During the review of PR #12 (Robust fitting), CodeRabbit flagged that join_and_score() assumes rows are in ascending date order when computing diff(csum).
If the joined reference data are not sorted, negative increments can arise and CRPS will be incorrect.
Discussion thread: #12 (comment)
Action: enforce chronological ordering (e.g., sort or key by date within sample, slide) before calling diff().
Reported by @coderabbitai, requested by @jamesmbaazam.