Article summary
Use Bolivia +591 WhatsApp Registration to build a unique valid-phone baseline before any later Activity queue experiment.
A Bolivia Registration page should not repeat an Activity tutorial. Its distinct job is to build a baseline: before adding time fields, establish how many unique, format-valid +591 phones were successfully checked and observed as registered.
Keep the baseline funnel to four layers
| Layer | Count |
|---|---|
| Source | Raw rows and provenance |
| Preparation | Unique format candidates |
| Execution | Successfully checked |
| Observation | Registered, not observed, unknown |
Exclude activity and profile fields
When the question is registration, do not collect activity time, age, gender or avatar. Fewer fields make the later experiment easier to explain.
+591 exceptions stay visible outside the denominator
Preserve phone_raw and record duplicated code, length conflict and country_unknown. Report quarantined counts instead of deleting them and making provenance look cleaner.
Deduplicate execution while retaining source weight
Check one phone once, but let the crosswalk retain every provenance occurrence. Reconnection does not count one registered phone multiple times or lose source-quality evidence.
Separate unknown and not observed
Task uncertainty is not a negative state. Report registered, not_observed and unknown separately and carry the same cohort definition into any later Activity test.
Apply permission before the baseline
Records with no current purpose, unknown provenance or opt-out do not return through technical analysis. Registration does not alter permission.
Freeze a version for the next experiment
Store cohort_id, rule_version and checked_at. Sample later Activity work from this defined baseline instead of comparing a different list.
How the two Bolivia pages cooperate
This page decides who enters the experiment; the Activity page tests whether queue outcomes improve. Their roles differ, while cohort_id connects the evidence.
Baseline acceptance
Every row can explain provenance, format, deduplication, task state and permission. With that baseline, later metrics do not drift with the denominator.
