Article summary
Quarantine Japan +81 formatting exceptions before defining a defensible denominator, unknown states and small-cell rules for a WhatsApp Gender and Age report.
The credibility of a Japan WhatsApp Gender and Age report is usually decided before any percentage is calculated. If +81 conversion, duplicate phones and task unknowns are not explained in the denominator, even a polished age chart cannot be reproduced.
Build the denominator through five counts
| Layer | Contains | Purpose |
|---|---|---|
| source_rows | Original rows | Count conservation |
| unique_candidates | Deduplicated candidates | Execution size |
| format_valid | Accepted phone form | Phone quality |
| successfully_checked | Completed checks | Task denominator |
| profile_observed | Usable profile observations | Reporting denominator |
Do not handle a national trunk zero by intuition
Preserve phone_raw and explicit country evidence and create an international candidate under current Japan numbering rules in a separate column. Never add +81 twice. Unsupported national notation enters review; a bulk first-digit deletion is not normalization.
Age and gender describe a dated observation
The task may return phone, age, gender, avatar and mapped phone. These are not user self-declarations and do not prove legal age, identity or real-world gender. Retain checked_at and task version.
Unknown cannot be folded into a category
Unknown age does not enter the oldest band, and unknown gender is not completed from name or avatar. Report unknown count and rate separately; a more complete-looking table can otherwise carry more bias.
Set the small-cell rule before analysis
If crossing region, source, age band and gender leaves too few records, combine age ranges or stop the split. Set the threshold before viewing results so grouping is not changed to obtain a preferred conclusion.
Give language preference separate evidence
+81, a Japanese-script name or an avatar must not populate a Japanese-language preference. Language comes from user choice, support confirmation or content behaviour, with source and date.
TXT has an execution-only role
Upload one phone per line. Name, region, permission and contact_id remain in a private crosswalk. Excel reconnects in staging through batch_row, and mapping conflicts do not merge automatically.
Use repeat samples to examine stability
Keep a small reviewed control cohort and recheck it at different times to separate changing observations from workflow defects. The exercise does not support a fixed accuracy claim across countries or dates.
Limitations belong beside the chart
Publish check date, valid denominator, unknown count, provenance coverage and prohibited uses with every chart. The value of a Japan country page is not finer human labels; it is making every percentage traceable.
