Psi Research Bulletin · 25 September 2026
Psi research, metascience and changes to OSF
Experimental findings, replication, fraud prevention and the OSF transition: the 25 September 2026 bulletin, including selected studies published earlier this year.
This issue brings together recent findings in parapsychology, metascience and changes to research infrastructure, alongside anomalistic studies and academic books. September announcements are accompanied by a clearly dated retrospective selection from the main journal issues published in 2026.
Key points
- Radin and Cline report a small signal in their human-AI retrocausal experiment. The late registration of the analysis plan and the treatment of dependencies limit its confirmatory value.
- Wahbeh et al.’s analysis of more than 25 million trials shows how optional stopping can shape apparent performance on psi platforms.
- OSF is retaining preregistrations while progressively withdrawing its active project workspaces. A workflow combining a registry, a durable repository and a working environment is becoming necessary.
- SCORE distinguishes two problems: reproducing an analysis using the original data and replicating a finding with new data. Both checks remain essential.
- James E. Kennedy locates fraud prevention within experimental design. Mindfield, the psychology of belief and new books complete this issue.
Psi research
Physics Essays
Volume 39, issue 3, published online on 4 September 2026. Published article
Human observation and previously recorded optical data
Dean Radin and Ron Cline use previously recorded photometric data from the two first-order spots of a diffraction grating. One of the two signals drives an animated flame during ten intention phases and ten relaxation phases. The study analyses 1,000 human sessions, contributed by 295 people, and 1,000 sessions involving a Claude agent. For the displayed channel in the human condition, the intention-relaxation contrast in the regression residual is d = -0.030, p = 0.0016 by permutation. The non-displayed channel yields p = 0.113, and the displayed channel in the AI condition p = 0.785.
The plan was registered after data collection had begun. The article distinguishes 817 human sessions preceding preregistration and a subsequent window extended to 308 sessions, partly exploratory. The latter yields d = -0.035, with a 95% confidence interval of [-0.079; +0.011]. A similar effect size does not retrospectively turn the earlier observations into a prospective test. The claimed equivalence relies in particular on a margin of ±0.10 d units, exceeding the effect being investigated.
The main analysis uses 10,000 pairs of blocks nested within 1,000 sessions from 295 participants. Permutation at the block level requires justification in view of dependencies between blocks, sessions and participants. The article alone does not demonstrate that an analysis accounting for all these levels would retain the same uncertainty. A replication should specify this statistical structure before data collection.
The factorial figure presents eight cell-wise tests. The p = 0.00007 reported in the abstract corresponds to a parametric slope; the intention-relaxation contrast by permutation yields p = 0.0016. Significance in one cell and its absence in others do not demonstrate a difference between cells: the claimed specificity of human observer × displayed sensor × intention requires a direct interaction test.
Finally, the photodetectors have already absorbed the photons, produced an electrical signal and recorded numerical values. At most, the protocol tests a retrospective correlation between later human observation and earlier optical time series. It does not directly measure wavefunction collapse and does not distinguish the von Neumann-Wigner framework from other interpretations. The article is a priority read because of its original design and reported signal, with low to intermediate evidential strength for an anomalous association and low evidential strength for the physical interpretation.
Primary source : Radin and Cline 2026
Preregistration : OSF c8p2s
Données and code : Zenodo
Journal of Anomalous Experience and Cognition
Volume 6, issue 1, published on 2 April 2026. Official contents
Retrospective selection: 2026
Memory Within Nature
Georgia Black, Bethany Butzer and Rupert Sheldrake test predictions of morphic resonance in three studies using Wordle. The first fails to find the predicted negative correlation. The second finds an improvement in solutions achieved in one or two attempts between morning and evening. The third does not replicate that finding.
The series is informative because of its internal inconsistency. It provides no robust support for the hypothesis and indicates that a new protocol should specify the temporal unit, primary contrast, exclusions and stopping rule in advance.
Primary source : Black, Butzer and Sheldrake 2026
Ian Stevenson’s Rabih Elawar at 60
James G. Matlock, Khattar Hatoum and Carole Louie re-interview, at ages 58-60, a subject studied by Ian Stevenson during childhood. Spontaneous memories are compared with published reports and field notes. The authors then present the subject with some details he had not spontaneously recalled.
The longitudinal concordance is notable. Family contacts, accumulated autobiographical cues and the cued-recall phase nevertheless preclude treating this follow-up as an independent test of the origin of the memories. The article primarily contributes to the study of autobiographical memory in reincarnation-type cases.
Primary source : Matlock, Hatoum and Louie 2026
Sensing Accuracy
Debra Lynne Katz, Jenifer Prather and Jimmy Akin conduct a phenomenological survey of 122 remote viewers recruited through snowball sampling. Of these, 89% report having experienced a feeling that a specific element was correct. Reported markers include suddenness, persistence, vividness and the unexpected nature of the impression.
The study measures neither the objective accuracy of these impressions nor their calibration. It provides a useful taxonomy for designing a prospective study in which confidence judgements would be collected on each trial before feedback and then compared with objective scores.
Primary source : Katz, Prather and Akin 2026
Questions about Questionable Research Practices
Lance Storm responds to analyses of questionable research practices applied to the ganzfeld and defends the stability of the effect after the exclusion of certain studies. This is an analytical commentary, not a new replication. Its strength depends on the comparability of exclusion rules and coding choices, and on access to the data used in the competing analyses.
Primary source : Storm 2026
A Look for a Presentiment Model’s Reasonable Parameters
Ephraim Yeruham Levin proposes a model of presentiment based on decoherent histories and time symmetry. The paper adds no discriminating empirical test. Its relevance will depend on quantitative predictions capable of distinguishing the model from ordinary physiological, temporal and analytical explanations.
Primary source : Levin 2026
Journal of Scientific Exploration
Volume 40, issue 2, published on 7 July 2026. Official contents
Retrospective selection: 2026
Psychic Hacking
Scotch Wichmann reports an exploratory online study involving 146 participants and 584 free-response sessions. Three judges assess each session. Targets include images, videos, PINs and passphrases. Significant results are reported for images, videos and some digit matches.
The computational controls are original. Evidential strength remains limited by the exploratory status, multiple outcomes, dependencies between sessions and participants, and the role of human judgement. The proposed implications for cybersecurity go beyond what the study design supports. The priority is an independent preregistered replication with a single target type, a primary outcome and a fully specified judging procedure.
Primary source : Wichmann 2026
The Myth of Publication Bias in Psi Research
Patrizio Tressoldi and Lance Storm compare 165 published studies, comprising 243 ESP experiments, with 40 preregistered confirmatory experiments from the Koestler Parapsychology Unit registry and with general psychology datasets. They find more published null results in parapsychology.
This finding documents a potentially less selective publication culture. The proportion of p values above 0.05 is nevertheless not a direct estimator of publication bias. It also depends on power, effect sizes, heterogeneity and study designs. The article weakens the idea of a literature devoid of null results, without establishing that psi meta-analyses are free from selection.
Primary source : Tressoldi and Storm 2026
Debate on the decline effect
Dick J. Bierman and James Spottiswoode find a decline in seven out of ten protocol families and a pronounced relationship between effect size and sample size. Tressoldi and Storm respond that the phenomenon is not general: two out of six anomalous cognition protocols would show a decline, compared with four out of four psychokinesis protocols.
The disagreement arises partly from the grouping of protocols, the historical duration of the series and the correlation between year and power. The results do not yet warrant a psi mechanism of decline. A preregistered hierarchical reanalysis, with protocol as a level and explicit dependencies, would provide stronger evidence.
Primary source : Bierman and Spottiswoode 2026
Primary source : Response by Tressoldi and Storm 2026
Zeitschrift für Anomalistik
Volume 26, issue 1, 2026. Contents and full texts
Retrospective selection: 2026
Probing Top Performers in a Forced Choice Clairvoyant Task
Helané Wahbeh, Michael Kriegsman, Beth Glick, Arnaud Delorme and Dean Radin analyse more than 25 million trials from an online clairvoyance task. The protocol is preregistered. Fixed-length series of 5, 10, 25 or 100 trials remain at chance level. No user meets the preregistered false discovery rate (FDR) criterion for top performers.
Irregularities in freely terminated series are largely reproduced by a Monte Carlo simulation fitted to the empirical distribution of stopping points. This negative finding has substantial methodological value: open platforms must model optional stopping before interpreting patterns of performance. Moderator analyses conducted after failure to meet the primary criterion remain exploratory.
Primary source : Wahbeh and al. 2026
Colour and food erasure paradigms
In the unconfounded colour study, Markus A. Maier and colleagues initially observe an effect, BF10 = 39.77, followed by a decline during the study. A permutation analysis assigns a probability below 0.03 to the joint pattern of an effect followed by a decline.
In four food studies totalling 647 participants, the authors report pronounced effects in designs where restricted sets of stimuli are deliberately linked to conditions. Permutation and multilevel analyses reduce some explanations based on stimulus allocation. They do not fully remove the problem of causal identification created by deliberate confounding.
The pattern of an effect followed by decline is not a signature unique to generalised quantum theory. Other processes, including regression, temporal instability, analytical choices and stimulus-by-order interactions, can produce similar structures.
Primary source : Colour study
Primary source : Food studies
Primary source : Authors’ response to commentaries
Development of a Dual Mode Application for Psi Research
Brian Laythe and James Houran evaluate the Immersive Psi Test in a pilot involving 126 participants. No explicit effect above chance is found. A very small effect appears in an implicit narrative condition, p = 0.016 and h = 0.09.
The work primarily concerns the instrument and experimental engagement. Multiple outcomes, convenience sampling and the small effect size require a preregistered replication before any confirmatory interpretation.
Primary source : Laythe and Houran 2026
Mindfield and interdisciplinary literature monitoring
Mindfield 18(1), published on 19 June 2026, addresses relations between psi, media and the cultural imagination. Of particular interest is the bibliography by Maurice van Luijtelaar and Renaud Evrard, Articles Relevant to Parapsychology in Journals of Various Fields (XLVIII), announced as containing 262 articles from 216 journals. It provides an entry point to publications dispersed beyond specialist journals.
The contents include Gerhard Mayer’s discussion of media and AI, analyses of fictional representations and an anthropological approach to experiences in the Philippines. These are primarily cultural and conceptual contributions. They shed light on cultural representations and uses of psi; they provide no new experimental tests.
Mindfield 18(1) • Public contents, 19 June 2026
Metascience and research infrastructure
OSF retains registrations while phasing out active projects
On 11 August 2026, the Center for Open Science announced a reduction in OSF functionality. Creation of new projects and components will end on 16 November 2026. After 19 February 2027, public and private projects are scheduled to become read-only. Existing links and DOIs will continue to work; COS states that it envisages subsequently deleting private projects, with twelve months’ notice.
Registrations and preregistrations are not being removed. The official documentation states that registration content, including drafts and embargoed registrations, is unaffected by this transition. Preprints and content discovery are also retained under this announcement. A modifiable collaborative workspace must therefore be distinguished from a protocol registration. Closing the former does not imply the disappearance of the latter.
For a psi laboratory, the priority is to ensure continuity of working files, scripts and links to data. Exports should preserve the dated protocol, amendments, the variable dictionary and the mapping between files and analyses. A persistent existing OSF link remains useful for documenting the historical state of a project.
OSF Support • Transition details, updated on 3 September 2026
A suitable alternative for each function
Zenodo is a relevant option for archiving the data, materials and code versions associated with an article. A deposit receives a DOI and can link multiple versions. Citing the DOI of a specific version helps identify the files actually analysed. Preserving bytes does not, however, guarantee that software will remain executable, that variables will be clear or that an analysis is correct: the software environment and reproduction instructions should be included.
| Need | Suitable solution | Limitation to bear in mind |
|---|---|---|
| Prospective protocol | OSF Registrations, which is being retained | A data DOI does not replace a plan registered before analysis. |
| Archiving results | Zenodo and a version-specific DOI | Archiving also requires metadata, documentation and licences. |
| Code under development | GitHub, with versions archived on Zenodo | The working repository changes; also cite the archived version. |
This is a recommended workflow: retain the registry for prospective commitments, use an appropriate development environment and archive cited versions on Zenodo. It addresses the announced changes without imposing unnecessary migration of preregistrations. Data that cannot be made public should remain in a suitable institutional environment, with public metadata where possible.
SCORE measures replication using independent data
Tyner et al., Nature, 1 April 2026. This retrospective selection concerns a major replication study in the social and behavioural sciences. Of 274 claims, 151 yield a statistically significant finding in the original direction: 55.1%, 95% CI [49.2; 60.9]. The result weighted at the level of the 164 articles is 49.3%. These percentages answer different questions, since multiple claims may come from the same article.
The replication rate is neither the proportion of true theories nor a benchmark to apply mechanically to parapsychology. It depends on study selection, power and the success criterion. For psi, the main contribution is the need to specify what counts as replication: effect direction, a compatible interval, a minimum effect of interest or a confirmatory test. Compatibility between two estimates is not, by itself, sufficient to establish a non-zero effect.
SCORE also reveals obstacles to reproducing analyses
Miske et al., Nature, 1 April 2026. Of 600 articles for which data were requested, the authors of 144 provided data; another 38 datasets were reconstructed from accessible sources. The team assessed 143 of the 182 available datasets. After weighting the 551 claims by their number within each article, 53.6% of the assessed articles are judged precisely reproducible and 73.5% at least approximately reproducible.
The approximate criterion allows, among other things, a 15% difference in effects or a 0.05 difference in p values. This result therefore does not mean that 73.5% of the initial 600 articles are reproducible. Data availability already selects what can be checked. For a psi publication, depositing a file is a first step; a useful check involves having someone outside the original analytical process reproduce the main tables and results.
Open science reforms themselves require evaluation
OSIRIS and TIER2, Royal Society Open Science, 26 August 2026, DOI 10.1098/rsos.252472. This narrative review reclassifies interventions from a database of 105 studies assembled for a 2025 review, according to their mechanisms for changing research practices. The novelty lies in this analysis of mechanisms, not in a new set of replications.
Mandates and persuasion are the most studied approaches. Mandatory measures generally produce better results than voluntary recommendations for certain practices, notably data sharing; implementation remains incomplete without monitoring. Tools and practical support are promising. Training, incentives and learning by example have received much less evaluation.
For psi journals, we recommend combining explicit requirements with practical assistance: a deposit template, a minimal executable script and an availability check. A declared policy does not prove its effectiveness. The review’s heterogeneous evidence base does not support a universal ranking of reforms or a single causal effect attributable to them.
OSIRIS and TIER2 • Article on open science interventions
OSIRIS • Project summary of the findings, 7 September 2026
A comparative protocol on preregistration and exploration
Syed, Armstrong, Chan and Person propose comparing 300 matched articles across Registered Reports, preregistered articles and traditional publications. The project measures the presence of analyses explicitly described as exploratory. Its open evaluation was published on MetaROR on 5 May 2026. This is an evaluated protocol; results are still to come.
The reviewers highlight a decisive difficulty: measuring what authors report in their articles does not directly measure all the exploration actually conducted, or its creativity. Editorial policies can change how analyses are labelled. The authors’ response refocuses the project’s scope on reporting practices.
The lesson for psi research is precise: preregistration does not prohibit exploration; it identifies decisions made before and after examining the data. A Registered Report adds prior review and in-principle acceptance conditional on adherence to the protocol. These two arrangements must be distinguished when comparing rates of positive findings.
Syed and al. • Protocol, reviews and authors’ response on MetaROR
COS • How Registered Reports work
Fraud prevention begins with the protocol
James E. Kennedy, Researcher Fraud in Experimental Parapsychology, Psi Encyclopedia, updated on 23 September 2026. This synthesis revisits the Levy and Soal cases and the Parra case. It distinguishes established facts, suspicions and undetected fraud, whose frequency cannot be inferred from known cases alone. Fabrication or falsification of findings must also be distinguished from plagiarism.
The central argument concerns whether a researcher acting alone can alter results without independent oversight. Kennedy advocates prevention built into the protocol rather than trust resting primarily on subsequent investigations. His companion article, Preventing Researcher Fraud, updated on 9 July 2026, discusses raw-data archiving, audit trails, access controls and shared responsibilities.
Proposed applications to psi experiments include automatically recording raw data, separating target allocation from analysis, documenting exclusions and retaining independently controlled copies. These measures also reduce inadvertent errors. Preregistration documents analytical intentions; it does not guarantee the authenticity of the recorded data.
Kennedy • Researcher Fraud in Experimental Parapsychology
Kennedy • Preventing Researcher Fraud
Critical inquiry and anomalistics
Conspiracy beliefs and spirituality
Schofield, Howard and Rimmer, JSE 40(2), 7 July 2026. An online survey of 240 participants recruited through social media examines relationships between spirituality, moral purity, political orientation and Covid-19 conspiracy beliefs. The first three variables contribute to predicting these beliefs in a multiple regression; the authors supplement the analysis with mediation analyses and a network of relationships.
The study is relevant to the psychology of belief and the critical analysis of discourse. Voluntary recruitment and the observational design limit generalisation and causal interpretation. Cross-sectional statistical mediation does not establish the causal ordering of mechanisms. Nor do these associations support conclusions about the validity of experimental psi findings.
Schofield and al. • Conspirituality • JSE 40(2), 174-182
Ghosthunting Gruppen und Ghosthunter in Deutschland
Gerhard Mayer and Sonja Nowara describe 21 groups and 39 individuals using two online surveys. The study documents motivations, beliefs, instrumental practices and reported exceptional experiences.
It is useful for the sociology of anomalistic practices and for preparing field protocols. The small, self-selected sample permits no direct inference about the existence of the phenomena being investigated.
Primary source : Mayer and Nowara 2026
Academic books
Thomas Rabeyron’s theoretical synthesis of psi
The Psi Enigma: When Anomalous Experiences Transcend the Everyday World. Routledge, 19 May 2026, ISBN 9781041270140. This English-language publication translates Codex Anomalia, published in French in 2023. It therefore belongs to retrospective bibliographic coverage and the international dissemination of an existing work. The book connects exceptional experiences, experimental findings and theoretical proposals.
Its value lies in making a synthesis informed by French scholarship accessible to English-speaking readers. Nemo C. Mörck’s review for the SPR acknowledges this contribution while expressing reservations about some theoretical formulations and several factual details.
Routledge • The Psi Enigma catalogue entry
Gabriel Andrade on the psychology of hallucinations
The Psychology of Hallucinations. Routledge, listed in the catalogue as on sale from 15 July 2026, ISBN 9781041328537; the edition’s record carries a 2027 copyright date. This distinction avoids confusing the commercial release date with the publication year on the edition. The book covers, among other topics, visual and auditory hallucinations and their perceptually convincing character.
This introduction from an academic publisher is relevant to the study of exceptional experiences and to defining phenomena. It is noted as a contextual resource, without attributing psi experiences as a whole to hallucinations or conflating phenomenological description with an explanation of the origin of information.
Routledge • The Psychology of Hallucinations
Priority reading
- Radin and Cline, followed by Wahbeh et al.: two very different studies for examining confirmatory status, dependencies and optional stopping.
- OSF Changes and the transition documentation: direct implications for ongoing projects and preservation of research materials.
- The two SCORE articles in Nature: read reproduction of calculations and replication with independent data together.
- Kennedy, Researcher Fraud in Experimental Parapsychology: incorporate data controls into experimental design.
- Bierman and Spottiswoode, with Tressoldi and Storm’s response: compare the definitions and units of analysis used to study decline.
- OSIRIS-TIER2 and the protocol evaluated on MetaROR: subject open science practices themselves to empirical scrutiny.