Discovery Lab Answers

RNG Psi Games — What the ESP-Nexus Library Shows

Coverage note (read this first)

The figures below come from the ESP-Nexus structured study database, which currently holds 39 studies on micro-PK / RNG effects spanning 1989–2025. The library’s share of the total published literature on this question has not been measured, so treat everything here as a summary of what the library holds, not a settled account of the field as a whole.

What “RNG psi games” means

In parapsychology, RNG (random number generator) or REG (random event generator) psi games are computerized tasks — PsiBall, Volition, Psi Invaders, and more recent web-based variants — where a participant tries to influence the output of a hardware or software random source, typically aiming to shift a bit stream above or below the 50% chance baseline. Effects, when reported, are very small and require large trial counts to detect statistically. The chart below plots effect size against publication year for the subset of results on a single comparable metric.

The evidence pattern across the library

The 39 studies yield 43 result rows, and they do not tell a uniform story.

Directions disagree. Fourteen of the 43 extracted results report null or below-chance outcomes — that is roughly one-third of the evidence base. Positive, null, and below-chance findings coexist across different labs, protocols, and participant populations, and that disagreement is a central empirical fact, not a peripheral detail.

The metrics are not poolable. The rows report standardized effect sizes (Cohen’s d, Pearson r), raw hit rates, z-scores, p-values only, and correlation-type metrics. These are different, non-comparable quantities. Averaging across them into a single “overall effect size” would be meaningless, so the table below organizes them by what each study actually measured.

Key findings by study

The Bösch (2006) large-scale meta-analysis — covering 380 intentional RNG studies — is a particularly important result to flag: the overall effect, while statistically significant on its own metric, landed below the chance expectation, not above it. That is a psi-missing outcome at scale, and it sits in direct tension with the positive pooled estimates from other analyses such as Williams (2021).

Early psi-game work at PRL

The earliest work in the library on computerized RNG psi games comes from Rick E. Berger and colleagues at Psychophysical Research Laboratories, using Apple II-based games — PsiBall, Volition, and Psi Invaders. A consistent finding across those experiments was that significant RNG deviations appeared in silent (no-feedback) conditions rather than, or in addition to, feedback conditions — a result that challenged the assumption that observation of RNG output is necessary for putative effects to occur. Cross-participant analysis across the three games also identified personality correlates: more introverted participants tended to perform better in silent conditions, while more extraverted participants did better in feedback conditions.

Unsettled signals you should know about

• Direction disagreement is substantial. Roughly one-third of extracted results are null or below-chance. The positive results are real in the data, but so are the nulls.
• Pre-registered confirmatory trials add complexity. Maier (2022)’s preregistered primary outcome (r = 0.01) is null, while exploratory subsets in the same study point positive. Dechamps (2019)’s preregistered Study 3 is null while earlier studies in the same series were positive.
• Protocol heterogeneity is high. Trials range from 80 to over 200,000; participants range from one (an experimenter as sole subject) to 4,092. Effect sizes where reported range from near zero to Cohen’s d = 0.49. Comparing across these studies is not straightforward.

Skeptical critiques

What critics argue. The library holds no sourced methodological critique of this work that meets the citation standard used elsewhere on this page, so none is stated here.

What the experimental data show. Williams (2021) computed an updated pooled retro-PK estimate across k = 42 studies and obtained z = 6.82 (p = 4.57 × 10⁻¹²), pointing in the opposite direction from Bösch’s aggregate. Null pre-registered trials — Maier (2022) primary (r = 0.01), Penberthy (2024) (z ≈ 0, p = 0.4995), Dechamps (2019) Study 3 (hit rate = 0.49775) — sit alongside positive pre-registered confirmatory results such as Mossbridge (2021) (p < 0.001).

Analysis. Bösch (2006) and Williams (2021) analyzed overlapping but differently scoped bodies of RNG literature and arrived at opposite directional conclusions. Pre-registered trials introduced after those meta-analyses have produced both confirmatory and null outcomes. No single pooled figure commands agreement across research groups using this paradigm, and the question of how much publication bias shapes the positive side of the literature has not been resolved by the studies in the library.

For deeper background on how RNG experiments are designed and interpreted, the RNG / Micro-PK Research methods page is a good starting point, and the Micro-PK (RNG / REG) evidence digest presents the full structured database this answer draws on.