or inheritance of acquired characters. Frontier Svensson argues separately that the extended synthesis rests on a biased historical narrative — that its advocates exaggerate the reach of the Modern Synthesis, present it as more monolithic than it was, and underweight post-synthesis developments including neutral theory and Wright’s emphasis on interaction and stochasticity.

Frontier The most useful critique is the one that offers a rival explanation for the neglect itself. Reviewing the programme’s 2024 book, David Houle accepts that the five phenomena it names are understudied and proposes that the reason is that “they are difficult to study”: interaction effects need larger samples, more complex designs and harder analyses than holding variables constant. Frontier He also presses the directional point — “creativity and steering imply not just change, but change in a particular direction” — and invokes endogenization, the idea that neglected phenomena get absorbed into existing theory rather than forcing its replacement. Speculative That is a testable sociological claim and nobody has tested it.

Frontier The programme’s own current statement asks for extension rather than replacement, which changes what would count as winning. Evolution Evolving advocates an extension rather than a replacement of the core of evolutionary theory, and the reviewer who calls it a tipping point says so in the same breath. Speculative An extension that nobody denies is not a dispute; the live question is which specific predictions the extension makes that the core does not, and section 7 asks for exactly that.

Frontier Can developmental bias predict the direction of evolutionary change before it happens? Every persuasive case is retrospective: the bias is characterised in extant species and the divergence it matches has already occurred. The prospective version — characterise the bias in an ancestral or source population, register the predicted direction, then impose selection — is the test the field has not run. Speculative Its absence is not an oversight but a consequence of the measurement problem the programme’s own review states: bias and selection are not easily disentangled, and a prediction made after both have acted is not a prediction.

Frontier Is a three-to-five generation memory long enough to matter evolutionarily? This is the question that decides whether epigenetic inheritance belongs in evolutionary theory or in physiology. A variant that persists for four generations and then resets can change the trajectory of a population under rapidly fluctuating selection and cannot accumulate the way a genetic variant does. Speculative The calculation required is a straightforward one in population genetics and this brief could not find it done for a measured inheritance duration and a measured fitness effect. Until it is, the evolutionary significance of the best-documented case of non-genetic inheritance is an assertion on both sides.

Frontier Is niche construction a cause of selection or a redescription of ecology? Critics grant that organisms modify environments and deny that this requires theoretical change; proponents answer that the feedback creates dynamics standard models do not produce — new equilibria, altered rates, fixation of otherwise deleterious alleles, cyclical behaviour. Frontier The Clark analysis is the one place the argument touches data, and what it shows is that constructed selection is weaker and steadier, which is a signature rather than a mechanism. Handwave Whether a weaker, steadier selective regime constitutes a different evolutionary process or a different parameter value in the same one is exactly the kind of question that has no experiment attached to it.

Frontier Does the extended synthesis forbid anything? A framework earns its keep by ruling things out. The ten published predictions are mostly of the form “X will frequently occur”, which the standard framework also permits at some frequency. Speculative The sharpest available candidate for a genuine disagreement is prediction (ix) — that heritable variation is systematically biased towards variants that are adaptive and well integrated with the existing phenotype — because the standard framework’s randomness assumption is about fitness, and a systematic adaptive bias in the supply of variation is a real departure. Frontier It is also the prediction most vulnerable to equivocation, since “non-random” in the sense of non-uniform is uncontroversial and “non-random” in the sense of fitness-directed is not.

Speculative Is the disagreement about facts or about what is visible? Lewens argues the split reflects perceptual salience produced by different training rather than disagreement about the world. Handwave If that diagnosis is right, then the empirical programme in section 7 will settle individual claims and leave the framework dispute exactly where it was, and the honest thing for the field to do is stop describing a disagreement about explanatory taste as a disagreement about evidence.

Frontier Is neglect explained by bias or by difficulty? Houle’s alternative — that these phenomena are understudied because they are hard, needing larger samples and more complex designs — is a claim about the sociology of a discipline that could be checked against grant outcomes, sample sizes and statistical power in the published literature. Speculative Nobody has checked it, and it is the sort of question Science of Science is built to answer.

Speculative What is the measured fitness consequence of inclusive inheritance? The dung beetle result from the funded programme — maternal microbiome transmission affecting larval growth rate and horn size — is the clearest candidate, because horn size is under sexual selection and the transmission is non-genetic. Speculative Turning that into an evolutionary claim requires a selection coefficient attached to a transmission fidelity, and the brief found neither published together.

Frontier The first bottleneck is that nobody can measure developmental bias in a population without also measuring the selection acting on it. The quantity wanted is the distribution of phenotypes that perturbation would produce, in the absence of selection; what is observed is the distribution that survived. Speculative Mutation-accumulation lines, induced-mutation panels and controlled environmental perturbation are the available approximations, and each measures something adjacent to the target. Handwave This is the technical core of the dispute and it is discussed less than any part of it.

Established The second is the F3 barrier in mammals, which is a cost problem before it is a biology problem. Demonstrating germline transmission requires the third generation, because F1 and F2 can be explained by direct exposure of the fetus and its germ cells. For a mouse that is roughly a year of breeding per replicate; for humans it is a multi-decade cohort confounded by shared genotype, shared household and shared culture. Frontier Cross-fostering, embryo transfer and gamete-only transmission designs can separate these and multiply the cost again.

Frontier The third is statistical power, and the critics have the better of this argument. Every process the extended synthesis emphasises is an interaction — genotype by environment, development by selection, organism by constructed environment — and interactions need far larger samples than main effects. Speculative A field that mostly publishes single-system demonstrations with tens of replicates is structurally unable to estimate the effects it is arguing about, which is an argument for consortium-scale work that nobody is funding.

Frontier The fourth is the bundling problem, and it is the one this brief was commissioned to take seriously. “The extended evolutionary synthesis” names a package whose components have different evidence: a comparative dataset of 1,230 selection gradients, a body of comparative morphology with no prospective test, and an inheritance phenomenon robust in nematodes and unestablished in mammals. Established Bundling lets a strong result in one component be cited as support for a weak claim in another, and it is why this page separates them at every point.

Frontier The fifth is that evolutionary biology has almost no pre-registration culture. The disputes here are about whether predictions were made in advance, and the discipline has no standard mechanism for demonstrating that they were. Speculative Registered reports exist in the journals of neighbouring fields and are rare in this one; their absence is why a retrospective match between a developmental bias and a phylogenetic pattern can be published as a confirmed prediction. Handwave The cost of fixing this is an editorial decision, which makes it the cheapest intervention available and explains nothing about why it has not happened.

Frontier The developmental-bias programme depends on regulatory-network biology it does not produce. The claim that perturbation produces some phenotypes more readily than others is a claim about gene-regulatory architecture, and it becomes quantitative only when networks can be reconstructed and perturbed at scale. The mammalian-tooth models that predict morphology across species and match the fossil record are the existence proof; extending that to traits under selection in the field is the dependency. Frontier Noncoding Genome Function owns the regulatory-sequence question this rests on.

Established The epigenetic-inheritance programme depends on germline assays that track marks through reprogramming. Bird’s objection is mechanistic: methylation is stripped in gametes and early embryos, so a carrier must either escape erasure or be re-established from something else. Small RNAs are the candidate in nematodes; the mammalian equivalent is the missing measurement.

Frontier Everything in section 7 depends on long-term experimental evolution infrastructure. Multi-generation selection lines with archived ancestors are the only apparatus that lets a prediction registered at generation zero be scored at generation fifty, and they are maintained by individual laboratories on grant cycles far shorter than the experiments. Speculative This is a facility problem of the kind Cloud Laboratories addresses in a different domain, and no equivalent exists for evolution.

Speculative The niche-construction programme depends on ecological time series it borrows from conservation biology. Measuring how a constructed environmental component alters selection requires both the selection gradient and the state of the constructed component over years, which is the data Biodiversity Restoration generates for other reasons.

Frontier And the whole dispute depends on a methodological import: registered reports. This is not a scientific dependency but an editorial one, and it is the only dependency on this list that could be satisfied in a year.

Frontier The tests below are ranked by how much each would move the argument, and the first is the only one that could settle a framework-level claim rather than a local one.

Speculative Tier one, and the decisive experiment for this brief: pre-registered experimental evolution in which developmental bias is measured in the ancestral population first, the predicted direction of change is registered, and only then is selection imposed. The design is specific. Characterise the ancestral population’s response to perturbation — mutational, environmental, or both — to estimate the axes along which phenotypic variation is produced most readily. Register the predicted direction and magnitude of divergence. Impose a selection regime orthogonal to, and one aligned with, the predicted axis. Compare realised divergence against both the registered developmental prediction and a null in which only the additive genetic variance matters. Frontier This is the one experiment that could show developmental bias doing work the standard framework does not already do, because it forces the prediction to be made before the outcome exists. Handwave No such experiment has been published, the programme’s own review says bias and selection are not easily disentangled, and its critics say the phenomena are neglected because they are hard — which are two descriptions of why this test has not been run.

Speculative Tier two is the mammalian F3 test with genetic and cultural transmission controlled out. Expose F0, breed to F3, and separate germline transmission from fetal programming and from behavioural transmission by embryo transfer and cross-fostering. Established The bar is set by the failures already recorded: the agouti result that did not replicate at larger sample size, the Swedish famine paternal-grandfather effect that did not reproduce in a larger cohort, and the Holocaust-survivor study with 32 participants. Frontier A single adequately powered, pre-registered F3 mammalian result would do more for this component than the entire existing literature, in either direction.

Frontier Tier three is niche construction with the construction removed. The prediction is that selection from organism-constructed sources is weaker and less variable, and it currently rests on a meta-analysis whose categories were assigned by the theory’s own authors. Speculative The experimental version removes or replaces the constructed component — the burrow, the mucus-modified soil, the shared microbiome — and measures the selection gradient with and without it in the same population. Speculative An independent replication of the comparative result with pre-registered category assignment is the cheaper version and would be worth almost as much.

Frontier Tier four converts the nematode result into an evolutionary one. The duration of small-RNA inheritance is now a manipulable variable: met-2 mutants inherit stably, stress resets inheritance through p38 MAPK and SKN-1, and selection can extend the response beyond 80 generations. Speculative The missing experiment is a fitness assay: compete lineages differing only in inheritance duration under fluctuating selection and measure whether a three-to-five generation memory changes allele frequencies enough to matter. Handwave Until somebody does this, the evolutionary significance of the best-characterised case of non-genetic inheritance rests on the plausibility of the mechanism rather than on a measured effect.

Established Tier five has been run once and should be run again in other systems. The spadefoot design — compare populations that differ in how often an induced phenotype is expressed, and test whether higher expression frequency predicts better refinement — is a clean test of a plasticity-led prediction and it returned 35 wins to 16 with assimilated jaw morphology. Frontier Its value is that the prediction is quantitative and directional; its limitation is that it is one genus, and replication in an unrelated taxon is the obvious next step.

Speculative Tier six is an audit rather than an experiment, and it is overdue. The ten predictions published in 2015 are on the record. Scoring them against the decade of literature since — which were tested, which were confirmed, which were confirmed in a form the standard framework also predicts — is a bounded piece of work that would tell the field whether its own programme is progressing. Frontier The funded project completed 22 studies and reported them as supportive; no independent audit against the published predictions exists.

Handwave Tier seven is the test nobody has designed: find a phenotype the standard framework forbids. A framework is distinguished from an emphasis by what it rules out, and no extended-synthesis prediction currently rules out an observation the Modern Synthesis permits. Speculative If, after another decade, no such prediction has been formulated, the honest conclusion is Houle’s: the phenomena are real, difficult, and being absorbed — and the framework dispute was about salience all along.

Established Nothing here needs a new instrument class; it needs throughput and duration. The apparatus of this dispute is breeding rooms, growth chambers, sequencers and phenotyping rigs, all of which exist. What does not exist is the scale: interaction effects need sample sizes an order of magnitude above the norm in the field, and the multi-generation designs need lines maintained past the length of a grant.

Frontier The binding engineering requirement is automated high-throughput phenotyping. Measuring the shape of available phenotypic variation means measuring many phenotypes on many individuals under many perturbations; the limiting step in every design in section 7 is how many organisms can be measured precisely enough to resolve an axis of variation. Speculative Imaging-based morphometrics and automated behavioural assays have cut that cost in model systems, and they do not exist for most of the taxa where the comparative evidence was gathered.

Frontier The second is germline epigenome measurement through reprogramming. Tracking a mark from an exposed parent through gamete formation and early embryogenesis into an F3 germline requires single-cell resolution on very small cell populations at defined developmental stages. Speculative The methods that would make this routine are the ones Genetic Engineering and Microbiome Engineering develop for other purposes, and the evolutionary question is a beneficiary rather than a driver.

Speculative The third is a records requirement rather than a device: archived ancestral populations. A registered prediction about the direction of evolution is only scorable against a frozen ancestor, and cryopreserved founder stocks are cheap, unglamorous and routinely not kept.

Frontier The closest neighbour on this map is a subject in physics, and the parallel is structural rather than decorative. Dark Matter Alternatives describes two programmes that agree on the measurements, disagree about what counts as an explanation, and each absorb failures by adding a component. The mechanism of stalemate is the same one this page documents, and in both cases the cheapest available remedy is a pre-registered protocol nobody is obliged to agree.

Established The regulatory-genome question is a direct input. Developmental bias is a claim about what regulatory architecture makes easy, and Noncoding Genome Function owns the dispute about how much of that architecture is functional — a dispute with the same shape and better data.

Frontier Somatic evolution is the place these ideas are already operational. Theories of Carcinogenesis treats a system in which selection, developmental state, epigenetic inheritance across cell generations and niche construction by the tumour all act at once and on a timescale that can be observed — which makes it the best available natural laboratory for claims this brief cannot test in whole organisms.

Frontier The historiography belongs next door and this page does not duplicate it. Scientific Revolutions covers whether conceptual change is punctuated at all, and reports that the one quantitative measurement recovered gradual modification rather than abrupt replacement. Speculative That result, if it generalises, is bad news for the framing of this dispute as a succession crisis and good news for the idea that the components will be absorbed one at a time. Philosophy of Science owns the question of what makes an explanation adequate, which is the half of this argument no experiment reaches.

Speculative Applied evolution is where the claims would pay or fail. Resistance management, biological control and crop improvement under a shifting climate all need short-horizon predictions of the direction of change; Biodiversity Restoration and Microbiome Engineering are where a real predictive gain would show up first, and neither currently reports one.

Established One foundation paid for the decisive phase of this dispute, and that fact is part of the dispute. The John Templeton Foundation awarded $7,480,634 to St Andrews for Putting the Extended Evolutionary Synthesis to the Test, running from September 2016 to May 2019 and producing 22 studies. Frontier A single private funder underwriting the empirical programme of one side of a framework dispute is an unusual arrangement, and it is cited by critics as a reason for suspicion and by proponents as evidence that public funders would not take the risk. Both readings are available and neither is testable.

Frontier The review process for framework-level claims has visibly failed at least once. In January 2025 Nature published a review of the programme’s book-length statement written by a scholar who had co-authored a manifesto for the same position with two of the book’s authors, and declined a letter pointing this out. Speculative The substantive problem is not the individual choice but that a small field has no mechanism for finding a reviewer who is competent and uncommitted, which is a general problem with framework disputes and an acute one here.

Frontier Journals could settle the pre-registration question unilaterally and have not. Registered reports would convert most of the arguments in section 3 from irreproducible disagreements about who predicted what into checkable records. Speculative The obstacle is that the format fits confirmatory work and much of this field is exploratory, which is a real objection and not a sufficient one for a dispute whose core complaint is about prediction.

Speculative The programme’s framing has drifted in a way that changes what institutions should do about it. The 2015 statement described an alternative conceptual framework; the 2024 book advocates extension rather than replacement. Handwave If the claim is now extension, the appropriate institutional response is a series of funded experiments on specific mechanisms rather than an argument about synthesis, and the field’s conference and review structures have not caught up with its own moderation.

Frontier Finally, there is a boundary-policing cost nobody wants to pay. The vocabulary of this dispute — that the standard account is inadequate, that development directs evolution — is routinely appropriated by anti-evolution advocacy, and the effect is to make working biologists more defensive of the standard framework than the evidence alone warrants. Speculative That is a real distortion of the scientific argument by an external one, and it runs in the opposite direction to the funding distortion above.

Established The ethically loaded claim in this brief is transgenerational epigenetic inheritance in humans, and it is the claim with the weakest evidence. The proposition that a grandparent’s famine, trauma or smoking is written into a descendant’s biology has moved into clinical, legal and policy language well ahead of the data: the famine result did not replicate in a larger cohort, the trauma study had 32 participants, and the definitional standard that would make any of it germline transmission — an effect surviving to F3 — is met by none of the human cases.

Frontier The harm is specific rather than abstract. A weakly supported claim of inherited biological damage attaches to populations who were harmed once already, and it supplies a biological framing for disadvantage whose social causes are documented and whose epigenetic causes are not. Speculative It also invites the inverse error — treating a non-replicating molecular result as the reason to act, when the historical injustice was reason enough.

Frontier There is a second-order duty on the communicators. The mechanisms are genuinely exciting, the nematode work is beautiful, and the distance from a worm switching off inherited small RNAs to a human inheriting an ancestor’s experience is enormous and rarely stated. Speculative The field’s own critics have done most of the stating, which suggests the incentive structure rewards the leap.

Frontier The funding arrangement raises a narrower question of disclosure rather than of legitimacy. Private money for unorthodox programmes is a defensible mechanism and the amounts here are public. What is not routine is the disclosure of intellectual commitment in review — the 2025 episode turned on an undeclared co-authorship rather than on money. Handwave A norm requiring reviewers of framework-level claims to declare prior public positions would cost nothing and does not exist.

Speculative Finally, the appropriation problem cuts both ways and deserves to be named rather than managed. Suppressing a live scientific dispute because bad actors quote it is a worse failure than the quoting; the correct response is the one this page attempts, which is to state precisely which component has which evidence.

Frontier If the extended synthesis is right in its strong form, the practical change is a forecasting capability on a decadal horizon, and that is a smaller prize than the rhetoric implies. Predicting which resistance phenotype will appear, which invasive lineage will establish, or which crop trait will assimilate under a warming climate are all short-horizon directional questions, and all of them are currently answered by observing what happened.

Speculative If it is right in its weak form — an extension rather than a replacement — the civilisational consequence is close to zero and the scientific one is real. Evolutionary biology absorbs development, inheritance broadens, and the textbooks change a chapter. Established That is the modal outcome and it is what the programme’s own current statement asks for.

Frontier The larger effect is on how a society reads biological inheritance. A public that believes acquired experience is heritable reasons differently about responsibility, about compensation for historical harm and about what medicine owes to descendants. Speculative That shift is already happening on evidence the specialists describe as inconclusive, which makes this an unusual case of a scientific dispute whose social consequences are running ahead of its resolution.

Handwave And there is a cautionary structural reading. A dispute in which both sides accept the observations, disagree about what an explanation requires, and have no experiment that either could fail is a dispute that can run indefinitely while producing literature. Speculative The same shape appears in Dark Matter Alternatives, and the remedy in both cases is the same unglamorous one: register the prediction before the result.

These horizons track what could be decided and by what, because there is no technology here to track:

• 10 yr: Frontier Expect the nematode inheritance system to be turned into a quantitative evolutionary model — duration is now a manipulable variable and the fitness assay is straightforward — and expect that to be the component that moves first. Frontier Expect at least one adequately powered mammalian F3 study with embryo transfer and cross-fostering controls, and expect it to be reported as negative or ambiguous. Speculative Expect the prospective developmental-bias experiment to remain unrun unless a journal commissions it as a registered report.
• 25 yr: Speculative This is the window in which the audit becomes possible: the ten predictions of 2015 will have had thirty years of literature to be scored against, and the answer will be legible. Speculative The modal outcome is absorption — developmental bias and inclusive inheritance become standard chapters, niche construction becomes a term in ecological modelling, and no framework is replaced. Handwave The minority outcome, in which a prospective developmental prediction beats a quantitative-genetic null in a registered test, would be the first genuine framework result this dispute has produced.
• 50 yr: Handwave On this horizon the interesting question is no longer which side won but whether the dispute was of the kind that has winners. Speculative If Lewens is right that the split is about perceptual salience, the record in fifty years will show two literatures that stopped talking rather than one that was refuted, and the useful residue will be the specific mechanisms rather than the synthesis.
• 100 / 250+ yr: Handwave Beyond useful forecasting. The defensible statement is structural: frameworks in biology have historically been extended rather than overturned, and the components at issue here are already being extended into. Handwave Any claim about what evolutionary theory will look like in two centuries is an assertion, and this page does not make one.

• Depends on Two results on this map and neither blocks. Noncoding Genome Function owns the regulatory architecture that developmental bias is a claim about, and its dispute over how much of the genome is functional sets the terms for how measurable bias can be. Theories of Carcinogenesis owns the one system where selection, developmental state, heritable epigenetic change and niche construction are all observable on a human timescale. Nothing on this map produces a result this brief waits on; the dependencies that bind are measurements and an editorial norm, typed below.
• Requires (not on this map) Six conditions, in the order of the tokens above. The prospective measurement. A way to estimate the axes along which a population produces phenotypic variation most readily, before selection acts on it, good enough to register a directional prediction; the programme’s own review says bias and selection are not easily disentangled, and this is that problem stated as a requirement. The editorial norm. Registered reports as the default format for experimental-evolution claims, which would convert most of the arguments in section 3 into checkable records at the cost of an editorial decision. The phenotyping. Automated morphometric and behavioural measurement at thousands of individuals per perturbation, because every effect at issue is an interaction and interactions need samples an order of magnitude above this field’s norm. The germline assay. Single-cell tracking of candidate epigenetic carriers through gamete formation and early embryogenesis, which is the measurement that would answer the reprogramming objection in mammals rather than argue with it. The disclosure norm. A requirement that reviewers of framework-level claims declare prior public positions, absent in January 2025 when it was needed. The applied pull. A breeding, biocontrol or resistance-management programme whose own economics justify three-generation experiments, because no basic-science grant cycle covers them.
• Enables A validated prospective prediction of evolutionary direction would give Biodiversity Restoration and Microbiome Engineering a forecasting tool they currently lack, and would matter to any programme that has to anticipate resistance. A negative result enables tidying: the components get absorbed, the framework claim lapses, and the field stops arguing about synthesis. No typed enabling edge is claimed, because neither outcome is scheduled and the honest expectation is absorption without resolution.
• Adjacent Within this map: Dark Matter Alternatives for the same structure of stalemate in physics, Scientific Revolutions for whether framework change is punctuated at all, Philosophy of Science for what makes an explanation adequate, Science of Science for whether neglect tracks difficulty, Eukaryogenesis for a macroevolutionary transition where developmental argument has no purchase, and the central dogma and genome modules for the inheritance machinery the dispute is about.

Established “The extended evolutionary synthesis is one claim.” It is at least three, and they are not equally supported. Niche construction has a comparative dataset of 1,230 selection gradients from 31 species behind its central prediction. Developmental bias has mechanism, striking comparative pattern, and no prospective test. Transgenerational epigenetic inheritance is robust in nematodes, real in plants, and unestablished in mammals. Frontier Any sentence that assigns one level of confidence to “the EES” is averaging over three different states of evidence.

Established “Epigenetic inheritance means your grandparents’ experiences are written into you.” In humans this is not established, and the specific cases have failed. An effect in a child or grandchild after a maternal exposure is fetal programming, because the fetus and its germ cells were directly exposed; only an effect surviving to the third generation demonstrates germline transmission. Established The Dutch famine outcomes are fetal programming on this standard; the Swedish paternal-grandfather effect failed to reproduce in a larger cohort; the Holocaust-survivor study had 32 participants. Frontier The mechanism objection is separate and harder: DNA methylation is largely stripped from gametes and early embryos, and no validated route exists by which the external environment writes to the germline epigenome.

Frontier “Small-RNA inheritance in worms lasts more than twenty generations, so epigenetic memory is deep.” This is the claim this brief was commissioned against, and it is the wrong way round. The standard duration is a three-to-five generation bottleneck. Inheritance past twenty generations requires breaking the machinery that ends it — met-2 mutants inherit stably — or artificial selection, which can push the response beyond 80 generations. Established The real finding is that termination is active and regulated: starvation, heat and osmotic stress each reset ancestral responses through the p38 MAPK pathway and SKN-1, which means the organism has evolved machinery to stop inheriting. Speculative A theory of heredity built on the long tail has selected the artefact and discarded the result.

Established “This vindicates Lamarck.” No, and the people making the scientific case mostly do not say it. Nothing in the evidence supports use-and-disuse driving adaptive change, and the strongest critique of the programme rests precisely on the distinction between variation that is non-uniform and variation that is directed toward fitness. Frontier Conflating the two is the single most common error in popular coverage, and it is made in both directions — by advocates claiming more than the data and by critics dismissing genuine non-uniformity as Lamarckism.

Frontier “The Modern Synthesis denied development, plasticity and inheritance beyond genes.” This is disputed history, and the dispute matters. Svensson argues that advocates exaggerate the reach and monolithic character of the Modern Synthesis, underweight post-synthesis developments including neutral theory, and overlook Wright’s persistent emphasis on gene interaction and stochasticity. Speculative If the historical premise is overstated, the case for a new framework weakens without any observation changing, which is why the historiography is being fought over rather than cited.

Frontier “Niche construction is the extended phenotype under a new name.” Partly, and the part that is not is the part with data. That organisms modify their environments is not in dispute. The distinctive claim is that this feedback changes evolutionary dynamics — new equilibria, altered rates, fixation of otherwise deleterious alleles — and the one quantitative test found constructed selection weaker and less variable than non-constructed. Speculative That is a signature, not a mechanism, and it was produced by the theory’s own authors from literature they categorised; it needs independent replication more than it needs defence.

Frontier “The Templeton programme tested it and it passed.” The programme produced demonstrations, not tests. $7,480,634 over roughly three years and 22 studies established that developmental constraint shapes butterfly eyespots, that plastic anole traits are more evolvable, and that a maternally transmitted microbiome affects dung beetle growth and horn size. Speculative Each result shows that a proposed process occurs; none shows the standard framework predicting the outcome wrongly, which is what a test of a framework would be.

Speculative “Doubting the Modern Synthesis is creationism.” It is not, and treating it as such damages the science. The anti-evolution use of this vocabulary is real and it makes working biologists defend the standard framework more firmly than the evidence alone warrants. Handwave The remedy is not to suppress the dispute but to state which component carries which evidence, which is the whole method of this page.

Handwave “One side is going to win.” The strongest thing this research found is that the current statements of both positions are compatible. The programme’s 2024 book asks for extension rather than replacement; its most serious critic grants the phenomena are understudied and says the reason is difficulty. Speculative A reader who leaves this page expecting a winner has read a dispute about explanatory emphasis as a dispute about facts, which is the error the page exists to prevent.