Case Studies

Case: the missing repeated session

A learner removes duplicate CSV rows to “clean” a report. The original count 3 becomes 2 even though the rows describe three study events. The defect is a mismatch between entity identity and equal values. Preserve repeated rows unless the input contract provides a reliable event identifier and a deduplication rule.

Case: a trusted-looking fingerprint

A report has a SHA-256 value, but nobody recorded which source bytes were hashed. The digest cannot establish the claimed run. Repair the provenance record with the actual input, command and code version; do not label a newly computed digest as historical evidence.