CI / test (pull_request) Successful in 43s
- collection/etl/runs.py fingerprints(): a form or register run's id is now <prefix>-<hash of the row's non-empty answers, timestamp included> instead of its row number, so deleting or re-sorting rows in the sheet moves no other run's id, and a column added later changes none. The 32 rows that are exact copies of another row get -2, -3 (identical, so which is which does not matter). A row edited in the sheet gets a new id. Every id in runs_clean.csv and overlap.json changes once; nothing else does (no statistic or model reads the ids). - tests/core/test_site_stats.py (9 tests): measures.yaml conditions, source groups and presets, bond denominators without deliberate deaths (cards, encounters, route map), the riders' paths, and a JSON-safe stats.json, on a fixed sample of the published runs. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ