Expand the Square
With both slits open and coherent illumination, let A1 and A2 be the complex probability amplitudes for a photon to arrive at a particular position on the screen. The detection probability density at that position is the squared magnitude of their sum.
= |A1|² + |A2|² + 2 Re(A1A2*)
The last term varies across the screen as the relative phase changes. Where the two contributions line up, counts rise. Where they oppose one another, counts fall. A physical path marker can correlate the two paths with distinct states of a detector or another degree of freedom. When those marker states are perfectly distinguishable and you ignore the marker when plotting the screen, the cross term disappears. The detector can remain unread. Feynman Lectures, Vol. III, Ch. 1 · Englert (1996)
In the earlier companion artwork, Max Planck evokes the origins of quantum theory. The psytrance colours, psychic motifs, fractals, cymatics, sacred geometry, matrix code and mycelium evoke possible connections at the level of imagery. The endoDMT-inspired aesthetic and imagined Akashic field are artistic metaphors.
What the Experiments Tested
In the linked IONS programme, the experimental variable was what a person was asked to attend to. The optical outcome might be a spectral ratio, fringe contrast or counts at a dark fringe. Those metrics can respond to changes in the apparatus or analysis as well as to changes in coherence. Establishing a causal effect of attention would require ruling out ordinary channels and distinguishing a fringe change from a specific physical measurement mechanism. IONS Double-Slit Project
2012 · Six Experiments With Laser Light
The IONS team reported that their spectral ratio fell in the predicted direction across 250 sessions contributed by 137 people (z = −4.36, p ≈ 0.000006). The measured outcome was a ratio derived from the optical pattern. It did not directly measure which path a photon took. Radin et al. (2012)
2015 · Single-Photon Test
The team predicted more counts at an interference minimum during directed attention, as expected if the dark fringe were filled in. Across six formal experiments the mean count moved the other way, by about two standard deviations on their reported test scale. The authors reported increased variance across experiments. This left the directional prediction unsupported. Radin et al. (2015) · BIAL Project Report
2016 · Online Attention and a Later Data Reanalysis
The authors reported a 5.72-sigma result for human sessions at the centre of the pattern, with 2,985 human sessions and 5,738 automated controls. Tremblay reanalysed the same dataset. The corrected 2021 version discusses how lag selection and the treatment of different years affect the result and finds the data fall short of strong evidence. PLOS republished the reanalysis after its own trimming procedure was found to produce underestimated p-values. This was a second analysis of existing records. Radin et al. (2016) · Tremblay, Corrected Analysis · Correction Notice
2019–2021 · What Happened When Nobody Was There?
In a commissioned study, the IONS team ran human and empty-room sessions. Walleczek and von Stillfried’s analysis found no confirmation in the participant mean comparisons and reported a nominal p = 0.021 for one sham comparison. Radin and colleagues argued that none of the eight mean comparisons survived a false-discovery-rate correction and pointed to a variance result. The two groups disagreed about how to treat the comparisons. IONS’s subsequent paper also reported a null planned analysis and positive exploratory analyses. These publications discuss the same commissioned experiment. 2019 Analysis · IONS Commentary · Response · 2021 IONS Report
Independent Studies · Mixed Results
Gabriel Guerrer built a separate apparatus at the University of São Paulo. He reports that an 80-session preregistered follow-up failed to reproduce the finding from 160 exploratory sessions. A further exploratory analysis yielded another lead, which lacked a fresh confirmatory test. The work remains a preprint. Vujanac and colleagues reported positive optical results in a separate 26-person study with EEG. Their proposed brain–apparatus entanglement is a mechanistic interpretation that would require a direct quantum test. Guerrer’s Project Account and Preprint · Vujanac et al. (2019)
2025 · A Preregistered Diffraction-Grating Study
Forty-seven experienced participants received custom optical devices. The combined data supported none of the three preregistered hypotheses. An outward- versus inward-focus comparison and a subsequent trend analysis produced positive results. These findings give a follow-up study specific predictions to test on fresh data. Radin (2025) · BIAL Study Summary
2026 · Humans and AI Viewed a Prerecorded Signal
Radin and Cline used diffraction-grating readings recorded months earlier to drive an online candle animation. They report 1,000 human and 1,000 AI-agent sessions and one significant result among eight outcomes after correction, in the human-observed intention condition (p = 0.00007). The reported result concerns a regression statistic on selected records. To establish a human–AI difference, inspect the direct interaction and its uncertainty. To interpret the result as retrocausal, verify how recordings were fixed, how windows were selected and assigned, how reuse was handled and whether the permutation preserved the assignment structure. Treat the sessions collected before and after preregistration separately when judging prospective evidence. Radin and Cline (2026) · Preprint Record
What Would Change the Answer?
- Validate the instrument. Use a known physical path marker as a positive control. Run masked attention, rest and empty-room sham blocks alongside a reference channel. Record temperature, vibration, laser power and detector behaviour.
- Fix the inference. Preregister one primary outcome, one timing window and a smallest effect size worth detecting. Randomise blocks or sessions at the level used in the analysis and account for repeated sessions and shared runs. Report effects and uncertainty in addition to p-values.
- For the prerecorded claim, lock the archive first. Hash the entire optical archive before recruiting participants and retain a tamper-evident log of eligible windows, assignments and exclusions. Preregister the direct human–AI interaction and run human, AI and no-agent controls against the same assignment pipeline.
- Repeat independently. Different teams should run the frozen protocol with fresh data and share the raw traces, assignment logs and analysis code. A reproducible effect specific to the intention condition would be grounds to investigate mechanisms. Precise null results would constrain the effect size under the tested protocol.