mikusama99/cs2-prompt-label-comparison-20260828

CS2 Prompt / Timeline Lab

Interactive review gallery for a small CS2 video-labeling comparison:

• —9 action classes
• —3 clips per action (stationary, move, and move-camera)
• —27 clips and 108 model responses
• —v1 / v2 prompt versions × action / chunk time policies

Use the filters at the top to narrow the gallery. Each card contains the selected clip, parsed event and chunk timeline, English and Chinese JSON, and the four result variants. action uses the parsed event interval; chunk uses the union of the active one-second chunks covering that event.

The accompanying comparison.md, comparison.csv, and comparison.json contain aggregate and per-case comparison statistics. The page is a static research preview; no API credentials or raw prompt transcripts are included.