A 311 team must design cleansing rules for category codes. What should they do first?

Profiling reveals how often category is missing, how long description fields run, and which distributions or outliers will break naive cleansing. Those facts drive whether rules should map free text, impute, quarantine, or reject—and prevent gold SLA dashboards from being built on misunderstood 311 shapes.

Profiling ticket channel mix and duplicate rates before designing cleansing rules for category codes fails by relevance: channel distribution and duplicate counts say nothing about category_code null rates or free-text lengths, so rules would be designed without evidence on the fields they must actually cleanse. Profiling ticket agent workload and resolution times before designing cleansing rules for category codes misdirects discovery: workload and timing metrics describe operational performance, not category missingness or description shapes, leaving the null rates and text distributions that drive mapping, imputation, or quarantine decisions unexamined. Profiling ticket location precision and closure dates before designing cleansing rules for category codes profiles the wrong attributes: location accuracy and closure timing cannot substitute for category_code null rates and free-text length percentiles, so cleansing would still be built without knowing how often category is missing or what the free text looks like.

Exam caveat: Profiling precedes cleansing design and gold publication. Exam scenarios that profile channel mix and duplicate rates, agent workload and resolution times, or location precision and closure dates before designing category-code rules are testing whether you know discovery must target the relevant fields first. Profile nulls, lengths, and categorical frequencies, then write rules against evidence.

Operational check: Run profiling on a representative 311 bronze sample—null rate for category_code, length percentiles for free text, top unmapped phrases. Document findings, draft cleansing rules (including reference-table mapping), and only then prototype gold SLA metrics. Re-profile after cleansing to confirm null categories drop without silent mass deletion of tickets.