Memory In Nature: A Study of Morphic Resonance Predictions on Learning

Black, G., Butzer, B., & Sheldrake, R. (2026). Memory In Nature: A Study of Morphic Resonance Predictions on Learning. Journal of Anomalous Experience and Cognition, 6(1), 11–37. https://doi.org/10.31156/jaex.27438

AI Assessment

Three studies testing a prediction of Rupert Sheldrake’s morphic resonance theory, that the daily Wordle puzzle should get easier to solve as the day goes on, because earlier solvers seed a shared “morphic field.” The results are mixed and mostly unsupportive: Study 1 found no effect, Study 2 found players did slightly better on their first two guesses by evening (but worse on later guesses), and Study 3 failed to replicate Study 2 and mostly ran the opposite way. Every effect size was tiny (Cohen’s h around 0.01). Two problems dominate: the “significant” results come from samples of hundreds of thousands to millions of players, where trivial differences reach p < .001, and morning versus evening players are simply different populations (time zones, who plays when), a confound that can produce time-of-day differences with no morphic resonance at all. The authors report the mixed picture honestly and conclude any effect must be subtle. Sheldrake, who originated the theory, is a co-author. This audit reports what the studies found and how; it takes no position on whether morphic resonance is real.

Provenance

DOI. 10.31156/jaex.27438 · Journal of Anomalous Experience and Cognition 2026, 6(1), 11 to 37. Open access under CC-BY.

Study type. Three quantitative observational studies (one questionnaire-based, two using large-scale public data) testing a specific prediction of morphic resonance.

Authors. Georgia Black and Bethany Butzer (The Alef Trust) and Rupert Sheldrake (Temenos Academy). Sheldrake is the originator of the morphic resonance hypothesis being tested, which is relevant author context.

Data availability. Available from the corresponding author on request (not openly deposited).

Source basis. Every figure below is taken from the article’s own Abstract, Highlights, and Results.

What the paper reports

Morphic resonance predicts that a skill becomes easier to learn as more people practise it, because the pattern of learning is embedded in a “morphic field” that later learners can tune into.¹ Applied to Wordle, the daily global puzzle, this predicts that people playing later in the day should need fewer attempts, having benefited from earlier solvers. The three studies test that prediction with different data sources, and the authors conclude that if such an effect exists it is subtle and easily masked.

Taken together, the findings suggest that if morphic resonance is at play, it is likely a subtle effect that is easily inhibited by factors such as conscious cognitive processes.

How it was run

• Study 1. A questionnaire collected Wordle scores and completion times from an experimental group (playing the official New York Times Wordle) and a control group (playing a custom Wordle), analyzed with Spearman rank correlations (distributions were non-normal by Shapiro-Wilk).
• Studies 2 and 3. Used WordleBot to sample Wordle players worldwide, comparing the percentage of players solving on each guess (1 through 6) between morning and evening, with six two-sample z-tests for proportions per study and Cohen’s h for effect size.
• Sample sizes. The public-data studies drew on very large samples: average morning samples of 641,303 (Study 2) and 696,690 (Study 3) players, and average evening samples of 1,745,345 (Study 2) and 1,779,495 (Study 3).
• Registration. The studies were not preregistered.

Results, as reported

Values are reproduced from the article’s Abstract and Results. The pattern is internally inconsistent: Study 1 is null, Study 2 supports the prediction on early guesses but contradicts it on later ones, and Study 3 mostly runs opposite to Study 2, with all effects negligibly small.

Eleven-dimension audit

Pre-registration

Not preregistered. With three studies and six per-attempt tests in each of the two large-sample studies, and no registered analysis plan, there is substantial room for some contrasts to reach significance by chance, and the direction of the “significant” results is not consistent across studies.

Randomization

Study 1 uses an experimental and a control group, but the core comparison in Studies 2 and 3, morning versus evening, is observational: players are not assigned to a time of day, they self-select, so time-of-day differences confound with who plays when.

Sensory leakage

Not applicable in the perceptual sense. The decisive analogous threat is a sampling confound, addressed under the adversarial record: morning and evening player populations differ systematically.

Blinding

Not applicable to an observational analysis of public game data; there are no raters or conditions to blind.

Optional stopping

Not a classic concern for the fixed public datasets, but the use of three studies and multiple per-guess tests, without a registered plan, gives the analysis flexibility that a single pre-committed test would not.

Outcome measure

Wordle attempts (Study 1) and the percentage of players solving on each guess (Studies 2 and 3). These are reasonable operationalizations of “easier to solve,” but breaking the outcome into six per-guess tests creates a set of contrasts that can point in opposite directions, as they did.

Effect size

Uniformly negligible: Cohen’s h values around 0.01 are far below any conventional threshold for even a small effect. Whatever their p-values, effects of this size are practically meaningless and are exactly what one would expect from trivial differences amplified by enormous samples.

Multiple comparisons

Six z-tests per large-sample study, across three studies, with no stated correction. Some reach significance in one direction (guesses 1 and 2 in Study 2) and others in the opposite direction (later guesses, and most of Study 3), which is the signature of noise rather than a coherent effect.

Internal replication

The built-in replication failed: Study 3 did not reproduce Study 2, and Study 1 found nothing. A hypothesis that survives only in one of three attempts, and only on part of that one, has not been internally replicated.

External replication

The paper adds to a small prior morphic-resonance literature and is itself a mixed, largely unsupportive contribution to it. It does not establish a replicable effect.

Transparency

Honest in reporting: the null Study 1, the contradictory within-Study-2 pattern, and the failed Study 3 are all reported plainly, and the conclusion is appropriately modest. The transparency limits are that the data are only available on request and the studies were not preregistered.

The adversarial record

• Huge samples, trivial effects. The “significant” results come from samples of hundreds of thousands to nearly two million players. At that scale, a Cohen’s h of 0.01 reaches p < .001 while meaning essentially nothing. Statistical significance here is a function of sample size, not of a substantive effect.
• The morning-versus-evening confound. Players who play in the morning and players who play in the evening are different populations (different time zones, routines, and skill mixes), and evening samples are roughly three times larger. Any systematic difference in their performance can arise from who is playing when, with no morphic resonance involved. The design cannot separate the two.
• Incoherent direction. Study 2’s support on the first two guesses is undercut by significant effects in the opposite direction on later guesses, and Study 3 mostly reverses Study 2. A real facilitation effect should not flip sign across guesses and studies.
• Author context. The originator of the hypothesis is a co-author. That does not invalidate an honestly reported near-null, but it is part of the context, and to the authors’ credit the paper does not overstate the mixed findings.
• What is done well. Using a natural, massive dataset is creative, the built-in replication is good practice, and reporting the failure to replicate rather than burying it is exactly right. The honest conclusion is that these three studies do not provide clear support for the morphic-resonance prediction they set out to test.