ucsc-vlaa/visualclaw/semantic-precision-matching

Match Precise Semantic Details to Video Content

1. Identify key semantic distinctions between similar options (e.g., "grind and polish" vs. "sharpen" vs. "cut").
2. For each option, extract the core claim: What specific process or action is being described?
3. Cross-reference each claim against what the video actually demonstrates—look for exact verb and object matches.
4. Pay attention to frequency markers: "repeatedly," "once," "consistently" must align with observed behavior.
5. Reject options where the semantic content (the actual process described) does not match the video, even if the option is well-constructed.

Example: Options describe: (A) grind and polish, (C) sharpen, (E) cut Video shows: grinding and polishing motions Correct: (A) — exact semantic match Incorrect: (C) or (E) — different processes not observed

Anti-pattern: Selecting an option because it uses similar keywords without verifying the core semantic claim (the actual process or action) is demonstrated in the video.