fix: count annotator history from both HF snapshot + local file to survive restarts

fix: initialize annotator count from historical data on login

feat: task pool with prompt_interaction on all subsets + updated questions

feat: prompt+caption for all subsets, 6 radios for env, updated question wording

fix: use CSS class with !important to force text visibility in aux info box

feat: regenerated task pool with action_interaction on all subsets

feat: camera question on all subsets, 5th radio, mask overlay for Object, fix text visibility

fix: remove MBench-V tabs, fix white-on-white text in aux info

feat: add task_pool.json for MBench-A (280 tasks)

feat: add MBench-A pairwise annotation tab (4 subsets, multi-dim questions)

Upload requirements.txt with huggingface_hub

Upload app.py with huggingface_hub

trigger restart: pick up cleaned annotations from dataset

fix: disable SSR (fixes /_app/immutable 404 + /video 308 redirect)

fix: keep Gradio native launch; monkey-patch /video onto its own FastAPI (fixes /_app/immutable 404)

fix: narrow-screen layout (wrap rows, min_width on columns) to prevent right half from vanishing

feat: add colored category badge (Causal/Human/Spatial&Object) in bottom-left meta

fix: proxy videos through Space same-origin to avoid cross-origin <video> blocks

i18n: UI copy to Simplified Chinese

feat: binary Y/N verdict; 2x2 layout; shared claim queue (no cross-annotator overlap)

fix: show 5-segment prompt (level_3) with segment headers

feat: add skyreels+longcat (pool 4672); per-annotator dedup; global progress bar

fix: render video via native <video> HTML to bypass gradio SSRF-download of HF xet CDN

fix: empty gr.State + api_name=False (workaround gradio_client schema bug)

fix: simplify UI to single page; no visibility toggling; plain-value returns

fix: bump to gradio 5.9.1 (HfFolder removed in new huggingface_hub)

fix: pin python 3.11 (avoid py3.13 audioop removal)

init: minimal Gradio annotation template (6 models, 3504 items)

initial commit