Cold fusion is the claimed nuclear reaction at or near room temperature announced by Martin Fleischmann and Stanley Pons at the University of Utah in March 1989: their heavy-water electrolysis cell with a palladium cathode appeared to produce “excess heat” far beyond what chemistry could supply, which they attributed to a nuclear process. The claim promised clean, effectively abundant energy, triggered a worldwide replication frenzy within weeks, was officially rejected within months — and then refused to die for 35 years, until a privately funded decade-long program called Project Callisto ran every credible variant of it to ground between 2015 and 2024 and published a definitive null result in September 2026.¹²

The 1989 claim and the first verdict

The motivating physics was a naïve reading of the Nernst equation: because the pressure term sits inside a logarithm, a small voltage applied to a gas-evolving electrode reaction implies enormous gas pressures — read literally, half a volt implies pressures beyond the center of the sun. Fleischmann and Pons hoped electrolysis would load deuterium into palladium at pressures extreme enough to fuse it. Their cell appeared to produce copious heat; they also claimed neutrons, gamma rays, and other nuclear signatures that other labs could not reproduce.³

The announcement came by press conference rather than peer review, and replication attempts began within days. Most failed to find the effect. The US Department of Energy convened a 22-scientist Cold Fusion Panel under the Energy Research Advisory Board, whose interim (August 1989) and final (November 1989, DOE/S-0073) reports concluded that the excess-heat results reported to date “do not present convincing evidence that useful sources of energy will result from the phenomena attributed to cold fusion,” that no convincing evidence associated the anomalous heat with a nuclear process, and that no special programs or research centers were justified — while staying sympathetic to modest support for carefully focused experiments within the normal funding system.⁴

The afterlife: LENR as counter-consensus

The mainstream verdict did not take. A number of respectable researchers — Robert Huggins at Stanford, John Bockris at Texas A&M, groups at national laboratories — reported their own excess heat, and a self-contained research counter-culture formed: its own journal (the Journal of Condensed Matter Nuclear Science), its own international conference series, a rebranding as “low-energy nuclear reactions” (LENR), and eventually more than 300 published claims of positive replication. Well-funded efforts followed from Toyota, NASA, Google, SRI, the US Navy, and the US Department of Energy, with new projects starting at Stanford and MIT as late as the mid-2020s; cumulative spending ran to hundreds of millions of dollars. The cultural tone shifted accordingly — from 1980s scientific blooper to J. Storrs Hall’s Where’s My Flying Car position that the many excess-heat reports are “a solid hint that there’s something in the physics that we don’t yet understand.”⁵

Project Callisto (2015–2024)

Project Callisto was a privately funded effort to settle the question experimentally. More than 100 scientists, engineers, technicians, machinists, and operations staff contributed over nine years — several anonymously even now, a measure of the stigma the topic carried — working in 50,000 square feet of lab space. Its two campaigns attacked the claim from both directions the literature proposed.⁶⁷

Electrochemical replication. Callisto performed more than 40 formal and modified replications of the Fleischmann–Pons experiment — prolonged electrolysis, cutting-fluid additives, oxygen bubbling, high-input-power operation — plus replications of the canonical later claims by McKubre, Staker, Hubler, Storms, Miles, Boss, Barham, Energetics, and Letts, in some cases sourcing vintage components through correspondence with the original researchers. Across the program they investigated more than 500 samples and accumulated more than 200,000 sample-hours of calorimetry, a scale comparable to all prior LENR efforts combined. In a 60-day Staker-like run, palladium loading stayed above 80% and the steady-state heat-output-to-input ratio stayed near one.⁸⁹

Ultrapressure gas loading and stimulation. Because high deuterium loading is the proposed necessary condition, Callisto built pressure machinery far beyond standard capability: 150,000 psi as a routine environment and advanced cells reaching 600,000 psi, achieving full (stoichiometric) palladium-deuteride loading cryogenically at 150 ksi and −100 °C, and sustaining the 450+ ksi that room-temperature full loading requires. They then tried to trigger reactions in fully loaded material: femtosecond laser pulses, heating to melting, intense electric fields, and microfabricated stimulation-and-sensing chips (the NanoTI campaign: 448+ devices, 1.5 million+ pulses, including operation at and above 450,000 psi).¹⁰

Result: null, everywhere. No excess heat, no nuclear signatures, in any regime searched. What distinguishes Callisto from prior null results is what it did with the apparent positives: the team repeatedly observed signals that initially looked like excess heat, and traced every one to a conventional cause — electrical-power accounting errors (a 1988-vintage power supply indicating 3A at 1.8V while actually delivering 3.12A), changing calorimeter response, the different reaction entropies of hydriding versus hydrogen evolution, faulty sensors, and stored chemical energy released when evolved hydrogen and oxygen recombined. A suspected deuterium enhancement in the laser-stimulation campaign turned out to be an optical-response difference between hydrogen- and deuterium-loaded samples, not heat. Their literature analysis concludes that most if not all reported excess-heat events trace back to this catalog of issues.¹¹¹²

The reports (dated February 2024 and June 2025) were publicly released on September 8, 2026; the release notes thank four people, including Orca Sciences’ Ian McKay, “for pushing the publication of this report long after the project concluded.”¹³

The Langmuir frame

In December 1953, Nobel laureate Irving Langmuir gave a colloquium at GE’s Knolls Research Laboratory on what he called “pathological science” — “the science of things that aren’t so.” Transcribed by R. N. Hall from a recovered recording, the talk anatomized N-rays, the Davis-Barnes effect, the Allison effect, mitogenetic rays, ESP, and flying saucers, and extracted six characteristic symptoms:¹⁴

1. The maximum effect is produced by a causative agent of barely detectable intensity, and the effect’s magnitude is largely independent of the cause’s intensity.
2. The effect remains close to the limit of detectability; many measurements are needed because of very low statistical significance.
3. Claims of great accuracy.
4. Fantastic theories contrary to experience.
5. Criticisms are met by ad hoc excuses devised on the spur of the moment.
6. The ratio of supporters to critics rises to somewhere near 50%, then falls gradually to oblivion.

Cold fusion maps onto nearly the whole list: a watt-scale effect at the edge of calorimetric detectability arising from an invisible lattice process (1, 2, 4), precision claims from calorimetry that Callisto showed was routinely mis-accounted (3), and a four-decade supply of ad hoc explanations for each null result (5). Where the episode departs from Langmuir’s template is symptom 6: the supporter ratio never fell to oblivion, because the counter-consensus built its own journals and conferences and stopped depending on mainstream critics. Langmuir’s deeper point fits exactly, though — his cases involved no dishonesty, only scientists “tricked into false results by a lack of understanding about what human beings can do to themselves in the way of being led astray by subjective effects, wishful thinking or threshold interactions.” Callisto’s own verdict is nearly the same sentence with 2026 instrumentation: every claimed positive result was most likely “incomplete understanding of an experimental apparatus, incorrect data interpretation, cherry-picking of data, or in the worst case, deception.”¹⁵¹⁶

The sociology, per a participant

Ian McKay, who worked on Callisto in 2016–2017 as a graduate student, draws three lessons. First, the moonshot hazard: a tiny probability times a civilizational-scale prize produces extreme and unresolvable disagreements about what to fund, and staffs projects badly — a funder optimizing portfolio expected value rationally buys double-or-nothing coin tosses, but a scientist’s career dies on the first failed toss, so moonshots either can’t hire or get subverted into professional wins and tail off into jobs programs (his examples: ITER, JCAP). Second, the conspiracy shape: cold fusion is huge if true, honestly believable through measurement error, and practically impossible to conclusively refute — and the mainstream stigma was itself an epistemic cost, forcing anonymity on researchers and pushing in-house work that outsiders could have checked. “Greed is a great reason to keep a project secret. Embarrassment is not.” Third, what survives contact: Fleischmann–Pons-style work should now be closed, but adjacent claims are live — DARPA’s MARRS program on lattice-confinement fusion (one electron-beam group has claimed fusion-like signals; McKay suspects a prosaic explanation) and muon-catalyzed fusion, which is real physics with a bad energy balance that startups are trying to fix. Eli Dourado’s proposed evidentiary bar for future claims: no attention without non-calorimetric signatures — alpha particles, helium production — since calorimetry alone has now been shown to be uninformative.¹⁷

Connections

• llm-replication-economics — the two ends of the audit-scale spectrum. Schwartz–Andrews–Shapiro audits 4,452 economics papers at machine speed and flags discrepancies; Callisto spent nine years and 200,000 sample-hours auditing one 1989 anomaly by hand. Both treat “why did this number appear” — not just “does it reproduce” — as the object of the audit, and both leave behind a reusable catalog of failure modes.
• adversarial-collaboration — the dispute that never had one. Kahneman’s protocol binds rival camps to shared data and an arbiter; cold fusion’s two camps never shared anything, and the skeptics’ verdict (ERAB 1989) was simply rejected by the counter-consensus, which built parallel institutions instead. Callisto closed the question the other way: not by agreement but by out-engineering the claim until no parameter space was left.
• scientific-idea-diffusion-decline — the insulation case. A claim that never diffused outward and never died: 35 years of survival inside a self-contained venue set, with its own journals, conferences, and funders. Diffusion failure as protection from falsification.
• tabbys-star — the mirror image. Boyajian’s star is an anomaly that diffused widely, attracted mainstream effort, and resolved into prosaic explanations within a few years; cold fusion is the same story with the mainstream exit blocked, stretching the resolution to 37 years.
• superposition-and-polysemanticity — instrument artifacts in both directions. Callisto’s catalog of apparent positives traced to power-accounting errors, calorimeter response drift, and recombination heat — false positives generated by the apparatus. The interpretability “misreading” story runs the mirror image: apparent polysemanticity attributed to a poorly tuned microscope, false negatives about the object. In both cases the surprising signal turned out to be a claim about the measuring instrument.

Sources

• 2026 — Project Callisto
• 2026 — Callisto Electrochemical Loading Research Final Report, Chapter 1: Introduction
• 1989 — Cold Fusion Research: A Report of the Energy Research Advisory Board to the United States Department of Energy (DOE/S-0073)
• Irving Langmuir; transcribed and edited by R. N. Hall 1953 — Pathological Science (colloquium transcript, GE report 68-C-035)
• Ian McKay 2026 — Real cold fusion has never been tried

Footnotes

1. 2026 — Project Callisto ↩

2. Ian McKay 2026 — Real cold fusion has never been tried ↩

3. Ian McKay 2026 — Real cold fusion has never been tried ↩

4. 1989 — Cold Fusion Research: A Report of the Energy Research Advisory Board to the United States Department of Energy (DOE/S-0073) ↩

5. Ian McKay 2026 — Real cold fusion has never been tried ↩

6. 2026 — Project Callisto ↩

7. Ian McKay 2026 — Real cold fusion has never been tried ↩

8. 2026 — Project Callisto ↩

9. 2026 — Callisto Electrochemical Loading Research Final Report, Chapter 1: Introduction ↩

10. 2026 — Project Callisto ↩

11. 2026 — Project Callisto ↩

12. 2026 — Callisto Electrochemical Loading Research Final Report, Chapter 1: Introduction ↩

13. 2026 — Project Callisto ↩

14. Irving Langmuir; transcribed and edited by R. N. Hall 1953 — Pathological Science (colloquium transcript, GE report 68-C-035) ↩

15. Irving Langmuir; transcribed and edited by R. N. Hall 1953 — Pathological Science (colloquium transcript, GE report 68-C-035) ↩

16. 2026 — Callisto Electrochemical Loading Research Final Report, Chapter 1: Introduction ↩

17. Ian McKay 2026 — Real cold fusion has never been tried ↩