Rerun Scope Is Part of the Data Contract
Why rerun scope, support-data snapshots, and processing windows must be explicit in batch workflows.
Details are generalized and sanitized; numbers are rounded where appropriate to preserve confidentiality while keeping the engineering lessons accurate.
Context
Rerun behavior can drift in delivery pipelines where core data is table-based but support data still arrives as snapshots.
Problem
Current-period runs and targeted historical reruns did not always resolve support-data snapshots the same way. When snapshot assumptions were implicit, historical reruns could read the wrong reference state.
Approach
I separated two values that were previously coupled: the metadata anchor used to select support-data snapshots, and the processing window used to compute output periods.
This kept the standard current-period path unchanged while making targeted historical reruns explicit about input assumptions.
Result
Reruns became easier to review because operators could see both the output period and the selected support-data snapshot. The short-term fix was explicit run parameters; the long-term fix is versioned support data.
Reusable pattern
- Separate metadata anchor from processing window.
- Make support-data snapshot assumptions visible.
- Use targeted reruns only when input assumptions are explicit.
- Prefer versioned support data for long-term reproducibility.