When designing a MapReduce job that totals citations by violation type, what primarily determines which values a reducer sees together?

When designing a MapReduce job that totals citations by violation type, intermediate keys primarily determine which values a reducer sees together because the framework groups (and typically sorts) values by key for reducers—key design is aggregation design. The color theme of the city's open-data portal CSS does not control reducer grouping. Random UUID filenames that ignore the emitted keys entirely would prevent meaningful aggregation by violation type. A rule that every reducer must process every key in the cluster contradicts partitioned reduce assignment. Exam caveat: choose map output keys as the grouping grain for the civic total you need. Operational check: emit violation-type as the intermediate key, confirm each reducer receives only its key set, and match totals to a citation control file before enforcement analytics publish. In municipal platform operations, record the rationale in the architecture decision record so future on-call staff know which pattern was chosen and which failure modes the rejected options would create for city services. Re-validate after the next maintenance window that monitoring, access paths, and recovery steps still match the documented design for agencies that depend on this platform.