A child sits still for about eight minutes while a flexible pad rests against the side of the neck and shoulder, held there by an elastic band. The pad carries no current, the paper says. Its manufacturer calls it a conveyor, whose job is to channel asymmetrically il debole flusso di correnti radioelettriche — the weak flow of radio-electric currents the machine generates — onto the area being treated. The machine driving it is a BENE Mod 110, built by ASMED Srl in Scandicci, on the edge of Florence, and whoever is operating it cannot change the settings. The protocols are fixed in the device and, in the company's own phrase, non modificabili dall'operatore — not modifiable by the operator.

The child comes back three or four times a day, at least an hour apart, for one to two weeks. Eighteen sessions. Before the first one there is a separate procedure, Neuro Postural Optimization, delivered through a hand-held probe with a gold tip touched to the outer ear. It lasts, the paper says, a few milliseconds.

Thirty-nine children went through this: thirty-one boys, eight girls, average age seven years and ten months. They were the whole of one clinic's autistic caseload across the first seven months of a year the paper does not specify, at a centre the paper does not name. Within a week of the final session their parents filled in a questionnaire they had already filled in once before.

That questionnaire is the study. Everything else is apparatus.

The questionnaire is the Autism Treatment Evaluation Checklist, developed at the Autism Research Institute by Bernard Rimland and Stephen Edelson. It was built so parents could watch for change in their own children: four subscales, a total score, lower meaning less impairment.

The most careful independent look at it I could find came from Iliana Magiati, Joanna Moss, Rachel Yates, Tony Charman and Patricia Howlin in the Journal of Intellectual Disability Research in 2011. They followed twenty-two children over five to six years and reported "preliminary evidence of the ATEC's potential value for monitoring progress ... over time," alongside "considerable variation in the patterns of scores shown by individual children." That is the checklist's strongest endorsement from outside the people who built it, and it is a long way from an instrument that can carry the weight of proving a treatment works.

The Florence paper states that the ATEC "has been validated for ASD research" and cites two studies for it: a longitudinal analysis co-authored by Stephen Edelson, who co-created the checklist, and a 2024 comparison paper. Magiati's evaluation is not in its reference list.

There is a second thing known about parent-rated autism outcomes, and it was published five years before this study. In Molecular Autism in August 2020, Spyridon Siafis and colleagues pooled eighty-six randomised placebo-controlled trials of drugs and dietary supplements in autism, covering 2,360 participants who received placebo. Nineteen per cent of them were rated at least much improved — on placebo. Among the design features associated with a larger placebo response in at least one core symptom domain, the first they list is this: "Caregiver (vs. clinician) ratings." Their recommended remedy appears in the conclusions: "the use of measurements of change not solely dependent on caregivers."

This study has no placebo arm, no control group, no clinician rating, no blinded observer and no neurophysiological measure. Its primary outcome is a checklist completed by parents who knew their child had just been treated.

A parent cannot tell, after a fortnight of sessions, whether a change came from the device, from growing up, from expectancy or from noise. Autism is developmental, fluctuating and heterogeneous, and nothing in this study was built to separate those things.

The results, as printed. Mean total ATEC fell from 67.76 to 56.25, a mean reduction of 11.51 points, p < 0.0001, Cohen's dz of 0.78, with a 95 per cent confidence interval running from 0.43 to 1.11 — thirty-nine children cannot locate an effect precisely in either direction. Twenty-three of the thirty-nine children — 59 per cent — improved by at least eight points, the paper's threshold for a clinically meaningful change. Twelve showed no clinically relevant change.

One comparison is available. Standardised against the spread of the scores the children started with rather than the spread of their changes — a different denominator from the paper's own dz — the shift is 11.51 over 16.11, about 0.71, against the pooled placebo change of −0.36 that Siafis found across the drug trials. Twice as far as inert placebo, then, and the arithmetic flatters the treatment: the ATEC is not among the scales he pooled, his analysis preferred clinician ratings wherever they existed, and caregiver rating headed his list of features associated with a larger placebo response. The caregiver-rated benchmark is higher than 0.36, and nobody knows by how much.

For four children the parent-rated total went the other way by at least eight points. In Table 3 that row is labelled "Clinically significant worsening," with a 95 per cent confidence interval running from 3.3 to 25.0 per cent. Four out of thirty-nine is a number so small that the honest range around it spans nearly an order of magnitude: a study this size cannot measure harm. It can only fail to find it.

The authors disclosed everything.

The conflict-of-interest statement reads, in full: "S.R. and V.F. are co-inventors of patents related to REAC technology, and founders of ASMED Srl, the manufacturer of REAC medical devices. A.R. is the daughter of S.R. and V.F. H.A.B.M. declares no conflict of interest." Salvatore Rinaldi and Vania Fontani invented the technology, hold patents on it, and founded the company that builds the machine. Arianna Rinaldi, the first author, is their daughter. It is all there, on the page, in a journal anyone can read for nothing.

So is the rest. The study was approved by the Institutional Review Board of the Rinaldi Fontani Institute — the authors' own institute — under protocol IRB-RFI-2025-07-1-2, on 1 July 2025. The paper adds that "given its retrospective nature and use of standard clinical practice, additional local IRB approvals were not required." Funding: "This research received no external funding." The article-processing charge that made the paper free to read was paid, the acknowledgements say, by the International Scientific Society of Neuro Psycho Physical Optimization with REAC Technology.

Every relationship a sceptical reader would want to know about is printed where a sceptical reader will find it.

That is the point.

Now the four children.

The study first appeared as a preprint on 18 September 2025. That version's body text reads: "clinically significant worsening (≥ 8point increase) was observed in 4 participants (10.3%)... The subgroup of participants showing worsening did not report adverse behavioral effects but exhibited fluctuations in caregiver perception of symptom severity."

The peer-reviewed version appeared in the Journal of Clinical Medicine on 23 October 2025. There, the same finding reads: "In 10.3% of participants, caregiver ratings suggested an apparent worsening, primarily in the Health/Physical/Behavior domain (Table 3). No objective deterioration was observed, and no adverse effects occurred. This phenomenon was interpreted as a transient disorientation phase in the process of functional reorganization, rather than a true negative outcome."

The discussion elaborates: "most likely a temporary adaptive fluctuation, akin to transient reorganization phenomena documented in post-stroke recovery and neuromodulation contexts ... rather than a genuine worsening."

The preprint's abstract mentioned no worsening at all; it closed on safety, reproducibility and tolerability. The published abstract carries the 10.3 per cent, and carries a limitations sentence the preprint abstract did not have: "the absence of a control group, lack of objective neurophysiological measures, and no long-term follow-up limit causal inference." The discussion had already conceded the control-group problem in September, word for word; what changed between the versions is what a reader who reads only the abstract is told. The paper also acquired in October an effect-size comparison with randomised trials of transcranial direct-current and magnetic stimulation — in September it had compared those methods only on convenience — and with it the concession that its findings "should therefore be interpreted as consistent with, rather than superior to, these emerging neuromodulation methods." Whether a reviewer asked for that or an author volunteered it, I cannot say. The reports are not public.

Review missed smaller things. Table 1 reports the cohort as 31 male (77.5%) and 8 female (22.5%) — percentages of forty, not thirty-nine — directly beneath a caption stating that percentages are relative to a total sample of thirty-nine, and a few lines below body text giving the correct 79.5 and 20.5.

And in the same revision that raised the harm signal, the worsening acquired an adjective. It became apparent. It acquired a mechanism — transient disorientation during functional reorganisation — and three citations supporting the mechanism. The signal was lifted into the abstract, and the solvent for it travelled up alongside — not in the same sentence, but in the next two, so that a reader meeting the harm meets the explanation before the paragraph ends.

Table 3 still says "Clinically significant worsening." Only the sentences changed.

Here is the difficulty, and it requires nobody to have acted in bad faith. The caregiver questionnaire is the study's only evidence that the treatment did anything at all. When those scores fall, the paper reads them as the protocol working. When they rise, the paper reads them as the parents' perception lagging behind the protocol working.

A measure that means the treatment succeeded whichever way it moves is not measuring the treatment.

Walk the chain and ask, at each link, who was positioned to return a different answer.

The device is made by a company founded by two of the authors, which says on its own website that it was born dalla stretta sinergia — out of the close synergy — with the institute where the study was coordinated, and calls itself il braccio operativo, the operating arm that turns the discovery of Drs Rinaldi and Fontani into certified medical devices. Under the study's design and setting, the work was "conducted at a single clinical center under the coordination of the Rinaldi Fontani Institute & Foundation (Florence, Italy)": the coordinating institute is named and placed, the treating centre is neither. Approval came from the coordinating institute's own review board. The outcome measure was completed by parents who had chosen the treatment. Interpretation belonged to the authors. The publication fee was paid by a society named after the technology.

That leaves the prior literature, which is the paper's stated reason for believing the mechanism is real before the study begins: REAC "exerts its effects selectively in tissues and neural circuits where alterations of endogenous bioelectrical activity are present, without interfering with healthy structures," followed by a run of references.

So I counted. In OpenAlex, a phrase search on "radio electric asymmetric conveyer" across titles and abstracts returns eighty-two works. Sixty-four of them sit in one institutional cluster, which the index misnames and which is unmistakably the group that invented the technology. The next-largest cluster is the University of Sassari at twenty-seven, and the University of Florence appears separately with ten — figures that cannot be added up, because a paper with authors at two institutions is counted under both, which is why sixty-three institution groups sum to 220 entries across eighty-two works. And the five references the paper cites for that claim about selectivity, [23] to [27], all carry Rinaldi, Fontani, or both, as authors.

That is one index and one phrase, and I have given the query so it can be checked. And it matters whose previous studies "have demonstrated improvements in cognitive, behavioral, and emotional regulation domains".

ASMED's website has a link for scientific publications. Follow it and you are told you are leaving ASMED for the website of the Rinaldi Fontani Institute, and then: Sezione in costruzione. The scientific work behind REAC Technology will be added soon, the notice says; in the meantime you can consult the studies published on PubMed directly.

The company's evidence page sends you to the literature. Most of the literature is the company's.

Elsewhere on the same site, under corporate values, ASMED states that every therapeutic application is supported by controlled clinical studies and peer-reviewed publications attesting to efficacy and safety. The study discussed here is uncontrolled, and says so in its own abstract.

Under "Certifications," ASMED lists one thing: ISO 13485:2016, behind a link marked visualizza. It is a real and demanding standard, and it certifies process — how the devices are designed, documented and audited — not whether a treatment works. I could not establish the device's registration record through either the European database or the Italian one, which establishes nothing either way about its class or its conformity route.

The strange thing about disclosure is that we have known for twenty years it can make matters worse. Daylian Cain, George Loewenstein and Don Moore titled their 2011 paper in the Journal of Consumer Research "When Sunlight Fails to Disinfect": disclosure, the abstract says, "can backfire, hurting those whom it is intended to protect." Advisers who had declared an interest felt licensed to lean on it and exaggerated more; the people they advised failed to discount it nearly enough.

The remedy named in their fourth study is "explicitly and simultaneously contrasting biased advice with unbiased advice." There is no unbiased advice to contrast with here. A parent reading the disclosure statement is not choosing between an interested account and a disinterested one. There is the interested account, and a note explaining why it is interested.

Underneath the psychology there is something harder. Quality here stays hidden from the buyer even after the thing has been bought. Siafis does not merely describe that condition; he measures it: nineteen per cent of children given an inert pill were rated at least much improved by the people who knew them best.

In a market like that the scarce good is not the treatment. It is a disinterested evaluator, and there is not one anywhere in this chain — not in the clinic, not on the review board, not in the literature the paper leans on to justify its mechanism before the study starts. Where no disinterested evaluator exists, the signal that gets produced is the one that is cheapest to produce: a single-arm, retrospective, caregiver-rated series in a journal the authors pay to appear in, added to a literature of eighty-two papers, sixty-four of which carry the inventors' affiliation. Disclosure does not put a second observer in the room, and a second observer was the only thing that would have helped.

Now the case on the other side. Nobody has made it publicly, so it is mine rather than theirs.

Every one of those relationships exists because there is nobody else. Public money was never realistically on the table for a proprietary radiofrequency protocol devised by two clinicians in a private Florentine institute, and I have found nothing to suggest anyone applied for it. The realistic alternative to a study run by the inventors was probably not a study run by somebody else; it was thirty-nine children treated and no published record at all. On that reading, disclosure is not an alibi. It is the whole of what they had to give, and they gave all of it, including four children who got worse, in an abstract that a preprint reader would never have seen.

And the result might be real. A caregiver-rated improvement of about 0.71 baseline standard deviations, in a condition where nineteen per cent of children are rated much improved on an inert pill, is entirely consistent with a treatment that does something. I have no evidence that REAC does nothing, and nothing here should be read as claiming it. The smallness cuts both ways: a study that cannot measure harm cannot demonstrate benefit either. My claim is the narrower one — that this paper cannot tell anybody which it is.

That is the case as I would put it for them. They have not put it. Grant every word of it: that nobody else would have done this work explains why the inventors did it. It does not explain why the work was done this way.

The missing ingredients were never expensive. A blinded second rater for thirty-nine children is one to two days of a psychologist's time. A comparison group drawn from the clinic's own waiting list defers revenue rather than destroying it; the checklist is free and parents fill it in at home. Pre-registering the design, reporting the four worsened children individually rather than as one row, and testing baseline severity against change to rule out regression to the mean cost nothing at all. The paper cites no analysis of placebo response in autism trials, though its discussion concedes the ground anyway — "placebo effects or regression to the mean cannot be entirely excluded" — before arguing that "several considerations reduce the likelihood that the observed changes were solely due to placebo."

The retrospective frame is the point here, and it is not a technicality. A study assembled after the fact cannot have a control arm, because nobody scored the untreated children; it cannot have a second rater, because the weeks in question are gone. Choosing to look backwards is what puts the cheap safeguards out of reach, and it is the one choice that had to be made before any data existed. The authors' own conclusion says prospective randomised trials would be "both feasible and ethically justifiable." That is the strongest sentence in the paper.

A single-arm, caregiver-rated design does not save money. What it removes is the possibility of a disappointing result, and that was the one setting nobody could modify. Improvement confirms the protocol. No change is an unresponsive subgroup. Deterioration is transient reorganisation. There is no outcome this design can produce that would count as evidence against the thing being studied, and a procedure that cannot fail is not a test.

None of the people who understand this market best are talking about REAC. They are describing the conditions a study like this one enters.

Thomas VanCott paid $9,000 for a course of magnetic stimulation for his son Jake, who is eighteen and minimally verbal, and told the Los Angeles Times in 2024: "It's too much for most things, but not for the potential of my child speaking." On the doctor who sold it, he "seemed pretty confident. And his confidence gave me confidence." On the outcome: "It just did nothing." On the decision: "When you're a parent of a child and you think that this can help, it's like, FDA be damned, right?"

Zoe Gross of the Autistic Self Advocacy Network, in the same article, on the pressure families are put under: "People will be saying things like, 'Time's ticking, your kid's missing milestones ... you have to fix it now.'"

A mother quoted by the Guardian in June 2026, who asked not to be named, had already spent $20,000 out of pocket on her three-year-old son. "The lack of government assistance for children with autism is truly mind blowing," she said. On what is available instead: "We were encouraged to put my child in a 40-hour-a-week facility for his autism, yet we couldn't even get insurance to approve two hours of occupational therapy."

That is the demand side. It is arithmetic done by people with no good options.

Anna Wexler, an assistant professor at the University of Pennsylvania who studies the ethics of emerging technologies, was talking about magnetic stimulation, but the test is general: "If someone opts for an experimental therapy, that in itself is not problematic. What is problematic is if they are making that decision based on erroneous or incorrect beliefs about efficacy."

The paper is where that belief is formed. I could not establish what a course of this treatment costs a family, or how it is sold; the institute publishes no prices. What is establishable is what a deciding parent would read: an open-access, peer-reviewed clinical paper, in a journal with a named academic editor and a visible revision history, reporting a large effect size and clinically meaningful improvement in 59 per cent of treated children after eighteen sessions of about eight minutes each — a little under two and a half hours of treatment in all.

The four children are still in the table. Their parents reported them at least eight points worse after eighteen sessions. In the row that records this they are "clinically significant worsening"; in the sentences that discuss them they are "apparent."

Nobody outside the institute can look at them again. The paper's data availability statement says that all data supporting the findings are contained within the manuscript, and what is contained within the manuscript is that row.

Everything was disclosed. A parent who reads every word of it is still left with one account of what happened to these thirty-nine children, written by the only people in a position to report that it had not worked.

Sources

  • Rinaldi, A.; Benetti Mota, H.A.; Rinaldi, S.; Fontani, V. "Targeted Endogenous Bioelectric Modulation in Autism Spectrum Disorder: Real-World Clinical Outcomes of the REAC BWO Neurodevelopment–Autism Protocol." Journal of Clinical Medicine 2025, 14(21), 7500. DOI 10.3390/jcm14217500. Received 4 September 2025, revised 7 October, accepted 20 October, published 23 October 2025.
  • The same study as a preprint: Preprints.org, posted 18 September 2025, DOI 10.20944/preprints202509.1579.v1.
  • Siafis, S. et al. "Placebo response in pharmacological and dietary supplement trials of autism spectrum disorder (ASD): systematic review and meta-regression analysis." Molecular Autism 11:66, 26 August 2020. DOI 10.1186/s13229-020-00372-z.
  • Magiati, I.; Moss, J. (Joanna); Yates, R. (Rachel); Charman, T.; Howlin, P. "Is the Autism Treatment Evaluation Checklist a useful tool for monitoring progress in children with autism spectrum disorders?" Journal of Intellectual Disability Research, 4 January 2011. DOI 10.1111/j.1365-2788.2010.01359.x.
  • Cain, D.M.; Loewenstein, G.; Moore, D.A. "When Sunlight Fails to Disinfect: Understanding the Perverse Effects of Disclosing Conflicts of Interest." Journal of Consumer Research, 2011 (published online 27 August 2010).
  • Purtill, C. "Can MERT help kids with autism? There's little evidence." Los Angeles Times, 5 September 2024 — source of the quotations from Thomas VanCott, Anna Wexler and Zoe Gross.
  • "'Autistic kids are being experimented on': inside America's booming market for unproven stem cell infusions." The Guardian, 12 June 2026.
  • ASMED Srl, corporate and device pages (asmed.net): "Dispositivi medici," "Azienda," "Certificazioni." Captured 12 September 2026.
  • Istituto Rinaldi Fontani (irf.it), homepage. Captured 12 September 2026.
  • OpenAlex, works filtered on title_and_abstract.search:"radio electric asymmetric conveyer", grouped by authorships.institutions.id. Queried 12 September 2026: 82 works, 63 institution groups, 220 entries. The largest group, with sixty-four works, is labelled "Deutsches Historisches Institut Rom" — the German Historical Institute in Rome, which does not publish radiofrequency medicine. It is a disambiguation error, and the index repeats it in its own record for this paper, where the first author's Florence affiliation appears under the Rome history institute's name.

No person quoted in this article spoke to me. Every quotation is transcribed from the published source named beside it; the Italian passages are quoted in the original and translated by me. The passage presented as the strongest case for the other side is my own construction and is attributed to no one.