Article summary
Break Uganda +256 WhatsApp nonresponse down by provenance, format and observation state instead of filling unknowns with demographics.
The most dangerous response to missing gender and age output in a Uganda +256 list is to fill it from names, avatar style or historical averages. Build a nonresponse matrix and preserve why the team does not know.
Use two axes in the matrix
| Source State | Format exception | Not observed | Visibility limited | Task exception |
|---|---|---|---|---|
| Website | Count | Count | Count | Count |
| Store | Count | Count | Count | Count |
| Partner | Count | Count | Count | Count |
Repair only missingness that can be repaired
Format exceptions can return to phone_raw and country_evidence; lost provenance may be recovered from import logs. Unobservable rows remain unknown and cannot be “fixed” into a demographic label.
+256 is not a Ugandan profile
The code supports numbering rules but does not prove present location, citizenship or population attributes. Cross-border use and reassignment require business evidence.
The denominator changes with the question
Task success uses format-processable rows as its denominator. A profile distribution describes only usable observations and must also show their share of the original submission.
Mark columns that must never be imputed
Do not populate gender_observed or age_band_observed from names, titles, photos or region. A manual correction must retain its evidence and reviewer.
Respect the AppShai input-output boundary
Upload a TXT file with one phone per line. Returned Excel enters an isolated table and reconnects to provenance by batch_row; internal names and customer labels do not belong in the screening task.
Sample abnormal matrix cells
When one source is concentrated in one state, sample that cell and inspect it. This exposes collection templates, prefix handling and stale-list defects.
Do not use profiles to exclude
Inferred age or gender should not reduce service, price or contact priority. User intent, permission and actual interaction are more defensible segmentation signals.
The executable Uganda conclusion
Unknown is not waste; it is a diagnostic entry. The matrix tells a team where data can be repaired and where it must admit that no judgement is available.
