Hugging Face's logo

Datasets:
pooria
/
benchmark-accessibility-fix

Tasks:
Text Generation
Modalities:
Text
Formats:
json
Languages:
English
Size:
< 1K
Tags:
benchmark
llm-evaluation
ai-benchmark
model-evaluation
evals
grading
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
benchmark-accessibility-fix / README.md
pooria's picture
pooria
Publish benchmark task data
8b99f6f verified
preview code
|
Raw History Blame Contribute Delete
3.36 kB
metadata 
license: mit
language:
  - en
pretty_name: Accessibility Fix
tags:
  - benchmark
  - llm-evaluation
  - ai-benchmark
  - model-evaluation
  - evals
  - grading
  - accessibility
  - a11y
  - wcag
  - openrouter
  - openai
  - mcp
  - typescript
  - agent-benchmark
task_categories:
  - text-generation
size_categories:
  - n<1K
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/tasks.jsonl

Accessibility Fix

Fix a broken page so keyboard, screen-reader and contrast checks pass.

What it measures

Whether a fixed HTML page really repairs injected keyboard, screen-reader, and contrast defects without deleting or hiding content.

Category: coding. Status: alpha.

How to run 

$ npx benchmark-accessibility-fix list
$ npx benchmark-accessibility-fix run --model <model-id> --provider openrouter
$ npx benchmark-accessibility-fix score results.json

--provider mock runs the whole benchmark offline with no keys. Real models run through OpenRouter (OPENROUTER_API_KEY) or any OpenAI-compatible base URL (OPENAI_BASE_URL, OPENAI_API_KEY).

Data fields

data/tasks.jsonl holds 40 tasks, one JSON object per line (kinds: 40 single).

field type
id string
kind string
prompt string
expected string
metadata object

Provenance

Accessibility Fix provenance

This file is generated from the benchmark definition.

Synthetic accessibility fixtures

  • Kind: synthetic
  • URL: https://github.com/pooriaarab/benchmark-accessibility-fix/blob/main/scripts/gen-tasks.mjs
  • Licence: MIT
  • Retrieved: 2026-10-16T00:00:00.000Z
  • How to get it: Run npm run generate:tasks (or node scripts/gen-tasks.mjs --seed 20261015 after npm run build) from the repository root. The generator grades every reference page with the built package and re-draws until it scores 1. A hidden split uses another seed: node scripts/gen-tasks.mjs --seed <n> --split hidden --out <path>; hidden files use a11yh- ids and stay out of git and npm.
  • Contamination notes: Public tasks use seed 20261015. Treat them as possibly in training data; use a separately seeded hidden split for final comparisons.
  • Refresh: Change the seed, run the generator, and review the JSONL diff.

Grader

Deterministic static analysis of the fixed HTML: ten WCAG-cited checks score defects fixed, minus regressions, content loss, element removal, and hiding abuse. Model judging is off.

Limitations

This dataset contains only the public task split (40 tasks). Hidden evaluation splits use a private seed and are deliberately not part of this dataset. Source kinds: synthetic. These are evaluation prompts for the benchmark-accessibility-fix benchmark, not training data; scoring requires the package's deterministic grader. See PROVENANCE.md for contamination notes.

Citation 

@misc{benchmark_accessibility_fix,
  author = {Pooria Arab},
  title = {Accessibility Fix task data},
  year = {2026},
  url = {https://github.com/pooriaarab/benchmark-accessibility-fix}
}

Links