EqualifyEverything / benchmarks-ai-alt

The tool reaches the archived images with a path that climbs out of its own directory into the corpus construction project, so a server rooted in projects/corpus-validation cannot follow it. The README said to serve from that directory, which would have shown 78 broken images. Add .nojekyll so Pages copies the files rather than building them. Nothing here needs a site generator, and 965 image files would only slow a build. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>