[bug] MATH grader: ideographic thousands comma 1，000 extracts as "000" (Chinese fullwidth ，)

[marktech0813]

Problem

Chinese / fullwidth ideographic comma ， (U+FF0C) is a common thousands separator. The extract path splits on it and keeps only the last group:

extract_last_number("1，000")  # "000"
score_text("math500", "1，000", "1000")           # 0.0
score_text("math500", r"\boxed{1，000}", "1000") # 0.0

ASCII , and LaTeX {,} already work.

Expected

1，000 extracts/normalizes as 1000 and grades equal to gold 1000.

Environment

• origin/main

[github-actions]

Thanks for opening this issue. To help maintainers review it quickly, please add the missing details below:

• bug description
• expected vs actual behavior
• reproduction steps

For bugs, include the command you ran, the config snippet, and any error output. For benchmark or scoring issues, note the benchmark name and affected file path if known.