Article summary
Build a source ledger for a multi-source DR Congo +243 list and accept activity, profile and mapping fields separately instead of relying on one hit rate.
Quality problems in a DR Congo +243 phone pool often begin during list assembly rather than in the phones themselves. Stores, resellers, events and a legacy CRM use different collection rules. Running WhatsApp Full Format without a provenance ledger can produce many columns but very little explainable evidence.
Issue an identity card to each source
| Source field | Required record | If missing |
|---|---|---|
| source_owner | Accountable list owner | Pause main-batch admission |
| collected_at | Collection or confirmation date | Mark freshness unknown |
| country_evidence | Country choice, order or address | Do not guess +243 |
| allowed_use | Current business purpose | Quarantine for review |
Normalize without overwriting the source
Retain phone_raw and create a separate +243 international candidate. Never prepend the code twice. Convert national notation only when source evidence supports DR Congo. Conflicting country, length or service evidence enters an exception queue; padding and deletion must not manufacture a high pass rate.
Accept Full Format by field family
AIPUSH WhatsApp Full Format may include phone, activity time, active days, gender, age, avatar, skin-tone observation, avatar type, business-account observation and mapped phone. Base/activity, profile observation and account relationship are three evidence families. Calculate missingness and conflicts for each.
Province does not come from a phone or avatar
Operational province comes from an order, address or self-report. +243 supports numbering-plan classification, while an avatar cannot prove residence, community, language or purchasing power. When province evidence is absent, keep region_unknown instead of completing a chart with a guess.
Separate TXT execution from business context
The execution file is TXT with one phone per line. Name, order, province, permission and contact_id stay in a private crosswalk and reconnect through batch_row. Returned Excel enters staging rather than overwriting the CRM.
Start acceptance with count conservation
Submitted rows should reconcile to returned, format exception, processing unknown and explicitly excluded counts. Then compare field missingness by provenance. A reseller with a high exception rate needs a collection and formatting review; it is not evidence about people in DR Congo.
Route mappings to relationship review
A mapped WhatsApp phone is a dated relationship observation. One-to-many, many-to-one or disagreement with the current primary phone enters mapping_conflict. A human reviewer records the evidence rather than merging customers automatically.
Give profile columns narrower access
Gender, age, avatar and skin-tone observations must not support identity verification, credit, employment, insurance or differential pricing. A team that only operates a service queue does not need the profile worksheet. Delete person-level temporary Excel when the purpose ends.
A sound handoff is not “100% complete”
Acceptable output preserves unknowns, exceptions and limitations and records checked_at, task version, source coverage and deletion date. Full Format then becomes an auditable data operation, not merely an export with many columns.
