nx_imgcorpus.nx

nx_imgcorpus.nx -- DETERMINISTIC SYNTHETIC IMAGE CORPUS for reverse-image measurement. Every organ that measures the reverse-image stack (the ruler, the engine gate, the CLI self-test) must agree on what "an image" is, or their numbers are not comparable. One generator, imported everywhere. WHY SYNTHETIC: a fixture corpus on disk makes a gate environment-dependent (missing file -> RED for a reason unrelated to the code) and makes results irreproducible across machines. A procedural corpus is byte-identical on the laptop, in the buildroot, and on the NAS, forever. WHY *THIS* CORPUS: a perceptual-hash measurement is only meaningful on images with MULTI-SCALE structure. Flat images would flatter every hash (nothing to lose under transformation); pure noise would destroy every hash (nothing to preserve). Real photographs have a large-scale layout, a mid-scale structure, and fine detail, so each image here is the sum of exactly those three: coarse -- an 8x8 random control grid, nearest-upscaled: the large-scale light/dark layout struct -- stripes, checkerboard, concentric rings, or a diagonal ramp, at a per-image period fine -- low-amplitude deterministic detail The measured mean pairwise dHash Hamming over the resulting corpus is ~32 of 64 -- exactly the "unrelated images" separation the shipped nx_phash_test observes on real photographs (33), which is the evidence that this corpus is a fair stand-in rather than a convenient one. license_tier: ORIGINAL