Article summary
Audit character confusion and lost provenance in Afghanistan +93 lists copied from paper or chat before accepting WhatsApp Full Format fields.
When an Afghanistan +93 list comes from paper forms, forwarded chats or manual entry, the first question is not which WhatsApp fields are available; it is whether the phone was transcribed correctly. Full Format cannot restore a missing digit, and a transcription defect must not become “not registered.”
Locate three transcription risks
| Risk | Signal | Action |
|---|---|---|
| Character confusion | Letters or different numeral forms mixed in | Review source image |
| Truncation | Abnormal lengths cluster by source | Pause that group |
| Lost provenance | No form or collector can be located | Exclude from production |
Do not turn cleanup into digit guessing
Retain phone_raw, a reference to the original image or message, and the data-entry operator. Spaces and international-prefix notation may be standardized, but missing digits cannot be invented and apparently extra digits cannot be deleted just to fit +93.
Run a small double-entry sample
Take a sample from each source and have two people enter it independently, then compare differences. If errors cluster around one form, camera quality or keyboard, repair collection before scaling the checking task.
Accept Full Format in three passes
WhatsApp Full Format may return phone and activity, age/gender and avatar observations, plus business-account and mapped-phone relationships. Accept each family separately. Missingness in one family cannot prove that the submitted phone was wrong.
Never overwrite the original notation
The internal table keeps phone_raw, phone_candidate, transform_rule and review_state. TXT contains only reviewed phones, one per line. Returned Excel reconnects through batch_row, never through visual row order.
Do not infer language or identity from a profile
Avatar, name, phone and numeral style cannot reliably generate language, community, religion or identity labels. Language preference comes from user choice or an actual communication record.
An unknown row is quality evidence
Count transcription_unknown, format_exception, task_unknown and returned separately. A rising unknown share triggers investigation; it is not a reason to delete difficult rows and advertise a higher success figure.
Custody for mapping conflicts
If a mapped phone conflicts with the source or CRM, route it to human review with decision evidence. Until confirmed, it must not merge customers, trigger messaging or replace the master phone.
The correct order for this guide
Traceable source, double-entry sample, phone formatting, TXT execution, field-family acceptance and conflict review: this order takes slightly longer than direct upload, but prevents transcription defects from becoming false audience conclusions.
