Accorma RGAA Corpus

Labeled test-case corpus used to evaluate LLM-augmented accessibility conformance testing against the RGAA 4.1 referential, as described in the accompanying paper "How Much Does an LLM Layer Add to Automated Accessibility Auditing? A Controlled Comparison of axe-core and an LLM-Augmented Pipeline on RGAA 4.1" (Ndiaye, Chifu, 2026). The corpus contains 1,435 HTML test cases, each paired with ground-truth labels for one or more of the 33 RGAA 4.1 criteria that are judgment-dependent rather than fully mechanizable by automated tools such as axe-core. Cases are balanced to at least 20 examples per polarity (conforming / non-conforming) per targeted criterion. Sources: 135 cases adapted from official W3C ACT Rules test cases, 1,300 synthetic cases authored for this study. Each case is a directory containing page.html (the test page) and labels.json (gold labels and metadata). Full schema and methodology are documented in the README. Code and ongoing updates: https://github.com/0SansNom/accorma-rgaa-corpus