Pathological Science

Established The prize is a calibrated base rate: knowing, for a new anomalous claim, roughly how long it will take to resolve and which single test resolves it. The six documented episodes on this page give the first data points. N-rays died in about a year (1903–1904); polywater took about eleven years (1962–1973); the EmDrive ran twenty years (2001–2021); LK-99 was settled in weeks (July–August 2023). Against that spread, cold fusion is the outlier that defines the value of getting this right: thirty-four years after the 1989 claim it is still open, having absorbed two U.S. Department of Energy reviews (1989 and 2004) and, in 2023, a third arm of the same department — ARPA-E — putting up about $10 million across eight teams to prove or debunk it.

Frontier Stated in the units that matter, the upside is a decision rule about when to fund a decisive test rather than let a claim fester. ARPA-E’s eight awards total roughly $9.96 million and run about three years; Google’s parallel effort spent a comparable sum ($10 million over three years, reported in 2019) and returned a clean null plus new calorimetry and materials science. The comparison the natural history makes legible is blunt: a properly powered kill-or-confirm test costs on the order of $10 million and three years, against a controversy that has stood for over three decades and recurs every few years in a new material.

Frontier Who collects is funders, journal editors, and institutions that must triage extraordinary claims — not the working scientist at the bench. The gain is avoided waste on both sides: fewer decades spent not-quite-dismissing a claim, and fewer real anomalies killed by a premature “pathological” label. The history records almost no false burials, which is itself the finding.

Speculative The condition, stated first because the upside depends on it: the base rate is real only if the resolution mechanism generalises. If most episodes end the same way — someone blinds the observer or names the mundane artifact, and the effect vanishes — then resolution time is predictable and the decisive test is cheap. If cold fusion is not an outlier but the normal shape of a diffuse claim class, the base rate is small and the confidence it offers is false. Section 7 says what the one test is that would move cold fusion from the second column to the first.

In a colloquium at the General Electric Research Laboratory on 18 December 1953, Irving Langmuir — a Nobel laureate in chemistry — described what he called pathological science, “the science of things that aren’t so.” He never published it; the account survives as a transcript from a poor audio recording. His subject was not fraud. It was the way careful, honest, credentialled scientists convince themselves and a community of a real effect that turns out not to exist, and how such episodes run their course.

Established Langmuir gave the class six symptoms, and the sixth is a natural-history claim rather than a laboratory one. The effect is produced by a cause of barely detectable intensity and its magnitude is independent of that intensity; the effect stays near the limit of detectability and needs many measurements of low statistical significance; there are claims of great accuracy; fantastic theories contrary to experience are invoked; criticisms are met by ad hoc excuses; and — the sixth — the ratio of supporters to critics “rises up to somewhere near 50% and then falls gradually to oblivion.” That last is a statement about the life-cycle of a claim, and it is what this brief takes as its object.

This is not a brief about any one of the effects. It is a brief about how the episodes resolve. The claim class is well populated and still growing: Blondlot’s N-rays (1903), mitogenetic radiation, the Allison effect, polywater (1960s), Benveniste’s “water memory” (1988), Fleischmann and Pons’s cold fusion (1989), the OPERA superluminal-neutrino result (2011), the STAP stem-cell claim (2014), the EmDrive (2001–2021), and the LK-99 room-temperature-superconductor claim (2023). The meta-question is empirical and answerable from the record: how do such claims die, how long does it take, and by what mechanism — and what is different about the one that will not.

Frontier The organising finding is that the class has a bimodal outcome. Most episodes resolve, often fast, when someone runs a blinded replication or identifies the mundane artifact that produced the signal. A minority — in the modern record, essentially cold fusion alone — never gets a clean kill, acquires a durable minority community, and gets re-adjudicated by institutions. Cold fusion is the reason the class is interesting: it is the only claim a government department has formally ruled on twice and then reopened a third time, which makes the 2023 ARPA-E topic the single best-defined natural experiment the subject has ever had.

Established “Pathological” here is a diagnosis of a process, not a verdict on people. Blondlot, Deryagin, Fleischmann and the LK-99 authors were not frauds; they were competent scientists who saw a real reading and drew the wrong conclusion from it. Keeping that distinction is the whole discipline of the subject, because the same word is routinely used as a weapon to end inquiry rather than as a description of how an honest error propagates.

Established Langmuir’s framework is the spine, and its diagnostic content is the two experimental symptoms, not the sociological one. Symptoms one and two — a maximum effect from a barely detectable cause, with magnitude independent of cause intensity, and an effect that stays at the limit of detectability requiring many low-significance measurements — are what let honest observers deceive themselves. When the signal sits at the threshold of perception and does not scale with the input, the reading is dominated by the observer’s expectation. That is the common thread through every episode below, and it is why the resolution mechanism is almost always the same: remove the observer’s expectation, or name the thing at the threshold that is actually producing the reading.

Established N-rays are the cleanest case and set the template: about one year, resolved by blinding. René Blondlot, a respected physicist at Nancy, reported N-rays in 1903; dozens of papers followed from French laboratories, all reporting faint changes in the brightness of a spark or phosphorescent screen judged by eye — symptom two exactly. In 1904 the American physicist R. W. Wood visited Blondlot’s laboratory and, in the dark, surreptitiously removed the aluminium prism that supposedly dispersed the rays. Blondlot’s assistant continued to report the spectrum from an apparatus that could no longer produce it. The effect was in the observation, not the world, and once the observation was blinded the effect was gone. Established The whole class had essentially collapsed within about a year of Wood’s intervention.

Established Polywater is the slow case: about eleven years, resolved by identifying the contaminant. Nikolai Fedyakin observed anomalous water in narrow glass capillaries in the early 1960s; Boris Deryagin presented the work at Nottingham in 1966, Brian Pethica’s Unilever group reproduced the observations by 1969, and Ellis Lippincott’s spectroscopic paper in Science on 27 June 1969 gave it a name and a “honeycomb” structure. The reversal came from contamination analysis: Denis Rousseau published in Science in January 1971 that polywater spectra matched ordinary sweat — his decisive demonstration used perspiration from his own shirt after handball — and traced the anomalies to sodium, potassium, silicates leached from glass, and organic aerosols. Established Deryagin conceded in 1973 that the experiments “do not support the hypothesis of anomalous or polymeric water.” The episode ran roughly 1962–1973, and it died when the mundane cause was named rather than when the theory was refuted.

Established Cold fusion is the anomaly among the anomalies, and it is the reason this brief exists. On 23 March 1989 Martin Fleischmann and Stanley Pons announced, at a press conference in Utah rather than in a paper, that palladium electrodes loaded with deuterium produced excess heat attributable to nuclear fusion. Within weeks major laboratories failed to reproduce the neutron signature that fusion required, and the claim was widely judged dead. It was not. Established A U.S. Department of Energy panel — the Energy Research Advisory Board — reviewed it in November 1989 and found no convincing evidence that useful energy came from nuclear reactions in the cells, declining to recommend a special programme while leaving the door open to modest, peer-reviewed work.

Frontier The second Department of Energy review, in 2004, is the fact that makes cold fusion a re-adjudication case rather than a closed one. Fifteen years on, DOE convened a fresh panel of roughly eighteen reviewers. On whether the experiments produced excess power as heat the panel “split about evenly”; on whether that power came from nuclear reactions, “two thirds of reviewers found the evidence unconvincing, one person found it compelling, and the remainder were somewhat convinced.” The panel recommended that agencies entertain well-designed cold fusion proposals through ordinary peer review. Frontier Asked whether anything had changed since 1989, DOE’s James Decker allowed that “that may be a fair comment” — a state department, twice, declining to close a claim it could not confirm.

Frontier The 2019 Google-funded effort is the strongest single attempt to kill or confirm the effect, and it did neither cleanly. Curtis Berlinguette, Yet-Ming Chiang and colleagues reported in Nature in 2019 that a multi-institution programme, funded by Google over about three years at a cost near $10 million, had found “no evidence that the phenomenon exists” across three approaches — electrochemical loading of palladium, high-temperature calorimetry, and pulsed-plasma deuterium loading. Frontier But the authors declined to call the case closed: a by-product was “new insights into highly hydrided metals,” and they held that “there remains much interesting science to be done in this underexplored parameter space.” A well-funded, rigorous null that its own authors would not treat as a burial is the signature of this claim class.

Frontier ARPA-E’s 2023 topic is the anchor: the only case in which a state re-adjudicates a claim its own department has already ruled on twice. In February 2023, thirty-four years after the 1989 review, ARPA-E announced about $10 million across eight projects to establish whether low-energy nuclear reactions (LENR) are real, framed explicitly as an effort to break the field’s stalemate and to prove or debunk the effect. Established The awards, drawn from universities, a national laboratory and a small business, were: Amphionic LLC ($295,924) on nanostructured palladium composites; the Energetics Technology Center ($1,500,000) on scintillator detectors near co-deposition cathodes; Lawrence Berkeley National Laboratory ($1,500,000) quantifying nuclear reactions in metal hydrides below 500 eV; MIT ($2,000,000) on neutron emission from laser-stimulated metal hydrides; Stanford ($1,500,000) on nuclear-product detection from deuterated nanoparticles; Texas Tech ($1,150,000) on materials characterisation and nuclear-product detection; and two University of Michigan awards ($1,108,412 and $902,213) on excess-heat evaluation with microwatt-resolution calorimetry and on ionizing-radiation detection.

Established The design of the ARPA-E topic is itself the finding: it is built around the two things every earlier cold fusion claim lacked. The projects pair excess-heat calorimetry at microwatt resolution with direct detection of nuclear products — neutrons, gammas, ionizing radiation, transmutation ash. That pairing is the point. The 1989 claim rested on a heat measurement with a nuclear-ash budget that did not add up; the field’s thirty-year impasse is precisely that excess heat has been reported without the commensurate nuclear signature, and nuclear signatures without commensurate heat. Frontier ARPA-E’s wager is that modern instrumentation applied to both channels at once, on the same samples, will finally force the question to a yes or a no.

Established LK-99 is the fast modern case, and it shows the base rate shifting: weeks, resolved by global preprint replication. In late July 2023 a South Korean group posted two arXiv preprints claiming a copper-doped lead apatite, LK-99, superconducted above 100 °C at ambient pressure, with a video of partial levitation. Within days laboratories worldwide attempted synthesis; within about three weeks the resolution was in hand. Established The apparent resistivity drop was traced to a first-order structural transition of a copper(I) sulfide (Cu₂S) impurity near 104–112 °C, and the partial levitation to ordinary ferromagnetism, not the diamagnetic expulsion a superconductor shows; the pure material is a wide-gap insulator. Work synthesising these conclusions was subsequently peer-reviewed. The claim was not fraud and not superconductivity — it was a real measurement of an impurity.

Established The EmDrive is the case where a conservation law was the prior and the artifact was thermal: about twenty years, resolved by better metrology. Roger Shawyer proposed the reactionless microwave thruster around 2001; NASA’s Eagleworks group reported a thrust-to-power ratio of 1.2 ± 0.1 mN/kW in a peer-reviewed paper in 2016, despite the device violating conservation of momentum. In 2021 Martin Tajmar’s group at TU Dresden, replicating the Eagleworks setup on a high-resolution balance, showed the apparent thrust was an artifact of thermal warping of the mounting as the cavity heated, plus interaction with Earth’s magnetic field. Established Their conclusion was categorical: “Our measurements refute all EmDrive claims by at least 3 orders of magnitude.” The signal was real; it was the balance moving, not the drive.

Frontier The pattern across the six is consistent enough to state as a provisional law and honest enough to state its exception. Resolution comes by one of two moves — blind the observer so expectation cannot enter (N-rays, and the 1988 Nature team’s blinded re-run of Benveniste’s water-memory assays), or identify the mundane cause and show the effect is it (polywater’s sweat, LK-99’s Cu₂S, the EmDrive’s thermal drift). Resolution time has fallen sharply as the tools of replication have globalised: about a year in 1904, about eleven years in the 1960s–70s, weeks in 2023. Frontier Cold fusion breaks the law on both counts — no single blinding kills it because it is a diffuse family of claims rather than one experiment, and no single mundane cause has been shown to account for all reported excess heat — which is exactly why it, alone, gets re-adjudicated by the state.

Frontier Is there a predictive model of which anomalous claims resolve fast and which persist? The record suggests the discriminating variable is not the strangeness of the claim but its structure: a single, sharply specified experiment (N-rays, LK-99, the EmDrive) resolves fast because one decisive test addresses the whole claim, whereas a diffuse family of loosely specified protocols (LENR) has no single test whose failure the community treats as dispositive. This is a testable hypothesis about claim topology and nobody has fit it to the episode record.

Frontier Why is cold fusion uniquely persistent, and is that a property of the claim or of the sociology? Two accounts compete. One: the effect is a chimera and its persistence is symptom six stalled — a supporter community that never fell to oblivion because irreproducibility is reframed as material sensitivity. The other: there is a small, real, poorly understood effect in loaded metal hydrides that ordinary calorimetry cannot cleanly separate from systematic error. Frontier The two accounts make the same near-term prediction (mostly null results with occasional unreplicable positives), which is why thirty-four years have not separated them and why a purpose-built test is needed.

Speculative Has preprint culture permanently lowered the resolution time? LK-99 resolved in weeks because synthesis recipes and null results moved on arXiv and social media at the speed of the claim itself. Whether this is a durable shift or an artifact of one unusually easy-to-synthesise material is open: cold fusion’s materials are notoriously batch-dependent, so open protocols may not compress its timeline the way they compressed LK-99’s. Frontier The natural experiment is already running in every new claim.

Frontier Does state re-adjudication resolve claims or launder them? ARPA-E’s topic is the first chance to answer this directly. A funded, instrumented, pass-or-fail programme could end the controversy; equally, a government agency spending $10 million on LENR confers a legitimacy that a null result may not fully retract, and the field’s history is of null results that do not stick. Which effect dominates is unknown and is itself a datum this episode will produce.

Speculative How many real anomalies has the “pathological” label wrongly buried? The strongest defence of the sceptical machinery is that the count appears to be near zero: no episode Langmuir named, nor any modern one resolved as pathological, has later been vindicated. But the record is necessarily of the claims that got attention; a real effect dismissed early and never revisited would not appear in it. Handwave Whether the false-burial rate is genuinely zero or merely unobserved is not answerable from the surviving record, and asserting it is zero is where a confident sceptic’s argument works by assertion.

Frontier Can the excess-heat and nuclear-ash channels be reconciled for even one LENR material? This is the live scientific question under the sociological one. Every credible path to closing cold fusion runs through a single sample on which heat and nuclear products are measured together and either agree (a real reaction) or do not (an artifact). No published result has cleared that bar in a way the whole community accepts, and it is what the ARPA-E teams are built to attempt.

Established The first bottleneck is calorimetry, and it is the reason cold fusion did not resolve like N-rays. Measuring excess heat means comparing electrical power in against thermal power out with an error budget small enough to catch a few percent excess. Open cells lose energy to recombination, evaporation, and uneven mixing; closed cells fight heat capacity and calibration drift. A claimed excess a few times the stated uncertainty is a claim about the uncertainty, and disputes over the error model — not over the raw readings — are where the field has lived for three decades. This is Langmuir’s symptom two rendered in watts.

Established The second is materials irreproducibility, and it is structural rather than incidental. Palladium loading with deuterium depends on lot, surface, current history and impurities in ways that were never fully controlled; two electrodes from different batches are effectively different experiments. That is why “it worked in my lab and not yours” is chronic here and was not for the EmDrive, whose device could be copied to drawing. Frontier A field whose positive result cannot be transferred with the material has no way to run the one experiment that would settle it, which is exactly why ARPA-E funds materials characterisation as a first-class deliverable.

Established The third is that the two measurement channels are rarely closed on the same sample. Excess heat is an electrochemistry-and-calorimetry skill; neutron, tritium, gamma and transmutation accounting is a nuclear-instrumentation skill; the two communities and their apparatus seldom sit on one bench. The historical result is heat without commensurate ash and ash without commensurate heat, reported by different groups on different cells. The bottleneck is integration, and it is an instrumentation-and-staffing problem before it is a physics one.

Frontier The fourth bottleneck is blinding, which almost never happens in this class. N-rays fell in a year because Wood could blind the observer in real time; water memory fell because the 1988 Nature team insisted on coded samples. Excess-heat and levitation experiments are rarely run blind, because the experimenter necessarily builds and operates the apparatus. Designing a protocol in which the team measuring the output does not know which cell is “active” is the single cheapest methodological upgrade available and is structurally resisted.

Speculative The fifth is a base-rate bottleneck in the natural history itself: the sample of well-documented episodes is small. Six to a dozen episodes is enough to see a pattern and far too few to fit a resolution-time distribution with any confidence. The subject cannot yet say whether cold fusion is a one-in-fifty outlier or one shape among several, because it does not have fifty clean cases with dated onsets and resolutions. Building that catalogue is a historiographical bottleneck the field has not funded.

Frontier The one hard scientific dependency is a result nobody has produced: a blinded, cross-laboratory excess-heat measurement, on shared material, with the nuclear ash accounted. Everything else in the assessment turns on it. Until such a result exists, the natural history can describe how the other episodes ended but cannot say which column cold fusion belongs in, and the honest position stays “open” rather than “resolved either way.” This dependency is typed below as a missing scientific result, not a fundable certainty.

Frontier The natural history depends on historiography that has not been assembled to a common standard. A predictive account of resolution needs a catalogue of episodes with dated onset, dated resolution, resolution mechanism, and independent classification — the kind of dataset Science of Science builds for other questions and Research Front Detection touches from the bibliometric side. No such catalogue exists for pathological-science episodes specifically, so every generalisation here rests on the same half-dozen much-retold cases.

Established It depends on metrology that mostly exists. Microwatt calorimetry, calibrated neutron and tritium counting, and superconductor magnetometry are mature; the dependency is on applying them together under blinding, not on inventing them. That is why the ARPA-E awards read as an integration and staffing exercise rather than a search for new instruments, and why Metrology Infrastructure is the real upstream brief here.

Frontier And it depends on a settled account of how scientific claims are supposed to change at all. Whether resolution is falsification, contamination-identification, or the slow demographic decline of symptom six is a question Scientific Revolutions and Philosophy of Science own; this brief borrows their apparatus and supplies the anomalous-claim cases they treat only in passing.

Frontier The decisive test is the one ARPA-E was built to run: excess heat above ten times the calorimeter’s measurement uncertainty, reproduced blind by two independent teams working from a common material lot, with a nuclear-ash budget in tritium or neutrons that accounts for the energy. Each clause disarms a specific historical failure. Ten-times-uncertainty defeats the error-model disputes that have consumed the field; blind operation defeats experimenter expectation; a common lot defeats “it worked with my palladium”; the ash budget defeats heat-without-fusion. No result in thirty-four years has met all four at once, and meeting them is what would move cold fusion from the persistent column to the resolved one.

Established The general kill-test template is already validated by the episodes that resolved. Wood’s removal of the aluminium prism from Blondlot’s apparatus while the observer kept reporting the spectrum is the canonical blinded intervention; the 1988 Nature team’s coding of Benveniste’s samples is the same move on a modern claim. Established The complementary template — reproduce the anomaly from the mundane cause — is Rousseau showing that sweat gives polywater’s spectrum and the 2023 groups showing that a Cu₂S transition gives LK-99’s resistivity drop. A claim survives only until one of these two experiments is run competently.

Frontier For LENR specifically, the highest-value single experiment is co-located heat-and-ash on one sample, which the ARPA-E portfolio is designed to deliver. Michigan’s microwatt-resolution calorimetry paired with scintillation and mass spectrometry, and MIT’s and Stanford’s nuclear-product detection, are attempts to close both channels on the same material rather than reconcile results from different cells. Frontier If heat and a commensurate nuclear signature appear together and survive replication, the effect is real; if the heat persists with no ash, it is calorimetric error; either outcome ends the impasse.

Speculative The cheapest experiment on the natural history itself is a catalogue, and it needs no laboratory. Assemble every documented anomalous-claim episode with dated onset, dated resolution, resolution mechanism and an independent pathological-or-real classification, and fit the resolution-time distribution. That converts “cold fusion is an outlier” from an impression into a measured tail probability, and it is a dataset-building exercise, not a discovery.

Frontier A blinding protocol is the second no-new-hardware experiment. Have one group build and run a battery of cells, some deuterium-loaded and some hydrogen controls under a code held by a third party, and have a separate group measure heat and radiation without knowing which is which. Frontier The literature contains almost no blinded excess-heat measurements, so even a null from a properly blinded run would be a genuinely new datum rather than another entry in a contested pile.

Handwave The experiment that would falsify the whole natural-history framing would be a claim that resolved by none of the known mechanisms — neither blinding nor mundane-cause identification — and simply turned out true on its own terms. The framework predicts this essentially never happens; a single clean counterexample would show the sceptical machinery has a false-burial rate above zero. Watching for it is the discipline’s own control experiment, and it runs continuously in every new anomalous claim.

Established The engineering requirement is a bench that closes both channels on one sample under blinding, and each part exists separately. On the heat side: flow or Seebeck calorimetry with microwatt resolution, closed to recombination losses, calibrated against a known Joule source in the same cell geometry. On the nuclear side: calibrated neutron counters, tritium assay of the electrolyte, gamma spectroscopy, and post-mortem mass spectrometry for transmutation ash. Frontier The engineering novelty is integration and shielding, so that the heat measurement and the radiation measurement do not degrade each other, not any new sensor.

Established Materials fabrication with controlled, documented loading is the deliverable that the 1989 work never had. Reproducible deuterium loading of palladium (or the nanocomposites the ARPA-E teams favour) with recorded lot, surface preparation, loading ratio and current history is what makes a common-lot experiment possible. Texas Tech’s and Amphionic’s materials-characterisation deliverables are aimed squarely at this: a sample that can be shipped to a second laboratory and behave the same way.

Frontier The blinding infrastructure is procedural engineering, and it is the cheapest high-value item. A coded set of active and control cells, a code held by an uninvolved party, and a measurement team kept ignorant of assignments requires no new hardware and would transform the evidential status of any positive. It is borrowed directly from clinical-trial practice and from the water-memory investigation, and its near-total absence from the LENR literature is an engineering-culture gap, not a technical barrier.

Speculative For the natural history as a discipline, the engineering is a database and a classification standard. A structured record of episodes — onset, resolution, mechanism, funding, outcome — with a documented coding scheme, is what turns anecdote into base rate. The techniques are those of Science of Science applied to a corpus nobody has assembled: the retractions, the review-panel reports, and the histories, made comparable.

Within this map, the closest neighbours are the other studies of how science behaves rather than what it finds. Scientific Revolutions asks whether conceptual change is punctuated or gradual and finds the punctuated picture unmeasured; this brief supplies the failure cases — the “revolutions” that were announced and did not happen — that any theory of change has to accommodate. Philosophy of Science owns demarcation and confirmation; pathological science is the empirical stress test of demarcation, since these are the claims that look like science, are done by scientists, and are not.

Metrology Infrastructure is the real upstream dependency: nearly every resolution here is a metrology result — a better balance, a cleaner spectrum, a closed error budget. Science of Science and Research Front Detection own the bibliometric instruments that could, in principle, detect symptom six — a supporter community rising toward and falling from a peak — in citation data, though neither has been pointed at this class. Innovation History and Scientific Institutions Through History own the journals, panels and academies that adjudicate.

On the physics side the adjacent briefs are the fields these claims trespass on: Commercial Fusion and Aneutronic Fusion, whose real nuclear signatures are the standard cold fusion has to meet, and History of Electrogravitics, a sibling case of a claim class that trespasses on a conservation law much as the EmDrive does. Outside the map: the history of science, which supplies the cases; scientometrics, which could supply the instruments; and clinical-trial methodology, which supplies the blinding this class refuses to adopt.

Established The institutional fact that distinguishes cold fusion is state re-adjudication, and it has now happened three times in one department. DOE reviewed the claim in 1989 and 2004 through advisory panels and declined to close it; ARPA-E reopened it in 2023 as a funded, instrumented programme. No other episode in the class was ever formally ruled on by a government science agency — N-rays, polywater and the EmDrive died in the literature and in the failure of laboratories to reproduce them. Frontier Whether repeated state review is a mechanism for resolving diffuse claims or for keeping them alive is the open institutional question this brief cannot yet answer.

Established Peer review does not catch this class, and the record is unambiguous. Benveniste’s water-memory paper appeared in Nature in 1988; the OPERA superluminal-neutrino result was a competent collaboration’s announcement in 2011; the STAP stem-cell papers appeared in Nature in 2014; the EmDrive’s thrust ratio was in a peer-reviewed 2016 paper. Review filters incompetence and fraud unevenly and does not filter honest self-deception at the detection limit at all. Established Resolution comes after publication, from replication and re-measurement, which is an argument about where scrutiny should be resourced.

Frontier Funded replication is the institution that actually resolves these, and it is chronically under-provided. Google’s roughly $10 million over three years and ARPA-E’s roughly $10 million across eight teams are unusual precisely because paying skilled groups to try to kill a claim rarely attracts money or credit. The default institution — individual laboratories reproducing at their own expense — is what let cold fusion persist, because a scattered set of unfunded nulls never adds up to a verdict the community will act on.

Established The preprint server has become an adjudicating institution in its own right. LK-99 was raised and settled on arXiv within weeks, faster than any journal cycle, because synthesis recipes and null results circulated openly and globally. Frontier That speed is a genuine institutional change since the polywater era, though it works best for claims that are cheap to attempt and may not compress the timeline for materials as batch-sensitive as loaded palladium.

Frontier The interested parties on this page are named where they matter. The most detailed roster of the ARPA-E awards used here comes from a trade outlet (Green Car Congress) reporting the announcement; the ARPA-E figures are the programme’s own; the 2004 panel breakdown is quoted from independent coverage in Scientific American and Nature; and advocacy archives that champion LENR were deliberately not used for factual claims, because a community that reframes irreproducibility as sensitivity is exactly the interested party this subject warns about.

Established The core ethical asymmetry is between two errors, and the record says one is far commoner than the other. A Type I error — believing a pathological claim — wastes money and careers and, at worst, misdirects a field for years. A Type II error — dismissing a real anomaly as pathological — would be catastrophic for knowledge, and the striking empirical fact is that the class has essentially no confirmed instances of it. Frontier That imbalance justifies a strong sceptical prior, but it does not justify certainty, because the record cannot show the anomalies that were dismissed and never revisited.

Established “Pathological science” is a label that can end inquiry, and using it as a weapon is its own hazard. Applied as Langmuir intended — a diagnosis of a self-deception process — it is a tool. Applied to a person or to a field one dislikes, it becomes a way to make a claim unfundable and unpublishable without doing the experiment that would settle it. Frontier The cold fusion history contains both uses, and separating the legitimate diagnosis from the reputational attack is a live ethical problem for editors and panels.

Frontier There is a real human cost, and it falls on competent people. Blondlot’s career did not survive N-rays; Fleischmann and Pons were destroyed professionally; the lead STAP author’s co-author died by suicide during the aftermath. Honest error at the detection limit is punished as if it were fraud, which both harms individuals and teaches the field to hide anomalies rather than report them. Speculative A culture that could distinguish honest self-deception from misconduct in public would lose fewer good scientists to a single mistake.

Frontier State re-adjudication carries a legitimacy cost that is itself an ethical question. A government agency funding LENR signals to the public that the claim is live, and a null result may not undo that signal; equally, refusing ever to test it can look like suppression to a sceptical public. Established The defensible position is the ARPA-E one — fund a decisive test with a public pass-or-fail bar — precisely because it converts an open-ended legitimacy into a dated verdict.

Frontier The civilisational value of a calibrated natural history is triage: knowing which extraordinary claims to test hard, which to let die in the literature, and roughly how long each will take. Every generation produces its anomalous claims, and the cost of handling them badly is measured in misdirected funding on one side and missed discoveries on the other. A base rate — most resolve in weeks to a decade once someone runs the decisive test; almost none are wrongly buried — is a decision aid a scientific civilisation can actually use.

Frontier The cold fusion case carries a specific lesson about diffuse claims. A claim that is really a family of loosely specified protocols cannot be killed by a single experiment, so it persists regardless of its truth. The civilisational response is to force diffuse claims into decisive form — a shared material, a blinded protocol, a pass-or-fail bar — rather than to argue them indefinitely. Speculative ARPA-E’s 2023 topic is a test of whether an institution can do that on demand, and its outcome will be studied whichever way it falls.

Speculative The accelerating resolution time is the most encouraging civilisational trend on the page. From about a year (N-rays, 1904) to about eleven years (polywater) to weeks (LK-99), the time from claim to resolution has collapsed as replication has globalised and moved to preprints. If that trend holds, the standing cost of anomalous claims falls toward zero for the easy cases, leaving only the diffuse, materials-dependent ones — the cold fusion shape — as the residual problem worth institutional attention.

Handwave The reflexive risk is that the same tools that resolve claims faster also generate and amplify them faster. The preprint-and-social-media loop that killed LK-99 in weeks also spread it worldwide in days, and a future claim that is easy to state and hard to test could ride that loop into a durable public belief the way none of these six did. Speculative Whether the resolution machinery keeps outrunning the amplification machinery is the open civilisational question, and it is not one the historical record can settle.

These horizons track how the natural history and its anchor case are likely to develop, not what any effect will turn out to be:

• 10 yr: Frontier The ARPA-E LENR topic reports out; on its design the most likely outcome is a set of well-instrumented nulls with the excess-heat-and-nuclear-ash question closed for the specific materials tested, which would be the strongest evidence the field has ever produced without necessarily ending the controversy. Speculative Expect at least one new LK-99-class claim to appear and resolve within weeks on preprint servers.
• 25 yr: Speculative If the historiographical catalogue gets built, the subject acquires an actual resolution-time distribution and cold fusion’s outlier status becomes a measured tail probability rather than an impression. Speculative A settled institutional practice — shared materials, blinding, pass-or-fail funded tests — may become the default response to extraordinary claims, or may not, depending on whether ARPA-E’s topic is judged a success.
• 50 yr: Speculative Cold fusion is either resolved — almost certainly as an artifact, on the weight of the record — or it has become a permanent low-level research subject sustained by a minority community, in which case it will be the longest-running case in the class by a wide margin. Handwave The small chance that a real, weak effect in loaded metal hydrides is finally isolated cannot be excluded from this distance.
• 100 / 250+ yr: Handwave Beyond useful forecasting. The one durable datum is structural: as long as instruments have detection limits and observers have expectations, Langmuir’s symptoms will keep recurring, and the natural history will keep acquiring cases. Handwave The vocabulary of “pathological science” is itself about seventy years old and may be replaced, but the phenomenon it names will not be.

• Depends on Metrology Infrastructure, because almost every resolution on this page is a metrology result, and Scientific Revolutions and Philosophy of Science for the account of how scientific claims are supposed to change. No brief on this map produces the specific missing result this one waits on; that is typed below.
• Requires (not on this map) A blinded, cross-laboratory excess-heat result above ten times measurement uncertainty on a common material lot with the nuclear ash accounted — the one result that would move cold fusion from the persistent column to the resolved one, and which no group has produced in thirty-four years. Shared reference-material lots and blinding protocols for anomaly replication, borrowed from clinical-trial practice and almost entirely absent from this literature, without which the common-lot experiment cannot be run. A dated catalogue of anomalous-claim episodes with resolution mechanism and timing, the historiographical dataset that would turn “cold fusion is an outlier” into a measured tail probability. And a standing, funded mechanism for decisive replication of contested claims, of which the ARPA-E topic and the Google effort are one-off instances rather than an institution.
• Enables a triage rule for extraordinary claims — test hard, let die, or re-adjudicate — that funders and editors could apply, and a calibrated prior for Research Front Detection to use in distinguishing a genuine emerging front from a supporter community climbing toward Langmuir’s 50% peak.
• Adjacent Science of Science, Scientific Institutions Through History, Innovation History, and on the physics side Commercial Fusion, Aneutronic Fusion and History of Electrogravitics.

Established “Cold fusion was debunked in 1989.” It was widely judged dead in 1989 when the neutron signature failed to reproduce, but it was never cleanly killed. The Department of Energy reviewed it in 1989 and again in 2004, and in 2004 the panel “split about evenly” on excess heat; ARPA-E funded eight teams to settle it in 2023. Frontier “Debunked” is what most people believe and “unresolved after thirty-four years and three state reviews” is what the record says.

Frontier “Science self-corrects quickly.” Usually, yes — N-rays in a year, LK-99 in weeks — but the speed depends on the claim being sharp enough for one decisive test. A diffuse, materials-dependent claim can resist correction for decades, and cold fusion is the standing counterexample to the comfortable version of the self-correction story. Established Self-correction is a property of the claim’s structure as much as of science’s virtue.

Established “Pathological science means fraud.” It means the opposite: honest, competent scientists deceived by a real signal at the limit of detection. Langmuir coined it precisely to separate self-deception from misconduct, and the six symptoms are diagnostic of a process, not of dishonesty. Frontier Conflating the two is what makes the label a weapon and punishes honest error as if it were fraud.

Established “LK-99 was a hoax.” No. It was a real measurement of a real material whose apparent resistivity drop came from a copper(I) sulfide impurity transition near 104–112 °C and whose “levitation” was ordinary ferromagnetism. Established The authors saw something; they misattributed it. That is the textbook shape of an honest anomalous claim, resolved unusually fast.

Established “The EmDrive might still work.” The 2021 Dresden measurements refuted all EmDrive claims by at least three orders of magnitude and showed the apparent thrust was thermal warping of the balance mount. Established Combined with the conservation-of-momentum prior it violated from the start, the case is as closed as any in this brief.

Frontier “Peer review would have caught these.” It did not: water memory, STAP and the EmDrive thrust ratio all passed peer review at strong venues, and cold fusion’s original claim was announced by press conference and later published. Established Review is not a replication engine; resolution comes after publication, which is an argument about resourcing replication rather than tightening review.

Frontier “State re-adjudication means the government thinks cold fusion is real.” It means the government judged the claim un-closed and worth a decisive, instrumented test — ARPA-E’s framing is to prove or debunk. Handwave Reading a funded kill-or-confirm programme as an endorsement is exactly the legitimacy error the programme’s pass-or-fail design is meant to prevent, though whether a null will fully retract the signal is genuinely uncertain.

Speculative “The sceptics have surely buried real discoveries as pathological.” They may have, but the record shows no confirmed case — no episode resolved as pathological has later been vindicated. Handwave The claim that the false-burial rate is exactly zero cannot be proven from a record that only contains the claims that got attention, so the honest statement is “near zero and unmeasurable at the tail,” not “zero.”