MindFold3D: Topology-Guaranteed Generation and Bidirectional Cognitive-Geometric Mapping of Voxel Stimuli for Spatial-Cognition Games
Authors/Creators
Description
Games designed to assess or train spatial cognition need unlimited stimuli with known, controlled properties. Procedural content generation promises this, but ordinary generators cannot guarantee such properties. We present MindFold3D, a spatial-cognition game whose platform-agnostic engine generates voxel stimuli under a topological guarantee. The number of volumetric holes in every shape, which grounds its topological tiers, is enforced by construction plus per-shape validation. Ring templates reserve each hole as inviolable empty space while varying tunnel orientation, aperture shape, and depth, and a certification step computes the first Betti number directly, regenerating nonconforming shapes. A bidirectional cognitive–geometric mapping steers five more difficulty features and labels every shape by measured geometry. Across two studies, the Betti number was exact in all 5,000 shapes, tier fidelity was 99.7%, and first-pass acceptance was 92–100%. The system ships as a tagged public repository. Tiers are geometric design targets awaiting human calibration.
Notes
Version 4 is a corrected revision of version 3. The changes:
-
Study 1 now regenerates exactly. In the version 3 code, the hole-tier generator seeded each attempt from the operating system rather than from the study seed, so Study 1's three cyclic Structural Complexity tiers did not reproduce from the published seed. With the seeding corrected, repeated runs are identical on Linux and Windows.
-
Rerunning Study 1 changed three branching-factor values in Table 2: Medium 2.45 ± 0.57 to 2.44 ± 0.56, High 4.89 ± 0.34 to 4.86 ± 0.38, and Expert 9.43 ± 0.88 to 9.51 ± 0.76. All other results are unchanged, including Table 1, the 99.7% emitted-set tier fidelity, the 92–100% first-pass acceptance, and the exact Betti number in all 5,000 shapes.
-
The results regenerate from release tag paper-v9 of https://github.com/snhwang/MindFold3D. The release adds studies/verify_reproduction.py, which reruns both studies and checks them against the archived results. It also fixes two game defects that do not affect any reported result. The distractor tiers now run radical, moderate, and subtle as described, and hard levels no longer fall back to a placeholder trial.
-
The complete system (game, generator, study drivers, and result data) remains publicly released under a source-available license (OCVSAL v1.0), with a live browser demonstration at https://mindfold3d.replit.app.
Files
zenodo_revised.pdf
Additional details
Software
- Repository URL
- https://github.com/snhwang/MindFold3D