1 Title and one-sentence summary

Occluded-Surface Imagery Training: Spatial Transfer and Hidden-Feature Identification

A two-center randomized trial will test whether six weeks of backside and tactile imagery training improves spatial transfer and whether performance is fully explained by sensory information, memory, and inferential cues.

Applicant: Future Mind Institute, a US nonprofit; priority A; duration 24 months; request US$650,000. This expanded Project Brief follows the requested length and will exceed one printed page.

2 Scientific question and background

How broadly does spatial training transfer, and what information supports accurate imagery? Uttal et al. (2013), reviewing 217 studies, support spatial trainability without validating this particular exercise. Yang et al. (2020) support plasticity following childhood spatial training, which cannot establish efficacy in adults. Pearson et al. (2015) describe cognitive and neural mechanisms of imagery without demonstrating access to unencountered information. Uttal et al., 2013; Yang et al., 2020; Pearson et al., 2015.

The experiment separates inferring an unseen surface from previously visible structure from identifying an independently randomized surface. Ordinary transfer would extend spatial-training evidence. Random-feature identification would challenge the sufficiency of conventional information sources only after independent replication and a successful leakage audit. In 2026, automated object generation, isolated target storage, and standardized scoring make this behavioral test feasible without first adding neuroimaging.

3 Origin of the hypotheses

The Seth Material serves strictly as a hypothesis generator. suggestion-12, dreams-projections §2 describes an exercise imagining an object's backside and encouragement to practice. you-create-your-own-reality-10, unknown-reality-2 §744 proposes exploration through internal mental activity. Neither statement constitutes empirical evidence. “Inner tactile enveloping” is operationalized as imagining contact with successive surfaces of an object.

The material does not specify that unknown random surfaces must be identifiable. That stronger version is an explicit prospective commitment of this project, not a prediction retroactively attributed to the source. Source documentation will use the applicant-supplied Zenodo dataset, licensed CC BY 4.0, and research website. The team will archive the source version and checksum at project initiation.

4 Primary hypotheses and falsifiable predictions

H1—Spatial transfer: imagery training improves a composite of untrained spatial tasks relative to the expectancy control. The null is zero adjusted between-group difference. The planning effect is d=0.45, expressed in baseline standard-deviation units; the smallest meaningful effect is d=0.20.

H2—Random hidden-feature identification: the imagery group exceeds both 25% four-choice chance accuracy and the equally weighted average of the two control groups. The null is that at least one positive condition fails. Planning assumptions are 30% versus 25% mean accuracy and a participant-level accuracy standard deviation of 0.10; the smallest meaningful difference is 2.5 percentage points. Both tests must pass independently at both centers, with comparisons against each control pointing in the same direction. Above-chance performance alone would not demonstrate a training effect.

5 Research design

Each center will recruit 300 adults aged 18–65, stratify randomization by baseline spatial ability, and allocate participants 1:1:1 to backside/tactile imagery, mental rotation, or expectancy-matched nonspatial attention training. Training comprises twenty minutes per session, five sessions weekly, for six weeks. Assessment occurs at baseline, immediately after training, and three months after training ends.

Assessment uses new objects. Inferable and random backside tasks occupy separate, counterbalanced blocks; each random assessment contains 80 trials without correctness feedback. In the random condition, four exchangeable candidate surfaces are generated first. An isolated server then selects the target uniformly, independently of visible geometry. A secretly salted hash commitment locks the target before the response; the participant-facing software receives the candidates but never the target label. Scoring occurs after responses are locked. Seed reuse is prohibited. Cache behavior, network traffic, timestamps, and file metadata undergo audit, and testing uses controlled devices.

Participants cannot be blinded to training activities, but are not told which condition is predicted to perform best. Standardized explanations support expectancy matching, which is measured directly. Assessors, analysts, and participant-facing personnel cannot access target labels. Group codes remain masked until the analysis is signed and locked.

Allowing 15% attrition yields approximately 85 evaluable participants per arm. At two-sided α=0.025 per hypothesis family, H1 has approximately 87% power assuming baseline–outcome correlation r=0.50. For H2, an 85-versus-170 participant comparison has approximately 94% power under the stated assumptions; the chance comparison approaches 100%. Requiring success at both independent centers gives approximately 88% joint power for H2 and 76% for H1. These are planning approximations, not 80% power guarantees for the smallest meaningful effects. Participant-level simulation will finalize operating characteristics before recruitment; inadequate simulated power requires revised enrollment, budgeting, and registration.

Preregistration fixes contrasts, outcomes, stopping rules, exclusions, missing-data handling, and audit criteria. Center two's protocol is frozen before center one's unblinding, preventing outcome-driven redesign.

6 Data and analysis plan

The co-primary outcomes are post-training spatial composite score and random-feature identification accuracy. The composite combines prespecified mental rotation, perspective-taking, and spatial visualization tests, standardized against the baseline distribution and equally weighted. H1 uses baseline-adjusted ANCOVA.

H2 analyzes participant-level accuracy with robust standard errors. A target-label randomization test evaluates the 25% benchmark; a treatment-assignment permutation test evaluates the training contrast. Both components must pass at α=0.025 within this intersection–union family. Allocating α=0.025 to each hypothesis family controls overall error at no more than 0.05. Each center receives an independent verdict; pooled results cannot rescue failed replication.

Secondary outcomes include inferable-backside accuracy, three-month retention, separate control comparisons, confidence calibration, adherence, and expectancy. False-discovery-rate control applies to secondary comparisons. Primary analyses follow randomized assignment; multiple imputation and sensitivity analyses for nonrandom missingness address incomplete outcomes. Adherence-restricted analyses are supplementary.

Deidentified data, executable code, training materials, audit logs, and all protocol deviations will be released. Target labels and seeds remain withheld until both centers finish, avoiding contamination of replication. Trial counts improve measurement precision but do not substitute for independent participants in power calculations.

7 Milestones and duration

Months 1–4: establish adversarial collaboration, obtain ethics approval, conduct a separate usability pilot, complete power simulations, and preregister. Months 5–11: center-one recruitment, training, and follow-up. Months 9–18: center-two implementation and follow-up under the frozen protocol. Months 19–21: unblinding, independent analysis, and audit. Months 22–24: joint reporting, public data release, and submission.

Replication funding is reserved from the outset. A negative first-center result does not automatically cancel the second center, preserving a balanced assessment of both positive and negative findings.

8 Indicative budget

The requested US$650,000 comprises:

• Personnel and two-center coordination: $250,000.
• Participant compensation and retention: $120,000.
• Software development and target isolation: $65,000.
• Independent security audit: $45,000.
• Statistics and adversarial collaboration: $45,000.
• Ethics administration, data management, and open publication: $25,000.
• Institutional indirect costs: $65,000.
• Contingency: $35,000.

This planning estimate falls within the stipulated $450,000–$750,000 range and requires institutional quotations. Neuroimaging is excluded from this behavioral project.

9 PI profile and prospective collaborators

The PI should have expertise in adult spatial cognition, randomized trials, and open science. The team requires an independent statistician, a software-security lead, and co-designers who respectively support and question the stronger hypothesis.

Prospective advisers include David H. Uttal and Northwestern's STAR Lab for spatial transfer; Nora S. Newcombe at Temple University for spatial cognition; Joel Pearson and UNSW's Future Minds Lab for imagery; and Zoltan Kekecs for transparent replication methods. These are proposed invitations, not representations of participation or endorsement.

10 Risks and abandonment criteria

Geometry, random seeds, caches, or personnel could disclose target information. An audit finding a usable leakage pathway invalidates the affected random-target batch, handled under preregistered rules. Increased confidence without improved accuracy does not count as success.

If the upper bound of the pooled two-center 95% confidence interval lies below d=0.20, the team will stop promoting meaningful spatial-transfer benefits from this protocol. If the upper bound for either random accuracy above chance or the training advantage lies below 2.5 percentage points, it will abandon a practically meaningful strong version at this training dose. Wide intervals indicate uncertainty rather than exclusion.

Without independent replication, the project will make no claim challenging conventional information sources. A negative result cannot eliminate ontological propositions that were never operationalized. These boundaries prevent both overinterpretation of anomalies and retrospective expansion of the hypothesis after failure.

11 Ethics and compliance

Both centers require ethics approval and informed consent. Participants will be told that the study is neither diagnostic nor therapeutic; recruitment will not select by belief or imply possession of special abilities. Withdrawal is permitted, completed participation is compensated, and discomfort is monitored.

Data collection will be minimized, identity mappings stored separately, and international sharing governed by applicable requirements. Future Mind Institute will retain its neutral consciousness-and-life-sciences positioning. The Seth Material Research Program is one public program under the institute; funding provenance, conflicts of interest, and a commitment to publication regardless of outcome must be disclosed.

12 Fit with established funders

The project aligns with Bial's interest in scientifically analyzable research on mental phenomena. It intersects with Fetzer's engagement with science and inner experience, while requiring a clear public-value case. Governance draws on Templeton World Charity Foundation's Cogitate model: competing predictions, preregistration, and cross-laboratory testing.

BICS, UVA DOPS, and IONS are potential dialogue counterparts to assess, not presumed funders of this proposal. These are assessments of thematic and methodological fit, not claims of eligibility, an open funding call, or support. An interpretable negative result is a core deliverable with value independent of whether the stronger hypothesis succeeds.