Dozens of "statistically significant" cells light up in pure noise just because you tested 2,500 of them at p<0.05. Watch an actual Gaussian-Random-Field cluster-extent correction — the same analytical machinery used in brain-scan morphometry — erase almost all of them, live, from a closed-form formula instead of a permutation test.
Grounded in
arXiv:2603.11344 — "Hybrid eTFCE-GRF: Exact Cluster-Size Retrieval with Analytical p-Values for Voxel-Based Morphometry" (Yin, Chen, Miki, Yang). The paper fuses eTFCE's union-find exact cluster-size tracking with pTFCE's analytical Gaussian-random-field p-values, replacing permutation-based cluster testing (>1000x slower) with closed-form inference — whole-brain VBM in ~85s instead of permutation runs, with FWER controlled at nominal level on synthetic phantoms (0/200 null rejections). This demo runs the real 2D GRF Euler-characteristic cluster-size math on a 50×50 "games × prop-split" grid, not that paper's exact 3D pipeline.