Scrutinizer is grounded in published visual neuroscience: cortical magnification
(Schwartz 1980; Blauch, Alvarez & Konkle 2026, arXiv preprint), spatial frequency decomposition via
Difference-of-Gaussians, per-channel chromatic pooling (castleCSF), and feature congestion
scoring (Rosenholtz 2007). Some perceptual phenomena, such as texture synthesis for crowding
(mongrels, Balas, Nakano & Rosenholtz 2009; metamers, Freeman & Simoncelli 2011), are approximated
to meet real-time performance constraints.
Spatial-frequency attenuation reaches a composite r = 0.60, below the r > 0.9 target (
spatial acuity report).
Chromatic pooling results are in the
color search report.
Feature congestion matches a Python port of the Rosenholtz lab toolbox at ρ=0.93.
Crowding geometry is tested with
reference stimuli.
Wave 7c did not demonstrate the isolated-versus-flanked crowding asymmetry, and Mode 17
length-tuning and the oblique effect are not yet validated against human data.
Our
system paper (in preparation)
details the full pipeline. See also the
literature review.