Summary

This PR improves the robustness of contest type classification and fixes JSON serialization to better handle missing data fields.

Key Changes

• Added _parse_rated_range() function: Extracts rating range parsing logic into a dedicated function that handles both legacy ("~") and new ("-") AtCoder rating range formats. This makes the code more maintainable and resilient to format changes.

• Refactored infer_contest_type(): Simplified the contest type inference logic by using the new parsing function and threshold-based classification instead of exact string matching. This approach is more flexible and easier to extend.

• Fixed JSON serialization: Added exclude_none=True to the S3 upload to omit None fields from the JSON output. This prevents the frontend from receiving explicit null values, which it rejects for partial models (e.g., a problem with difficulty but no time model).

Implementation Details

• The _parse_rated_range() function uses regex to parse rating ranges and returns (lower, upper) tuples, with None representing open upper bounds (e.g., "2000 -" or "All").
• Contest type classification now uses threshold comparisons on the upper bound rather than exact string matching, making it more resilient to minor format variations.
• Added clarifying comments explaining the JSON serialization change and the fallback to ID-based classification for unrated contests.

https://claude.ai/code/session_01Hsq4dCo1x9ve8W8inoWxe2