Live, in this tab
The same integer kernel runs on your GPU and on your CPU. Both outputs are painted into one image, compared bit by bit and hashed. If they differ anywhere, the seam shows it.
The seal SHA-256, crypto.subtle
A XOR B black = identical
Every cell of A is XORed with the same cell of B. While the substrates agree this square stays black.
WebGPU is not available
The same claim, in the MIND compiler
Pinned on x86 and ARM, replayed here
The browser kernel above is hand-written. The compiler's own guarantee is tested in the public star-ga/mind repository: every workload fixture pins one SHA-256 for x86 (avx2) and one for ARM (neon), and continuous integration checks both on real runners.
Reproducible chaos: lorenz-q16 computing
A Lorenz attractor in Q16.16 fixed point, 1000 forward Euler steps from (0, 1, 0). Your browser reruns the fixture's exact integer recipe and hashes the 8 bytes of the final x, the way the repository's harness does.
- final x
- computing
- i64 LE
computing- this page
computing- avx2 pin
- neon pin
Computing the float comparison.
35 workload fixtures avx2 = neon in every one
Each tile is a directory under tests/cross_substrate_identity/. The pinned avx2 and neon digests are equal in all 35 at commit 05eb95a. Hover or tap a tile for both full hashes.